Update architectures

This was probably used in the prototyping stage, left as a reference,
and then forgotten. Nothing uses it any more.

.github/ISSUE_TEMPLATE/01_bugs.md

@@ -4,6 +4,8 @@ about: Use this template if you came across a bug or unexpected behaviour differ
 
 ---
 
+<!-- NOTE: For questions or install related issues, please open a Discussion instead. -->
+
 ## How to reproduce the behaviour
 <!-- Include a code example or the steps that led to the problem. Please try to be as specific as possible. -->
 

.github/ISSUE_TEMPLATE/config.yml

@@ -1,8 +1,5 @@
 blank_issues_enabled: false
 contact_links:
-  - name: ⚠️ Python 3.10 Support
-    url: https://github.com/explosion/spaCy/discussions/9418
-    about: Python 3.10 wheels haven't been released yet, see the link for details.
   - name: 🗯 Discussions Forum
     url: https://github.com/explosion/spaCy/discussions
     about: Install issues, usage questions, general discussion and anything else that isn't a bug report.

.github/azure-steps.yml

@@ -27,7 +27,6 @@ steps:
 
   - script: python -m mypy spacy
     displayName: 'Run mypy'
-    condition: ne(variables['python_version'], '3.10')
 
   - task: DeleteFiles@1
     inputs:
@@ -41,7 +40,7 @@ steps:
 
   - bash: |
       ${{ parameters.prefix }} SDIST=$(python -c "import os;print(os.listdir('./dist')[-1])" 2>&1)
-      ${{ parameters.prefix }} python -m pip install dist/$SDIST
+      ${{ parameters.prefix }} SPACY_NUM_BUILD_JOBS=2 python -m pip install dist/$SDIST
     displayName: "Install from sdist"
 
   - script: |
@@ -64,12 +63,12 @@ steps:
     displayName: "Run GPU tests"
     condition: eq(${{ parameters.gpu }}, true)
 
-  - script: |
-      python -m spacy download ca_core_news_sm
-      python -m spacy download ca_core_news_md
-      python -c "import spacy; nlp=spacy.load('ca_core_news_sm'); doc=nlp('test')"
-    displayName: 'Test download CLI'
-    condition: eq(variables['python_version'], '3.8')
+#  - script: |
+#      python -m spacy download ca_core_news_sm
+#      python -m spacy download ca_core_news_md
+#      python -c "import spacy; nlp=spacy.load('ca_core_news_sm'); doc=nlp('test')"
+#    displayName: 'Test download CLI'
+#    condition: eq(variables['python_version'], '3.8')
 
   - script: |
       python -m spacy convert extra/example_data/ner_example_data/ner-token-per-line-conll2003.json .
@@ -93,17 +92,17 @@ steps:
     displayName: 'Test train CLI'
     condition: eq(variables['python_version'], '3.8')
 
-  - script: |
-      python -c "import spacy; config = spacy.util.load_config('ner.cfg'); config['components']['ner'] = {'source': 'ca_core_news_sm'}; config.to_disk('ner_source_sm.cfg')"
-      PYTHONWARNINGS="error,ignore::DeprecationWarning" python -m spacy assemble ner_source_sm.cfg output_dir
-    displayName: 'Test assemble CLI'
-    condition: eq(variables['python_version'], '3.8')
-
-  - script: |
-      python -c "import spacy; config = spacy.util.load_config('ner.cfg'); config['components']['ner'] = {'source': 'ca_core_news_md'}; config.to_disk('ner_source_md.cfg')"
-      python -m spacy assemble ner_source_md.cfg output_dir 2>&1 | grep -q W113
-    displayName: 'Test assemble CLI vectors warning'
-    condition: eq(variables['python_version'], '3.8')
+#  - script: |
+#      python -c "import spacy; config = spacy.util.load_config('ner.cfg'); config['components']['ner'] = {'source': 'ca_core_news_sm'}; config.to_disk('ner_source_sm.cfg')"
+#      PYTHONWARNINGS="error,ignore::DeprecationWarning" python -m spacy assemble ner_source_sm.cfg output_dir
+#    displayName: 'Test assemble CLI'
+#    condition: eq(variables['python_version'], '3.8')
+#
+#  - script: |
+#      python -c "import spacy; config = spacy.util.load_config('ner.cfg'); config['components']['ner'] = {'source': 'ca_core_news_md'}; config.to_disk('ner_source_md.cfg')"
+#      python -m spacy assemble ner_source_md.cfg output_dir 2>&1 | grep -q W113
+#    displayName: 'Test assemble CLI vectors warning'
+#    condition: eq(variables['python_version'], '3.8')
 
   - script: |
       python .github/validate_universe_json.py website/meta/universe.json
@@ -111,7 +110,7 @@ steps:
     condition: eq(variables['python_version'], '3.8')
 
   - script: |
-      ${{ parameters.prefix }} python -m pip install thinc-apple-ops
+      ${{ parameters.prefix }} python -m pip install --pre thinc-apple-ops
       ${{ parameters.prefix }} python -m pytest --pyargs spacy
     displayName: "Run CPU tests with thinc-apple-ops"
-    condition: and(startsWith(variables['imageName'], 'macos'), eq(variables['python.version'], '3.9'))
+    condition: and(startsWith(variables['imageName'], 'macos'), eq(variables['python.version'], '3.10'))

.github/contributors/Lucaterre.md

@@ -0,0 +1,106 @@
+# spaCy contributor agreement
+
+This spaCy Contributor Agreement (**"SCA"**) is based on the
+[Oracle Contributor Agreement](http://www.oracle.com/technetwork/oca-405177.pdf).
+The SCA applies to any contribution that you make to any product or project
+managed by us (the **"project"**), and sets out the intellectual property rights
+you grant to us in the contributed materials. The term **"us"** shall mean
+[ExplosionAI GmbH](https://explosion.ai/legal). The term
+**"you"** shall mean the person or entity identified below.
+
+If you agree to be bound by these terms, fill in the information requested
+below and include the filled-in version with your first pull request, under the
+folder [`.github/contributors/`](/.github/contributors/). The name of the file
+should be your GitHub username, with the extension `.md`. For example, the user
+example_user would create the file `.github/contributors/example_user.md`.
+
+Read this agreement carefully before signing. These terms and conditions
+constitute a binding legal agreement.
+
+## Contributor Agreement
+
+1. The term "contribution" or "contributed materials" means any source code,
+object code, patch, tool, sample, graphic, specification, manual,
+documentation, or any other material posted or submitted by you to the project.
+
+2. With respect to any worldwide copyrights, or copyright applications and
+registrations, in your contribution:
+
+    * you hereby assign to us joint ownership, and to the extent that such
+    assignment is or becomes invalid, ineffective or unenforceable, you hereby
+    grant to us a perpetual, irrevocable, non-exclusive, worldwide, no-charge,
+    royalty-free, unrestricted license to exercise all rights under those
+    copyrights. This includes, at our option, the right to sublicense these same
+    rights to third parties through multiple levels of sublicensees or other
+    licensing arrangements;
+
+    * you agree that each of us can do all things in relation to your
+    contribution as if each of us were the sole owners, and if one of us makes
+    a derivative work of your contribution, the one who makes the derivative
+    work (or has it made will be the sole owner of that derivative work;
+
+    * you agree that you will not assert any moral rights in your contribution
+    against us, our licensees or transferees;
+
+    * you agree that we may register a copyright in your contribution and
+    exercise all ownership rights associated with it; and
+
+    * you agree that neither of us has any duty to consult with, obtain the
+    consent of, pay or render an accounting to the other for any use or
+    distribution of your contribution.
+
+3. With respect to any patents you own, or that you can license without payment
+to any third party, you hereby grant to us a perpetual, irrevocable,
+non-exclusive, worldwide, no-charge, royalty-free license to:
+
+    * make, have made, use, sell, offer to sell, import, and otherwise transfer
+    your contribution in whole or in part, alone or in combination with or
+    included in any product, work or materials arising out of the project to
+    which your contribution was submitted, and
+
+    * at our option, to sublicense these same rights to third parties through
+    multiple levels of sublicensees or other licensing arrangements.
+
+4. Except as set out above, you keep all right, title, and interest in your
+contribution. The rights that you grant to us under these terms are effective
+on the date you first submitted a contribution to us, even if your submission
+took place before the date you sign these terms.
+
+5. You covenant, represent, warrant and agree that:
+
+    * Each contribution that you submit is and shall be an original work of
+    authorship and you can legally grant the rights set out in this SCA;
+
+    * to the best of your knowledge, each contribution will not violate any
+    third party's copyrights, trademarks, patents, or other intellectual
+    property rights; and
+
+    * each contribution shall be in compliance with U.S. export control laws and
+    other applicable export and import laws. You agree to notify us if you
+    become aware of any circumstance which would make any of the foregoing
+    representations inaccurate in any respect. We may publicly disclose your
+    participation in the project, including the fact that you have signed the SCA.
+
+6. This SCA is governed by the laws of the State of California and applicable
+U.S. Federal law. Any choice of law rules will not apply.
+
+7. Please place an “x” on one of the applicable statement below. Please do NOT
+mark both statements:
+
+    * [x] I am signing on behalf of myself as an individual and no other person
+    or entity, including my employer, has or will have rights with respect to my
+    contributions.
+
+    * [ ] I am signing on behalf of my employer or a legal entity and I have the
+    actual authority to contractually bind that entity.
+
+## Contributor Details
+
+| Field                          | Entry         |
+|------------------------------- |---------------|
+| Name                           | Lucas Terriel |
+| Company name (if applicable)   |               |
+| Title or role (if applicable)  |               |
+| Date                           | 2022-06-20    |
+| GitHub username                | Lucaterre     |
+| Website (optional)             |               |

.github/contributors/fonfonx.md

@@ -0,0 +1,106 @@
+# spaCy contributor agreement
+
+This spaCy Contributor Agreement (**"SCA"**) is based on the
+[Oracle Contributor Agreement](http://www.oracle.com/technetwork/oca-405177.pdf).
+The SCA applies to any contribution that you make to any product or project
+managed by us (the **"project"**), and sets out the intellectual property rights
+you grant to us in the contributed materials. The term **"us"** shall mean
+[ExplosionAI GmbH](https://explosion.ai/legal). The term
+**"you"** shall mean the person or entity identified below.
+
+If you agree to be bound by these terms, fill in the information requested
+below and include the filled-in version with your first pull request, under the
+folder [`.github/contributors/`](/.github/contributors/). The name of the file
+should be your GitHub username, with the extension `.md`. For example, the user
+example_user would create the file `.github/contributors/example_user.md`.
+
+Read this agreement carefully before signing. These terms and conditions
+constitute a binding legal agreement.
+
+## Contributor Agreement
+
+1. The term "contribution" or "contributed materials" means any source code,
+object code, patch, tool, sample, graphic, specification, manual,
+documentation, or any other material posted or submitted by you to the project.
+
+2. With respect to any worldwide copyrights, or copyright applications and
+registrations, in your contribution:
+
+    * you hereby assign to us joint ownership, and to the extent that such
+    assignment is or becomes invalid, ineffective or unenforceable, you hereby
+    grant to us a perpetual, irrevocable, non-exclusive, worldwide, no-charge,
+    royalty-free, unrestricted license to exercise all rights under those
+    copyrights. This includes, at our option, the right to sublicense these same
+    rights to third parties through multiple levels of sublicensees or other
+    licensing arrangements;
+
+    * you agree that each of us can do all things in relation to your
+    contribution as if each of us were the sole owners, and if one of us makes
+    a derivative work of your contribution, the one who makes the derivative
+    work (or has it made will be the sole owner of that derivative work;
+
+    * you agree that you will not assert any moral rights in your contribution
+    against us, our licensees or transferees;
+
+    * you agree that we may register a copyright in your contribution and
+    exercise all ownership rights associated with it; and
+
+    * you agree that neither of us has any duty to consult with, obtain the
+    consent of, pay or render an accounting to the other for any use or
+    distribution of your contribution.
+
+3. With respect to any patents you own, or that you can license without payment
+to any third party, you hereby grant to us a perpetual, irrevocable,
+non-exclusive, worldwide, no-charge, royalty-free license to:
+
+    * make, have made, use, sell, offer to sell, import, and otherwise transfer
+    your contribution in whole or in part, alone or in combination with or
+    included in any product, work or materials arising out of the project to
+    which your contribution was submitted, and
+
+    * at our option, to sublicense these same rights to third parties through
+    multiple levels of sublicensees or other licensing arrangements.
+
+4. Except as set out above, you keep all right, title, and interest in your
+contribution. The rights that you grant to us under these terms are effective
+on the date you first submitted a contribution to us, even if your submission
+took place before the date you sign these terms.
+
+5. You covenant, represent, warrant and agree that:
+
+    * Each contribution that you submit is and shall be an original work of
+    authorship and you can legally grant the rights set out in this SCA;
+
+    * to the best of your knowledge, each contribution will not violate any
+    third party's copyrights, trademarks, patents, or other intellectual
+    property rights; and
+
+    * each contribution shall be in compliance with U.S. export control laws and
+    other applicable export and import laws. You agree to notify us if you
+    become aware of any circumstance which would make any of the foregoing
+    representations inaccurate in any respect. We may publicly disclose your
+    participation in the project, including the fact that you have signed the SCA.
+
+6. This SCA is governed by the laws of the State of California and applicable
+U.S. Federal law. Any choice of law rules will not apply.
+
+7. Please place an “x” on one of the applicable statement below. Please do NOT
+mark both statements:
+
+    * [x] I am signing on behalf of myself as an individual and no other person
+    or entity, including my employer, has or will have rights with respect to my
+    contributions.
+
+    * [ ] I am signing on behalf of my employer or a legal entity and I have the
+    actual authority to contractually bind that entity.
+
+## Contributor Details
+
+| Field                          | Entry                |
+|------------------------------- | -------------------- |
+| Name                           | Xavier Fontaine      |
+| Company name (if applicable)   |                      |
+| Title or role (if applicable)  |                      |
+| Date                           | 2022-04-13           |
+| GitHub username                | fonfonx              |
+| Website (optional)             |                      |

.github/spacy_universe_alert.py

@@ -0,0 +1,67 @@
+import os
+import sys
+import json
+from datetime import datetime
+
+from slack_sdk.web.client import WebClient
+
+CHANNEL = "#alerts-universe"
+SLACK_TOKEN = os.environ.get("SLACK_BOT_TOKEN", "ENV VAR not available!")
+DATETIME_FORMAT = "%Y-%m-%dT%H:%M:%SZ"
+
+client = WebClient(SLACK_TOKEN)
+github_context = json.loads(sys.argv[1])
+
+event = github_context['event']
+pr_title = event['pull_request']["title"]
+pr_link = event['pull_request']["patch_url"].replace(".patch", "")
+pr_author_url = event['sender']["html_url"]
+pr_author_name = pr_author_url.rsplit('/')[-1]
+pr_created_at_dt = datetime.strptime(
+    event['pull_request']["created_at"],
+    DATETIME_FORMAT
+)
+pr_created_at = pr_created_at_dt.strftime("%c")
+pr_updated_at_dt = datetime.strptime(
+    event['pull_request']["updated_at"],
+    DATETIME_FORMAT
+)
+pr_updated_at = pr_updated_at_dt.strftime("%c")
+
+blocks = [
+    {
+      "type": "section",
+      "text": {
+        "type": "mrkdwn",
+        "text": "📣 New spaCy Universe Project Alert ✨"
+      }
+    },
+    {
+      "type": "section",
+      "fields": [
+        {
+          "type": "mrkdwn",
+          "text": f"*Pull Request:*\n<{pr_link}|{pr_title}>"
+        },
+        {
+          "type": "mrkdwn",
+          "text": f"*Author:*\n<{pr_author_url}|{pr_author_name}>"
+        },
+        {
+          "type": "mrkdwn",
+          "text": f"*Created at:*\n {pr_created_at}"
+        },
+        {
+          "type": "mrkdwn",
+          "text": f"*Last Updated:*\n {pr_updated_at}"
+        }
+      ]
+    }
+  ]
+
+
+client.chat_postMessage(
+    channel=CHANNEL,
+    text="spaCy universe project PR alert",
+    blocks=blocks
+)

.github/workflows/explosionbot.yml

@@ -23,5 +23,5 @@ jobs:
         env:
           INPUT_TOKEN: ${{ secrets.EXPLOSIONBOT_TOKEN }}
           INPUT_BK_TOKEN: ${{ secrets.BUILDKITE_SECRET }}
-          ENABLED_COMMANDS: "test_gpu,test_slow"
+          ENABLED_COMMANDS: "test_gpu,test_slow,test_slow_gpu"
           ALLOWED_TEAMS: "spaCy"

.github/workflows/gputests.yml

@@ -0,0 +1,22 @@
+name: Weekly GPU tests
+
+on:
+  schedule:
+    - cron: '0 1 * * MON'
+
+jobs:
+  weekly-gputests:
+    strategy:
+      fail-fast: false
+      matrix:
+        branch: [master, v4]
+    if: github.repository_owner == 'explosion'
+    runs-on: ubuntu-latest
+    steps:
+      - name: Trigger buildkite build
+        uses: buildkite/trigger-pipeline-action@v1.2.0
+        env:
+          PIPELINE: explosion-ai/spacy-slow-gpu-tests
+          BRANCH: ${{ matrix.branch }}
+          MESSAGE: ":github: Weekly GPU + slow tests - triggered from a GitHub Action"
+          BUILDKITE_API_ACCESS_TOKEN: ${{ secrets.BUILDKITE_SECRET }}

.github/workflows/slowtests.yml

@@ -0,0 +1,38 @@
+name: Daily slow tests
+
+on:
+  schedule:
+    - cron: '0 0 * * *'
+
+jobs:
+  daily-slowtests:
+    strategy:
+      fail-fast: false
+      matrix:
+        branch: [master, v4]
+    if: github.repository_owner == 'explosion'
+    runs-on: ubuntu-latest
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v1
+        with:
+          ref: ${{ matrix.branch }}
+      - name: Get commits from past 24 hours
+        id: check_commits
+        run: |
+          today=$(date '+%Y-%m-%d %H:%M:%S')
+          yesterday=$(date -d "yesterday" '+%Y-%m-%d %H:%M:%S')
+          if git log --after="$yesterday" --before="$today" | grep commit ; then
+            echo "::set-output name=run_tests::true"
+          else
+            echo "::set-output name=run_tests::false"
+          fi
+
+      - name: Trigger buildkite build
+        if: steps.check_commits.outputs.run_tests == 'true'
+        uses: buildkite/trigger-pipeline-action@v1.2.0
+        env:
+          PIPELINE: explosion-ai/spacy-slow-tests
+          BRANCH: ${{ matrix.branch }}
+          MESSAGE: ":github: Daily slow tests - triggered from a GitHub Action"
+          BUILDKITE_API_ACCESS_TOKEN: ${{ secrets.BUILDKITE_SECRET }}

.github/workflows/spacy_universe_alert.yml

@@ -0,0 +1,30 @@
+name: spaCy universe project alert
+
+on:
+  pull_request_target:
+    paths:
+      - "website/meta/universe.json"
+
+jobs:
+  build:
+    runs-on: ubuntu-latest
+
+    steps:
+      - name: Dump GitHub context
+        env:
+          GITHUB_CONTEXT: ${{ toJson(github) }}
+          PR_NUMBER: ${{github.event.number}}
+        run: |
+          echo "$GITHUB_CONTEXT"
+
+      - uses: actions/checkout@v1
+      - uses: actions/setup-python@v1
+      - name: Install Bernadette app dependency and send an alert
+        env:
+          SLACK_BOT_TOKEN: ${{ secrets.SLACK_BOT_TOKEN }}
+          GITHUB_CONTEXT: ${{ toJson(github) }}
+          CHANNEL: "#alerts-universe"
+        run: |
+          pip install slack-sdk==3.17.2 aiohttp==3.8.1
+          echo "$CHANNEL"
+          python .github/spacy_universe_alert.py "$GITHUB_CONTEXT"

.gitignore

@@ -9,7 +9,6 @@ keys/
 spacy/tests/package/setup.cfg
 spacy/tests/package/pyproject.toml
 spacy/tests/package/requirements.txt
-spacy/tests/universe/universe.json
 
 # Website
 website/.cache/

.pre-commit-config.yaml

@@ -1,9 +1,10 @@
 repos:
 -   repo: https://github.com/ambv/black
-    rev: 21.6b0
+    rev: 22.3.0
     hooks:
     - id: black
       language_version: python3.7
+      additional_dependencies: ['click==8.0.4']
 -   repo: https://gitlab.com/pycqa/flake8
     rev: 3.9.2
     hooks:

CONTRIBUTING.md

@@ -144,7 +144,7 @@ Changes to `.py` files will be effective immediately.
 
 When fixing a bug, first create an
 [issue](https://github.com/explosion/spaCy/issues) if one does not already
-exist.  The description text can be very short – we don't want to make this too
+exist. The description text can be very short – we don't want to make this too
 bureaucratic.
 
 Next, add a test to the relevant file in the
@@ -233,7 +233,7 @@ also want to keep an eye on unused declared variables or repeated
 (i.e. overwritten) dictionary keys. If your code was formatted with `black`
 (see above), you shouldn't see any formatting-related warnings.
 
-The [`.flake8`](.flake8) config defines the configuration we use for this
+The `flake8` section in [`setup.cfg`](setup.cfg) defines the configuration we use for this
 codebase. For example, we're not super strict about the line length, and we're
 excluding very large files like lemmatization and tokenizer exception tables.
 

LICENSE

@@ -1,6 +1,6 @@
 The MIT License (MIT)
 
-Copyright (C) 2016-2021 ExplosionAI GmbH, 2016 spaCy GmbH, 2015 Matthew Honnibal
+Copyright (C) 2016-2022 ExplosionAI GmbH, 2016 spaCy GmbH, 2015 Matthew Honnibal
 
 Permission is hereby granted, free of charge, to any person obtaining a copy
 of this software and associated documentation files (the "Software"), to deal

MANIFEST.in

@@ -1,11 +1,8 @@
-recursive-include include *.h
-recursive-include spacy *.pyi *.pyx *.pxd *.txt *.cfg *.jinja *.toml
+recursive-include spacy *.pyi *.pyx *.pxd *.txt *.cfg *.jinja *.toml *.hh
 include LICENSE
 include README.md
 include pyproject.toml
 include spacy/py.typed
-recursive-exclude spacy/lang *.json
-recursive-include spacy/lang *.json.gz
-recursive-include spacy/cli *.json *.yml
+recursive-include spacy/cli *.yml
 recursive-include licenses *
 recursive-exclude spacy *.cpp

README.md

@@ -16,7 +16,7 @@ production-ready [**training system**](https://spacy.io/usage/training) and easy
 model packaging, deployment and workflow management. spaCy is commercial
 open-source software, released under the MIT license.
 
-💫 **Version 3.2 out now!**
+💫 **Version 3.3.1 out now!**
 [Check out the release notes here.](https://github.com/explosion/spaCy/releases)
 
 [![Azure Pipelines](https://img.shields.io/azure-devops/build/explosion-ai/public/8/master.svg?logo=azure-pipelines&style=flat-square&label=build)](https://dev.azure.com/explosion-ai/public/_build?definitionId=8)
@@ -32,19 +32,20 @@ open-source software, released under the MIT license.
 
 ## 📖 Documentation
 
-| Documentation              |                                                                |
-| -------------------------- | -------------------------------------------------------------- |
-| ⭐️ **[spaCy 101]**        | New to spaCy? Here's everything you need to know!              |
-| 📚 **[Usage Guides]**      | How to use spaCy and its features.                             |
-| 🚀 **[New in v3.0]**       | New features, backwards incompatibilities and migration guide. |
-| 🪐 **[Project Templates]** | End-to-end workflows you can clone, modify and run.            |
-| 🎛 **[API Reference]**      | The detailed reference for spaCy's API.                        |
-| 📦 **[Models]**            | Download trained pipelines for spaCy.                          |
-| 🌌 **[Universe]**          | Plugins, extensions, demos and books from the spaCy ecosystem. |
-| 👩‍🏫 **[Online Course]**     | Learn spaCy in this free and interactive online course.        |
-| 📺 **[Videos]**            | Our YouTube channel with video tutorials, talks and more.      |
-| 🛠 **[Changelog]**          | Changes and version history.                                   |
-| 💝 **[Contribute]**        | How to contribute to the spaCy project and code base.          |
+| Documentation                                                                                                                                                                                                             |                                                                                                                                                                                                                                                                                                                              |
+| ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| ⭐️ **[spaCy 101]**                                                                                                                                                                                                       | New to spaCy? Here's everything you need to know!                                                                                                                                                                                                                                                                            |
+| 📚 **[Usage Guides]**                                                                                                                                                                                                     | How to use spaCy and its features.                                                                                                                                                                                                                                                                                           |
+| 🚀 **[New in v3.0]**                                                                                                                                                                                                      | New features, backwards incompatibilities and migration guide.                                                                                                                                                                                                                                                               |
+| 🪐 **[Project Templates]**                                                                                                                                                                                                | End-to-end workflows you can clone, modify and run.                                                                                                                                                                                                                                                                          |
+| 🎛 **[API Reference]**                                                                                                                                                                                                     | The detailed reference for spaCy's API.                                                                                                                                                                                                                                                                                      |
+| 📦 **[Models]**                                                                                                                                                                                                           | Download trained pipelines for spaCy.                                                                                                                                                                                                                                                                                        |
+| 🌌 **[Universe]**                                                                                                                                                                                                         | Plugins, extensions, demos and books from the spaCy ecosystem.                                                                                                                                                                                                                                                               |
+| 👩‍🏫 **[Online Course]**                                                                                                                                                                                                    | Learn spaCy in this free and interactive online course.                                                                                                                                                                                                                                                                      |
+| 📺 **[Videos]**                                                                                                                                                                                                           | Our YouTube channel with video tutorials, talks and more.                                                                                                                                                                                                                                                                    |
+| 🛠 **[Changelog]**                                                                                                                                                                                                         | Changes and version history.                                                                                                                                                                                                                                                                                                 |
+| 💝 **[Contribute]**                                                                                                                                                                                                       | How to contribute to the spaCy project and code base.                                                                                                                                                                                                                                                                        |
+| <a href="https://explosion.ai/spacy-tailored-pipelines"><img src="https://user-images.githubusercontent.com/13643239/152853098-1c761611-ccb0-4ec6-9066-b234552831fe.png" width="125" alt="spaCy Tailored Pipelines"/></a> | Get a custom spaCy pipeline, tailor-made for your NLP problem by spaCy's core developers. Streamlined, production-ready, predictable and maintainable. Start by completing our 5-minute questionnaire to tell us what you need and we'll be in touch! **[Learn more &rarr;](https://explosion.ai/spacy-tailored-pipelines)** |
 
 [spacy 101]: https://spacy.io/usage/spacy-101
 [new in v3.0]: https://spacy.io/usage/v3
@@ -60,9 +61,7 @@ open-source software, released under the MIT license.
 
 ## 💬 Where to ask questions
 
-The spaCy project is maintained by **[@honnibal](https://github.com/honnibal)**,
-**[@ines](https://github.com/ines)**, **[@svlandeg](https://github.com/svlandeg)**,
-**[@adrianeboyd](https://github.com/adrianeboyd)** and **[@polm](https://github.com/polm)**.
+The spaCy project is maintained by the [spaCy team](https://explosion.ai/about).
 Please understand that we won't be able to provide individual support via email.
 We also believe that help is much more valuable if it's shared publicly, so that
 more people can benefit from it.

azure-pipelines.yml

@@ -11,12 +11,14 @@ trigger:
     exclude:
       - "website/*"
       - "*.md"
+      - ".github/workflows/*"
 pr:
-   paths:
+  paths:
     exclude:
       - "*.md"
       - "website/docs/*"
       - "website/src/*"
+      - ".github/workflows/*"
 
 jobs:
   # Perform basic checks for most important errors (syntax etc.) Uses the config

extra/DEVELOPER_DOCS/Code Conventions.md

@@ -137,7 +137,7 @@ If any of the TODOs you've added are important and should be fixed soon, you sho
 
 ## Type hints
 
-We use Python type hints across the `.py` files wherever possible. This makes it easy to understand what a function expects and returns, and modern editors will be able to show this information to you when you call an annotated function. Type hints are not currently used in the `.pyx` (Cython) code, except for definitions of registered functions and component factories, where they're used for config validation.
+We use Python type hints across the `.py` files wherever possible. This makes it easy to understand what a function expects and returns, and modern editors will be able to show this information to you when you call an annotated function. Type hints are not currently used in the `.pyx` (Cython) code, except for definitions of registered functions and component factories, where they're used for config validation. Ideally when developing, run `mypy spacy` on the code base to inspect any issues.
 
 If possible, you should always use the more descriptive type hints like `List[str]` or even `List[Any]` instead of only `list`. We also annotate arguments and return types of `Callable` – although, you can simplify this if the type otherwise gets too verbose (e.g. functions that return factories to create callbacks). Remember that `Callable` takes two values: a **list** of the argument type(s) in order, and the return values.
 
@@ -155,6 +155,13 @@ def create_callback(some_arg: bool) -> Callable[[str, int], List[str]]:
     return callback
 ```
 
+For typing variables, we prefer the explicit format.
+
+```diff
+- var = value    # type: Type
++ var: Type = value
+```
+
 For model architectures, Thinc also provides a collection of [custom types](https://thinc.ai/docs/api-types), including more specific types for arrays and model inputs/outputs. Even outside of static type checking, using these types will make the code a lot easier to read and follow, since it's always clear what array types are expected (and what might go wrong if the output is different from the expected type).
 
 ```python
@@ -448,6 +455,10 @@ Regression tests are tests that refer to bugs reported in specific issues. They
 
 The test suite also provides [fixtures](https://github.com/explosion/spaCy/blob/master/spacy/tests/conftest.py) for different language tokenizers that can be used as function arguments of the same name and will be passed in automatically. Those should only be used for tests related to those specific languages. We also have [test utility functions](https://github.com/explosion/spaCy/blob/master/spacy/tests/util.py) for common operations, like creating a temporary file.
 
+### Testing Cython Code
+
+If you're developing Cython code (`.pyx` files), those extensions will need to be built before the test runner can test that code - otherwise it's going to run the tests with stale code from the last time the extension was built. You can build the extensions locally with `python setup.py build_ext -i`.
+
 ### Constructing objects and state
 
 Test functions usually follow the same simple structure: they set up some state, perform the operation you want to test and `assert` conditions that you expect to be true, usually before and after the operation.

extra/DEVELOPER_DOCS/ExplosionBot.md

@@ -0,0 +1,56 @@
+# Explosion-bot
+
+Explosion-bot is a robot that can be invoked to help with running particular test commands.
+
+## Permissions
+
+Only maintainers have permissions to summon explosion-bot. Each of the open source repos that use explosion-bot has its own team(s) of maintainers, and only github users who are members of those teams can successfully run bot commands.
+
+## Running robot commands
+
+To summon the robot, write a github comment on the issue/PR you wish to test. The comment must be in the following format:
+
+```
+@explosion-bot please test_gpu
+```
+
+Some things to note:
+
+- The `@explosion-bot please` must be the beginning of the command - you cannot add anything in front of this or else the robot won't know how to parse it. Adding anything at the end aside from the test name will also confuse the robot, so keep it simple!
+- The command name (such as `test_gpu`) must be one of the tests that the bot knows how to run. The available commands are documented in the bot's [workflow config](https://github.com/explosion/spaCy/blob/master/.github/workflows/explosionbot.yml#L26) and must match exactly one of the commands listed there.
+- The robot can't do multiple things at once, so if you want it to run multiple tests, you'll have to summon it with one comment per test.
+
+### Examples
+
+- Execute spaCy slow GPU tests with a custom thinc branch from a spaCy PR:
+
+  ```
+  @explosion-bot please test_slow_gpu --thinc-branch <branch_name>
+  ```
+
+  `branch_name` can either be a named branch, e.g: `develop`, or an unmerged PR, e.g: `refs/pull/<pr_number>/head`.
+
+- Execute spaCy Transformers GPU tests from a spaCy PR:
+
+  ```
+  @explosion-bot please test_gpu --run-on spacy-transformers --run-on-branch master --spacy-branch current_pr
+  ```
+
+  This will launch the GPU pipeline for the `spacy-transformers` repo on its `master` branch, using the current spaCy PR's branch to build spaCy.
+
+- General info about supported commands.
+
+  ```
+  @explosion-bot please info
+  ```
+
+- Help text for a specific command
+  ```
+  @explosion-bot please <command> --help
+  ```
+
+## Troubleshooting
+
+If the robot isn't responding to commands as expected, you can check its logs in the [Github Action](https://github.com/explosion/spaCy/actions/workflows/explosionbot.yml).
+
+For each command sent to the bot, there should be a run of the `explosion-bot` workflow. In the `Install and run explosion-bot` step, towards the ends of the logs you should see info about the configuration that the bot was run with, as well as any errors that the bot encountered.

licenses/3rd_party_licenses.txt

@@ -127,3 +127,36 @@ distributed under the License is distributed on an "AS IS" BASIS,
 WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License.
+
+
+coval
+-----
+
+* Files: scorer.py
+
+The implementations of ClusterEvaluator, lea, get_cluster_info, and
+get_markable_assignments are adapted from coval, which is distributed
+under the following license:
+
+The MIT License (MIT)
+
+Copyright 2018 Nafise Sadat Moosavi (ns.moosavi at gmail dot com)
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
+

pyproject.toml

@@ -5,8 +5,7 @@ requires = [
     "cymem>=2.0.2,<2.1.0",
     "preshed>=3.0.2,<3.1.0",
     "murmurhash>=0.28.0,<1.1.0",
-    "thinc>=8.0.12,<8.1.0",
-    "blis>=0.4.0,<0.8.0",
+    "thinc>=8.1.0,<8.2.0",
     "pathy",
     "numpy>=1.15.0",
 ]

requirements.txt

@@ -1,14 +1,13 @@
 # Our libraries
-spacy-legacy>=3.0.8,<3.1.0
+spacy-legacy>=3.0.9,<3.1.0
 spacy-loggers>=1.0.0,<2.0.0
 cymem>=2.0.2,<2.1.0
 preshed>=3.0.2,<3.1.0
-thinc>=8.0.12,<8.1.0
-blis>=0.4.0,<0.8.0
+thinc>=8.1.0,<8.2.0
 ml_datasets>=0.2.0,<0.3.0
 murmurhash>=0.28.0,<1.1.0
-wasabi>=0.8.1,<1.1.0
-srsly>=2.4.1,<3.0.0
+wasabi>=0.9.1,<1.1.0
+srsly>=2.4.3,<3.0.0
 catalogue>=2.0.6,<2.1.0
 typer>=0.3.0,<0.5.0
 pathy>=0.3.5
@@ -16,22 +15,23 @@ pathy>=0.3.5
 numpy>=1.15.0
 requests>=2.13.0,<3.0.0
 tqdm>=4.38.0,<5.0.0
-pydantic>=1.7.4,!=1.8,!=1.8.1,<1.9.0
+pydantic>=1.7.4,!=1.8,!=1.8.1,<1.10.0
 jinja2
 langcodes>=3.2.0,<4.0.0
 # Official Python utilities
 setuptools
 packaging>=20.0
-typing_extensions>=3.7.4.1,<4.0.0.0; python_version < "3.8"
+typing_extensions>=3.7.4.1,<4.2.0; python_version < "3.8"
 # Development dependencies
 pre-commit>=2.13.0
 cython>=0.25,<3.0
-pytest>=5.2.0
+pytest>=5.2.0,!=7.1.0
 pytest-timeout>=1.3.0,<2.0.0
 mock>=2.0.0,<3.0.0
 flake8>=3.8.0,<3.10.0
 hypothesis>=3.27.0,<7.0.0
-mypy>=0.910
+mypy>=0.910,<0.970
 types-dataclasses>=0.1.3; python_version < "3.7"
 types-mock>=0.1.1
 types-requests
+black>=22.0,<23.0

setup.cfg

@@ -38,18 +38,17 @@ setup_requires =
     cymem>=2.0.2,<2.1.0
     preshed>=3.0.2,<3.1.0
     murmurhash>=0.28.0,<1.1.0
-    thinc>=8.0.12,<8.1.0
+    thinc>=8.1.0,<8.2.0
 install_requires =
     # Our libraries
-    spacy-legacy>=3.0.8,<3.1.0
+    spacy-legacy>=3.0.9,<3.1.0
     spacy-loggers>=1.0.0,<2.0.0
     murmurhash>=0.28.0,<1.1.0
     cymem>=2.0.2,<2.1.0
     preshed>=3.0.2,<3.1.0
-    thinc>=8.0.12,<8.1.0
-    blis>=0.4.0,<0.8.0
-    wasabi>=0.8.1,<1.1.0
-    srsly>=2.4.1,<3.0.0
+    thinc>=8.1.0,<8.2.0
+    wasabi>=0.9.1,<1.1.0
+    srsly>=2.4.3,<3.0.0
     catalogue>=2.0.6,<2.1.0
     typer>=0.3.0,<0.5.0
     pathy>=0.3.5
@@ -57,12 +56,12 @@ install_requires =
     tqdm>=4.38.0,<5.0.0
     numpy>=1.15.0
     requests>=2.13.0,<3.0.0
-    pydantic>=1.7.4,!=1.8,!=1.8.1,<1.9.0
+    pydantic>=1.7.4,!=1.8,!=1.8.1,<1.10.0
     jinja2
     # Official Python utilities
     setuptools
     packaging>=20.0
-    typing_extensions>=3.7.4,<4.0.0.0; python_version < "3.8"
+    typing_extensions>=3.7.4,<4.2.0; python_version < "3.8"
     langcodes>=3.2.0,<4.0.0
 
 [options.entry_points]
@@ -77,37 +76,43 @@ transformers =
 ray =
     spacy_ray>=0.1.0,<1.0.0
 cuda =
-    cupy>=5.0.0b4,<10.0.0
+    cupy>=5.0.0b4,<11.0.0
 cuda80 =
-    cupy-cuda80>=5.0.0b4,<10.0.0
+    cupy-cuda80>=5.0.0b4,<11.0.0
 cuda90 =
-    cupy-cuda90>=5.0.0b4,<10.0.0
+    cupy-cuda90>=5.0.0b4,<11.0.0
 cuda91 =
-    cupy-cuda91>=5.0.0b4,<10.0.0
+    cupy-cuda91>=5.0.0b4,<11.0.0
 cuda92 =
-    cupy-cuda92>=5.0.0b4,<10.0.0
+    cupy-cuda92>=5.0.0b4,<11.0.0
 cuda100 =
-    cupy-cuda100>=5.0.0b4,<10.0.0
+    cupy-cuda100>=5.0.0b4,<11.0.0
 cuda101 =
-    cupy-cuda101>=5.0.0b4,<10.0.0
+    cupy-cuda101>=5.0.0b4,<11.0.0
 cuda102 =
-    cupy-cuda102>=5.0.0b4,<10.0.0
+    cupy-cuda102>=5.0.0b4,<11.0.0
 cuda110 =
-    cupy-cuda110>=5.0.0b4,<10.0.0
+    cupy-cuda110>=5.0.0b4,<11.0.0
 cuda111 =
-    cupy-cuda111>=5.0.0b4,<10.0.0
+    cupy-cuda111>=5.0.0b4,<11.0.0
 cuda112 =
-    cupy-cuda112>=5.0.0b4,<10.0.0
+    cupy-cuda112>=5.0.0b4,<11.0.0
 cuda113 =
-    cupy-cuda113>=5.0.0b4,<10.0.0
+    cupy-cuda113>=5.0.0b4,<11.0.0
 cuda114 =
-    cupy-cuda114>=5.0.0b4,<10.0.0
+    cupy-cuda114>=5.0.0b4,<11.0.0
+cuda115 =
+    cupy-cuda115>=5.0.0b4,<11.0.0
+cuda116 =
+    cupy-cuda116>=5.0.0b4,<11.0.0
+cuda117 =
+    cupy-cuda117>=5.0.0b4,<11.0.0
 apple =
-    thinc-apple-ops>=0.0.4,<1.0.0
+    thinc-apple-ops>=0.1.0.dev0,<1.0.0
 # Language tokenizers with external dependencies
 ja =
-    sudachipy>=0.4.9
-    sudachidict_core>=20200330
+    sudachipy>=0.5.2,!=0.6.1
+    sudachidict_core>=20211220
 ko =
     natto-py==0.9.0
 th =

setup.py

@@ -23,6 +23,7 @@ Options.docstrings = True
 
 PACKAGES = find_packages()
 MOD_NAMES = [
+    "spacy.training.alignment_array",
     "spacy.training.example",
     "spacy.parts_of_speech",
     "spacy.strings",
@@ -33,6 +34,7 @@ MOD_NAMES = [
     "spacy.ml.parser_model",
     "spacy.morphology",
     "spacy.pipeline.dep_parser",
+    "spacy.pipeline._edit_tree_internals.edit_trees",
     "spacy.pipeline.morphologizer",
     "spacy.pipeline.multitask",
     "spacy.pipeline.ner",
@@ -81,7 +83,6 @@ COPY_FILES = {
     ROOT / "setup.cfg": PACKAGE_ROOT / "tests" / "package",
     ROOT / "pyproject.toml": PACKAGE_ROOT / "tests" / "package",
     ROOT / "requirements.txt": PACKAGE_ROOT / "tests" / "package",
-    ROOT / "website" / "meta" / "universe.json": PACKAGE_ROOT / "tests" / "universe",
 }
 
 
@@ -125,6 +126,8 @@ class build_ext_options:
 
 class build_ext_subclass(build_ext, build_ext_options):
     def build_extensions(self):
+        if self.parallel is None and os.environ.get("SPACY_NUM_BUILD_JOBS") is not None:
+            self.parallel = int(os.environ.get("SPACY_NUM_BUILD_JOBS"))
         build_ext_options.build_options(self)
         build_ext.build_extensions(self)
 
@@ -205,7 +208,11 @@ def setup_package():
     for name in MOD_NAMES:
         mod_path = name.replace(".", "/") + ".pyx"
         ext = Extension(
-            name, [mod_path], language="c++", include_dirs=include_dirs, extra_compile_args=["-std=c++11"]
+            name,
+            [mod_path],
+            language="c++",
+            include_dirs=include_dirs,
+            extra_compile_args=["-std=c++11"],
         )
         ext_modules.append(ext)
     print("Cythonizing sources")

spacy/__init__.py

@@ -32,6 +32,7 @@ def load(
     *,
     vocab: Union[Vocab, bool] = True,
     disable: Iterable[str] = util.SimpleFrozenList(),
+    enable: Iterable[str] = util.SimpleFrozenList(),
     exclude: Iterable[str] = util.SimpleFrozenList(),
     config: Union[Dict[str, Any], Config] = util.SimpleFrozenDict(),
 ) -> Language:
@@ -42,6 +43,8 @@ def load(
     disable (Iterable[str]): Names of pipeline components to disable. Disabled
         pipes will be loaded but they won't be run unless you explicitly
         enable them by calling nlp.enable_pipe.
+    enable (Iterable[str]): Names of pipeline components to enable. All other
+        pipes will be disabled (but can be enabled later using nlp.enable_pipe).
     exclude (Iterable[str]): Names of pipeline components to exclude. Excluded
         components won't be loaded.
     config (Dict[str, Any] / Config): Config overrides as nested dict or dict
@@ -49,7 +52,12 @@ def load(
     RETURNS (Language): The loaded nlp object.
     """
     return util.load_model(
-        name, vocab=vocab, disable=disable, exclude=exclude, config=config
+        name,
+        vocab=vocab,
+        disable=disable,
+        enable=enable,
+        exclude=exclude,
+        config=config,
     )
 
 

spacy/about.py

@@ -1,6 +1,6 @@
 # fmt: off
 __title__ = "spacy"
-__version__ = "3.2.1"
+__version__ = "3.4.0"
 __download_url__ = "https://github.com/explosion/spacy-models/releases/download"
 __compatibility__ = "https://raw.githubusercontent.com/explosion/spacy-models/master/compatibility.json"
 __projects__ = "https://github.com/explosion/projects"

spacy/attrs.pyx

@@ -1,3 +1,6 @@
+from .errors import Errors
+
+IOB_STRINGS = ("", "I", "O", "B")
 
 IDS = {
     "": NULL_ATTR,
@@ -64,7 +67,6 @@ IDS = {
     "FLAG61": FLAG61,
     "FLAG62": FLAG62,
     "FLAG63": FLAG63,
-
     "ID": ID,
     "ORTH": ORTH,
     "LOWER": LOWER,
@@ -72,7 +74,6 @@ IDS = {
     "SHAPE": SHAPE,
     "PREFIX": PREFIX,
     "SUFFIX": SUFFIX,
-
     "LENGTH": LENGTH,
     "LEMMA": LEMMA,
     "POS": POS,
@@ -87,7 +88,7 @@ IDS = {
     "SPACY": SPACY,
     "LANG": LANG,
     "MORPH": MORPH,
-    "IDX": IDX
+    "IDX": IDX,
 }
 
 
@@ -109,28 +110,66 @@ def intify_attrs(stringy_attrs, strings_map=None, _do_deprecated=False):
     """
     inty_attrs = {}
     if _do_deprecated:
-        if 'F' in stringy_attrs:
+        if "F" in stringy_attrs:
             stringy_attrs["ORTH"] = stringy_attrs.pop("F")
-        if 'L' in stringy_attrs:
+        if "L" in stringy_attrs:
             stringy_attrs["LEMMA"] = stringy_attrs.pop("L")
-        if 'pos' in stringy_attrs:
+        if "pos" in stringy_attrs:
             stringy_attrs["TAG"] = stringy_attrs.pop("pos")
-        if 'morph' in stringy_attrs:
-            morphs = stringy_attrs.pop('morph')
-        if 'number' in stringy_attrs:
-            stringy_attrs.pop('number')
-        if 'tenspect' in stringy_attrs:
-            stringy_attrs.pop('tenspect')
+        if "morph" in stringy_attrs:
+            morphs = stringy_attrs.pop("morph")
+        if "number" in stringy_attrs:
+            stringy_attrs.pop("number")
+        if "tenspect" in stringy_attrs:
+            stringy_attrs.pop("tenspect")
         morph_keys = [
-            'PunctType', 'PunctSide', 'Other', 'Degree', 'AdvType', 'Number',
-            'VerbForm', 'PronType', 'Aspect', 'Tense', 'PartType', 'Poss',
-            'Hyph', 'ConjType', 'NumType', 'Foreign', 'VerbType', 'NounType',
-            'Gender', 'Mood', 'Negative', 'Tense', 'Voice', 'Abbr',
-            'Derivation', 'Echo', 'Foreign', 'NameType', 'NounType', 'NumForm',
-            'NumValue', 'PartType', 'Polite', 'StyleVariant',
-            'PronType', 'AdjType', 'Person', 'Variant', 'AdpType',
-            'Reflex', 'Negative', 'Mood', 'Aspect', 'Case',
-            'Polarity', 'PrepCase', 'Animacy' # U20
+            "PunctType",
+            "PunctSide",
+            "Other",
+            "Degree",
+            "AdvType",
+            "Number",
+            "VerbForm",
+            "PronType",
+            "Aspect",
+            "Tense",
+            "PartType",
+            "Poss",
+            "Hyph",
+            "ConjType",
+            "NumType",
+            "Foreign",
+            "VerbType",
+            "NounType",
+            "Gender",
+            "Mood",
+            "Negative",
+            "Tense",
+            "Voice",
+            "Abbr",
+            "Derivation",
+            "Echo",
+            "Foreign",
+            "NameType",
+            "NounType",
+            "NumForm",
+            "NumValue",
+            "PartType",
+            "Polite",
+            "StyleVariant",
+            "PronType",
+            "AdjType",
+            "Person",
+            "Variant",
+            "AdpType",
+            "Reflex",
+            "Negative",
+            "Mood",
+            "Aspect",
+            "Case",
+            "Polarity",
+            "PrepCase",
+            "Animacy",  # U20
         ]
         for key in morph_keys:
             if key in stringy_attrs:
@@ -142,8 +181,13 @@ def intify_attrs(stringy_attrs, strings_map=None, _do_deprecated=False):
     for name, value in stringy_attrs.items():
         int_key = intify_attr(name)
         if int_key is not None:
+            if int_key == ENT_IOB:
+                if value in IOB_STRINGS:
+                    value = IOB_STRINGS.index(value)
+                elif isinstance(value, str):
+                    raise ValueError(Errors.E1025.format(value=value))
             if strings_map is not None and isinstance(value, str):
-                if hasattr(strings_map, 'add'):
+                if hasattr(strings_map, "add"):
                     value = strings_map.add(value)
                 else:
                     value = strings_map[value]

spacy/cli/__init__.py

@@ -14,6 +14,7 @@ from .pretrain import pretrain  # noqa: F401
 from .debug_data import debug_data  # noqa: F401
 from .debug_config import debug_config  # noqa: F401
 from .debug_model import debug_model  # noqa: F401
+from .debug_diff import debug_diff  # noqa: F401
 from .evaluate import evaluate  # noqa: F401
 from .convert import convert  # noqa: F401
 from .init_pipeline import init_pipeline_cli  # noqa: F401

spacy/cli/_util.py

@@ -12,7 +12,7 @@ from click.parser import split_arg_string
 from typer.main import get_command
 from contextlib import contextmanager
 from thinc.api import Config, ConfigValidationError, require_gpu
-from thinc.util import has_cupy, gpu_is_available
+from thinc.util import gpu_is_available
 from configparser import InterpolationError
 import os
 
@@ -360,7 +360,7 @@ def download_file(src: Union[str, "Pathy"], dest: Path, *, force: bool = False)
     src = str(src)
     with smart_open.open(src, mode="rb", ignore_ext=True) as input_file:
         with dest.open(mode="wb") as output_file:
-            output_file.write(input_file.read())
+            shutil.copyfileobj(input_file, output_file)
 
 
 def ensure_pathy(path):
@@ -462,6 +462,23 @@ def git_sparse_checkout(repo, subpath, dest, branch):
         shutil.move(str(source_path), str(dest))
 
 
+def git_repo_branch_exists(repo: str, branch: str) -> bool:
+    """Uses 'git ls-remote' to check if a repository and branch exists
+
+    repo (str): URL to get repo.
+    branch (str): Branch on repo to check.
+    RETURNS (bool): True if repo:branch exists.
+    """
+    get_git_version()
+    cmd = f"git ls-remote {repo} {branch}"
+    # We might be tempted to use `--exit-code` with `git ls-remote`, but
+    # `run_command` handles the `returncode` for us, so we'll rely on
+    # the fact that stdout returns '' if the requested branch doesn't exist
+    ret = run_command(cmd, capture=True)
+    exists = ret.stdout != ""
+    return exists
+
+
 def get_git_version(
     error: str = "Could not run 'git'. Make sure it's installed and the executable is available.",
 ) -> Tuple[int, int]:
@@ -554,5 +571,5 @@ def setup_gpu(use_gpu: int, silent=None) -> None:
         require_gpu(use_gpu)
     else:
         local_msg.info("Using CPU")
-        if has_cupy and gpu_is_available():
+        if gpu_is_available():
             local_msg.info("To switch to GPU 0, use the option: --gpu-id 0")

spacy/cli/debug_data.py

@@ -6,19 +6,21 @@ import sys
 import srsly
 from wasabi import Printer, MESSAGES, msg
 import typer
+import math
 
 from ._util import app, Arg, Opt, show_validation_error, parse_config_overrides
 from ._util import import_code, debug_cli
-from ..training import Example
+from ..training import Example, remove_bilu_prefix
 from ..training.initialize import get_sourced_components
 from ..schemas import ConfigSchemaTraining
 from ..pipeline._parser_internals import nonproj
 from ..pipeline._parser_internals.nonproj import DELIMITER
-from ..pipeline import Morphologizer
+from ..pipeline import Morphologizer, SpanCategorizer
 from ..morphology import Morphology
 from ..language import Language
 from ..util import registry, resolve_dot_names
 from ..compat import Literal
+from ..vectors import Mode as VectorsMode
 from .. import util
 
 
@@ -29,6 +31,12 @@ DEP_LABEL_THRESHOLD = 20
 # Minimum number of expected examples to train a new pipeline
 BLANK_MODEL_MIN_THRESHOLD = 100
 BLANK_MODEL_THRESHOLD = 2000
+# Arbitrary threshold where SpanCat performs well
+SPAN_DISTINCT_THRESHOLD = 1
+# Arbitrary threshold where SpanCat performs well
+BOUNDARY_DISTINCT_THRESHOLD = 1
+# Arbitrary threshold for filtering span lengths during reporting (percentage)
+SPAN_LENGTH_THRESHOLD_PERCENTAGE = 90
 
 
 @debug_cli.command(
@@ -170,29 +178,164 @@ def debug_data(
         show=verbose,
     )
     if len(nlp.vocab.vectors):
-        msg.info(
-            f"{len(nlp.vocab.vectors)} vectors ({nlp.vocab.vectors.n_keys} "
-            f"unique keys, {nlp.vocab.vectors_length} dimensions)"
-        )
-        n_missing_vectors = sum(gold_train_data["words_missing_vectors"].values())
-        msg.warn(
-            "{} words in training data without vectors ({:.0f}%)".format(
-                n_missing_vectors,
-                100 * (n_missing_vectors / gold_train_data["n_words"]),
-            ),
-        )
-        msg.text(
-            "10 most common words without vectors: {}".format(
-                _format_labels(
-                    gold_train_data["words_missing_vectors"].most_common(10),
-                    counts=True,
-                )
-            ),
-            show=verbose,
-        )
+        if nlp.vocab.vectors.mode == VectorsMode.floret:
+            msg.info(
+                f"floret vectors with {len(nlp.vocab.vectors)} vectors, "
+                f"{nlp.vocab.vectors_length} dimensions, "
+                f"{nlp.vocab.vectors.minn}-{nlp.vocab.vectors.maxn} char "
+                f"n-gram subwords"
+            )
+        else:
+            msg.info(
+                f"{len(nlp.vocab.vectors)} vectors ({nlp.vocab.vectors.n_keys} "
+                f"unique keys, {nlp.vocab.vectors_length} dimensions)"
+            )
+            n_missing_vectors = sum(gold_train_data["words_missing_vectors"].values())
+            msg.warn(
+                "{} words in training data without vectors ({:.0f}%)".format(
+                    n_missing_vectors,
+                    100 * (n_missing_vectors / gold_train_data["n_words"]),
+                ),
+            )
+            msg.text(
+                "10 most common words without vectors: {}".format(
+                    _format_labels(
+                        gold_train_data["words_missing_vectors"].most_common(10),
+                        counts=True,
+                    )
+                ),
+                show=verbose,
+            )
     else:
         msg.info("No word vectors present in the package")
 
+    if "spancat" in factory_names:
+        model_labels_spancat = _get_labels_from_spancat(nlp)
+        has_low_data_warning = False
+        has_no_neg_warning = False
+
+        msg.divider("Span Categorization")
+        msg.table(model_labels_spancat, header=["Spans Key", "Labels"], divider=True)
+
+        msg.text("Label counts in train data: ", show=verbose)
+        for spans_key, data_labels in gold_train_data["spancat"].items():
+            msg.text(
+                f"Key: {spans_key}, {_format_labels(data_labels.items(), counts=True)}",
+                show=verbose,
+            )
+        # Data checks: only take the spans keys in the actual spancat components
+        data_labels_in_component = {
+            spans_key: gold_train_data["spancat"][spans_key]
+            for spans_key in model_labels_spancat.keys()
+        }
+        for spans_key, data_labels in data_labels_in_component.items():
+            for label, count in data_labels.items():
+                # Check for missing labels
+                spans_key_in_model = spans_key in model_labels_spancat.keys()
+                if (spans_key_in_model) and (
+                    label not in model_labels_spancat[spans_key]
+                ):
+                    msg.warn(
+                        f"Label '{label}' is not present in the model labels of key '{spans_key}'. "
+                        "Performance may degrade after training."
+                    )
+                # Check for low number of examples per label
+                if count <= NEW_LABEL_THRESHOLD:
+                    msg.warn(
+                        f"Low number of examples for label '{label}' in key '{spans_key}' ({count})"
+                    )
+                    has_low_data_warning = True
+                # Check for negative examples
+                with msg.loading("Analyzing label distribution..."):
+                    neg_docs = _get_examples_without_label(
+                        train_dataset, label, "spancat", spans_key
+                    )
+                if neg_docs == 0:
+                    msg.warn(f"No examples for texts WITHOUT new label '{label}'")
+                    has_no_neg_warning = True
+
+            with msg.loading("Obtaining span characteristics..."):
+                span_characteristics = _get_span_characteristics(
+                    train_dataset, gold_train_data, spans_key
+                )
+
+            msg.info(f"Span characteristics for spans_key '{spans_key}'")
+            msg.info("SD = Span Distinctiveness, BD = Boundary Distinctiveness")
+            _print_span_characteristics(span_characteristics)
+
+            _span_freqs = _get_spans_length_freq_dist(
+                gold_train_data["spans_length"][spans_key]
+            )
+            _filtered_span_freqs = _filter_spans_length_freq_dist(
+                _span_freqs, threshold=SPAN_LENGTH_THRESHOLD_PERCENTAGE
+            )
+
+            msg.info(
+                f"Over {SPAN_LENGTH_THRESHOLD_PERCENTAGE}% of spans have lengths of 1 -- "
+                f"{max(_filtered_span_freqs.keys())} "
+                f"(min={span_characteristics['min_length']}, max={span_characteristics['max_length']}). "
+                f"The most common span lengths are: {_format_freqs(_filtered_span_freqs)}. "
+                "If you are using the n-gram suggester, note that omitting "
+                "infrequent n-gram lengths can greatly improve speed and "
+                "memory usage."
+            )
+
+            msg.text(
+                f"Full distribution of span lengths: {_format_freqs(_span_freqs)}",
+                show=verbose,
+            )
+
+            # Add report regarding span characteristics
+            if span_characteristics["avg_sd"] < SPAN_DISTINCT_THRESHOLD:
+                msg.warn("Spans may not be distinct from the rest of the corpus")
+            else:
+                msg.good("Spans are distinct from the rest of the corpus")
+
+            p_spans = span_characteristics["p_spans"].values()
+            all_span_tokens: Counter = sum(p_spans, Counter())
+            most_common_spans = [w for w, _ in all_span_tokens.most_common(10)]
+            msg.text(
+                "10 most common span tokens: {}".format(
+                    _format_labels(most_common_spans)
+                ),
+                show=verbose,
+            )
+
+            # Add report regarding span boundary characteristics
+            if span_characteristics["avg_bd"] < BOUNDARY_DISTINCT_THRESHOLD:
+                msg.warn("Boundary tokens are not distinct from the rest of the corpus")
+            else:
+                msg.good("Boundary tokens are distinct from the rest of the corpus")
+
+            p_bounds = span_characteristics["p_bounds"].values()
+            all_span_bound_tokens: Counter = sum(p_bounds, Counter())
+            most_common_bounds = [w for w, _ in all_span_bound_tokens.most_common(10)]
+            msg.text(
+                "10 most common span boundary tokens: {}".format(
+                    _format_labels(most_common_bounds)
+                ),
+                show=verbose,
+            )
+
+        if has_low_data_warning:
+            msg.text(
+                f"To train a new span type, your data should include at "
+                f"least {NEW_LABEL_THRESHOLD} instances of the new label",
+                show=verbose,
+            )
+        else:
+            msg.good("Good amount of examples for all labels")
+
+        if has_no_neg_warning:
+            msg.text(
+                "Training data should always include examples of spans "
+                "in context, as well as examples without a given span "
+                "type.",
+                show=verbose,
+            )
+        else:
+            msg.good("Examples without ocurrences available for all labels")
+
     if "ner" in factory_names:
         # Get all unique NER labels present in the data
         labels = set(
@@ -203,6 +346,7 @@ def debug_data(
         has_low_data_warning = False
         has_no_neg_warning = False
         has_ws_ents_error = False
+        has_boundary_cross_ents_warning = False
 
         msg.divider("Named Entity Recognition")
         msg.info(f"{len(model_labels)} label(s)")
@@ -217,7 +361,7 @@ def debug_data(
             if label != "-"
         ]
         labels_with_counts = _format_labels(labels_with_counts, counts=True)
-        msg.text(f"Labels in train data: {_format_labels(labels)}", show=verbose)
+        msg.text(f"Labels in train data: {labels_with_counts}", show=verbose)
         missing_labels = model_labels - labels
         if missing_labels:
             msg.warn(
@@ -237,17 +381,25 @@ def debug_data(
                 has_low_data_warning = True
 
                 with msg.loading("Analyzing label distribution..."):
-                    neg_docs = _get_examples_without_label(train_dataset, label)
+                    neg_docs = _get_examples_without_label(train_dataset, label, "ner")
                 if neg_docs == 0:
                     msg.warn(f"No examples for texts WITHOUT new label '{label}'")
                     has_no_neg_warning = True
 
+        if gold_train_data["boundary_cross_ents"]:
+            msg.warn(
+                f"{gold_train_data['boundary_cross_ents']} entity span(s) crossing sentence boundaries"
+            )
+            has_boundary_cross_ents_warning = True
+
         if not has_low_data_warning:
             msg.good("Good amount of examples for all labels")
         if not has_no_neg_warning:
             msg.good("Examples without occurrences available for all labels")
         if not has_ws_ents_error:
             msg.good("No entities consisting of or starting/ending with whitespace")
+        if not has_boundary_cross_ents_warning:
+            msg.good("No entities crossing sentence boundaries")
 
         if has_low_data_warning:
             msg.text(
@@ -564,7 +716,12 @@ def _compile_gold(
         "deps": Counter(),
         "words": Counter(),
         "roots": Counter(),
+        "spancat": dict(),
+        "spans_length": dict(),
+        "spans_per_type": dict(),
+        "sb_per_type": dict(),
         "ws_ents": 0,
+        "boundary_cross_ents": 0,
         "n_words": 0,
         "n_misaligned_words": 0,
         "words_missing_vectors": Counter(),
@@ -593,6 +750,7 @@ def _compile_gold(
                 if nlp.vocab.strings[word] not in nlp.vocab.vectors:
                     data["words_missing_vectors"].update([word])
         if "ner" in factory_names:
+            sent_starts = eg.get_aligned_sent_starts()
             for i, label in enumerate(eg.get_aligned_ner()):
                 if label is None:
                     continue
@@ -600,10 +758,66 @@ def _compile_gold(
                     # "Illegal" whitespace entity
                     data["ws_ents"] += 1
                 if label.startswith(("B-", "U-")):
-                    combined_label = label.split("-")[1]
+                    combined_label = remove_bilu_prefix(label)
                     data["ner"][combined_label] += 1
+                if sent_starts[i] and label.startswith(("I-", "L-")):
+                    data["boundary_cross_ents"] += 1
                 elif label == "-":
                     data["ner"]["-"] += 1
+        if "spancat" in factory_names:
+            for spans_key in list(eg.reference.spans.keys()):
+                # Obtain the span frequency
+                if spans_key not in data["spancat"]:
+                    data["spancat"][spans_key] = Counter()
+                for i, span in enumerate(eg.reference.spans[spans_key]):
+                    if span.label_ is None:
+                        continue
+                    else:
+                        data["spancat"][spans_key][span.label_] += 1
+
+                # Obtain the span length
+                if spans_key not in data["spans_length"]:
+                    data["spans_length"][spans_key] = dict()
+                for span in gold.spans[spans_key]:
+                    if span.label_ is None:
+                        continue
+                    if span.label_ not in data["spans_length"][spans_key]:
+                        data["spans_length"][spans_key][span.label_] = []
+                    data["spans_length"][spans_key][span.label_].append(len(span))
+
+                # Obtain spans per span type
+                if spans_key not in data["spans_per_type"]:
+                    data["spans_per_type"][spans_key] = dict()
+                for span in gold.spans[spans_key]:
+                    if span.label_ not in data["spans_per_type"][spans_key]:
+                        data["spans_per_type"][spans_key][span.label_] = []
+                    data["spans_per_type"][spans_key][span.label_].append(span)
+
+                # Obtain boundary tokens per span type
+                window_size = 1
+                if spans_key not in data["sb_per_type"]:
+                    data["sb_per_type"][spans_key] = dict()
+                for span in gold.spans[spans_key]:
+                    if span.label_ not in data["sb_per_type"][spans_key]:
+                        # Creating a data structure that holds the start and
+                        # end tokens for each span type
+                        data["sb_per_type"][spans_key][span.label_] = {
+                            "start": [],
+                            "end": [],
+                        }
+                    for offset in range(window_size):
+                        sb_start_idx = span.start - (offset + 1)
+                        if sb_start_idx >= 0:
+                            data["sb_per_type"][spans_key][span.label_]["start"].append(
+                                gold[sb_start_idx : sb_start_idx + 1]
+                            )
+
+                        sb_end_idx = span.end + (offset + 1)
+                        if sb_end_idx <= len(gold):
+                            data["sb_per_type"][spans_key][span.label_]["end"].append(
+                                gold[sb_end_idx - 1 : sb_end_idx]
+                            )
+
         if "textcat" in factory_names or "textcat_multilabel" in factory_names:
             data["cats"].update(gold.cats)
             if any(val not in (0, 1) for val in gold.cats.values()):
@@ -674,21 +888,222 @@ def _format_labels(
     return ", ".join([f"'{l}'" for l in cast(Iterable[str], labels)])
 
 
-def _get_examples_without_label(data: Sequence[Example], label: str) -> int:
+def _format_freqs(freqs: Dict[int, float], sort: bool = True) -> str:
+    if sort:
+        freqs = dict(sorted(freqs.items()))
+
+    _freqs = [(str(k), v) for k, v in freqs.items()]
+    return ", ".join(
+        [f"{l} ({c}%)" for l, c in cast(Iterable[Tuple[str, float]], _freqs)]
+    )
+
+
+def _get_examples_without_label(
+    data: Sequence[Example],
+    label: str,
+    component: Literal["ner", "spancat"] = "ner",
+    spans_key: Optional[str] = "sc",
+) -> int:
     count = 0
     for eg in data:
-        labels = [
-            label.split("-")[1]
-            for label in eg.get_aligned_ner()
-            if label not in ("O", "-", None)
-        ]
+        if component == "ner":
+            labels = [
+                remove_bilu_prefix(label)
+                for label in eg.get_aligned_ner()
+                if label not in ("O", "-", None)
+            ]
+
+        if component == "spancat":
+            labels = (
+                [span.label_ for span in eg.reference.spans[spans_key]]
+                if spans_key in eg.reference.spans
+                else []
+            )
+
         if label not in labels:
             count += 1
     return count
 
 
-def _get_labels_from_model(nlp: Language, pipe_name: str) -> Set[str]:
-    if pipe_name not in nlp.pipe_names:
-        return set()
-    pipe = nlp.get_pipe(pipe_name)
-    return set(pipe.labels)
+def _get_labels_from_model(nlp: Language, factory_name: str) -> Set[str]:
+    pipe_names = [
+        pipe_name
+        for pipe_name in nlp.pipe_names
+        if nlp.get_pipe_meta(pipe_name).factory == factory_name
+    ]
+    labels: Set[str] = set()
+    for pipe_name in pipe_names:
+        pipe = nlp.get_pipe(pipe_name)
+        labels.update(pipe.labels)
+    return labels
+
+
+def _get_labels_from_spancat(nlp: Language) -> Dict[str, Set[str]]:
+    pipe_names = [
+        pipe_name
+        for pipe_name in nlp.pipe_names
+        if nlp.get_pipe_meta(pipe_name).factory == "spancat"
+    ]
+    labels: Dict[str, Set[str]] = {}
+    for pipe_name in pipe_names:
+        pipe = nlp.get_pipe(pipe_name)
+        assert isinstance(pipe, SpanCategorizer)
+        if pipe.key not in labels:
+            labels[pipe.key] = set()
+        labels[pipe.key].update(pipe.labels)
+    return labels
+
+
+def _gmean(l: List) -> float:
+    """Compute geometric mean of a list"""
+    return math.exp(math.fsum(math.log(i) for i in l) / len(l))
+
+
+def _wgt_average(metric: Dict[str, float], frequencies: Counter) -> float:
+    total = sum(value * frequencies[span_type] for span_type, value in metric.items())
+    return total / sum(frequencies.values())
+
+
+def _get_distribution(docs, normalize: bool = True) -> Counter:
+    """Get the frequency distribution given a set of Docs"""
+    word_counts: Counter = Counter()
+    for doc in docs:
+        for token in doc:
+            # Normalize the text
+            t = token.text.lower().replace("``", '"').replace("''", '"')
+            word_counts[t] += 1
+    if normalize:
+        total = sum(word_counts.values(), 0.0)
+        word_counts = Counter({k: v / total for k, v in word_counts.items()})
+    return word_counts
+
+
+def _get_kl_divergence(p: Counter, q: Counter) -> float:
+    """Compute the Kullback-Leibler divergence from two frequency distributions"""
+    total = 0.0
+    for word, p_word in p.items():
+        total += p_word * math.log(p_word / q[word])
+    return total
+
+
+def _format_span_row(span_data: List[Dict], labels: List[str]) -> List[Any]:
+    """Compile into one list for easier reporting"""
+    d = {
+        label: [label] + list(round(d[label], 2) for d in span_data) for label in labels
+    }
+    return list(d.values())
+
+
+def _get_span_characteristics(
+    examples: List[Example], compiled_gold: Dict[str, Any], spans_key: str
+) -> Dict[str, Any]:
+    """Obtain all span characteristics"""
+    data_labels = compiled_gold["spancat"][spans_key]
+    # Get lengths
+    span_length = {
+        label: _gmean(l)
+        for label, l in compiled_gold["spans_length"][spans_key].items()
+    }
+    min_lengths = [min(l) for l in compiled_gold["spans_length"][spans_key].values()]
+    max_lengths = [max(l) for l in compiled_gold["spans_length"][spans_key].values()]
+
+    # Get relevant distributions: corpus, spans, span boundaries
+    p_corpus = _get_distribution([eg.reference for eg in examples], normalize=True)
+    p_spans = {
+        label: _get_distribution(spans, normalize=True)
+        for label, spans in compiled_gold["spans_per_type"][spans_key].items()
+    }
+    p_bounds = {
+        label: _get_distribution(sb["start"] + sb["end"], normalize=True)
+        for label, sb in compiled_gold["sb_per_type"][spans_key].items()
+    }
+
+    # Compute for actual span characteristics
+    span_distinctiveness = {
+        label: _get_kl_divergence(freq_dist, p_corpus)
+        for label, freq_dist in p_spans.items()
+    }
+    sb_distinctiveness = {
+        label: _get_kl_divergence(freq_dist, p_corpus)
+        for label, freq_dist in p_bounds.items()
+    }
+
+    return {
+        "sd": span_distinctiveness,
+        "bd": sb_distinctiveness,
+        "lengths": span_length,
+        "min_length": min(min_lengths),
+        "max_length": max(max_lengths),
+        "avg_sd": _wgt_average(span_distinctiveness, data_labels),
+        "avg_bd": _wgt_average(sb_distinctiveness, data_labels),
+        "avg_length": _wgt_average(span_length, data_labels),
+        "labels": list(data_labels.keys()),
+        "p_spans": p_spans,
+        "p_bounds": p_bounds,
+    }
+
+
+def _print_span_characteristics(span_characteristics: Dict[str, Any]):
+    """Print all span characteristics into a table"""
+    headers = ("Span Type", "Length", "SD", "BD")
+    # Prepare table data with all span characteristics
+    table_data = [
+        span_characteristics["lengths"],
+        span_characteristics["sd"],
+        span_characteristics["bd"],
+    ]
+    table = _format_span_row(
+        span_data=table_data, labels=span_characteristics["labels"]
+    )
+    # Prepare table footer with weighted averages
+    footer_data = [
+        span_characteristics["avg_length"],
+        span_characteristics["avg_sd"],
+        span_characteristics["avg_bd"],
+    ]
+    footer = ["Wgt. Average"] + [str(round(f, 2)) for f in footer_data]
+    msg.table(table, footer=footer, header=headers, divider=True)
+
+
+def _get_spans_length_freq_dist(
+    length_dict: Dict, threshold=SPAN_LENGTH_THRESHOLD_PERCENTAGE
+) -> Dict[int, float]:
+    """Get frequency distribution of spans length under a certain threshold"""
+    all_span_lengths = []
+    for _, lengths in length_dict.items():
+        all_span_lengths.extend(lengths)
+
+    freq_dist: Counter = Counter()
+    for i in all_span_lengths:
+        if freq_dist.get(i):
+            freq_dist[i] += 1
+        else:
+            freq_dist[i] = 1
+
+    # We will be working with percentages instead of raw counts
+    freq_dist_percentage = {}
+    for span_length, count in freq_dist.most_common():
+        percentage = (count / len(all_span_lengths)) * 100.0
+        percentage = round(percentage, 2)
+        freq_dist_percentage[span_length] = percentage
+
+    return freq_dist_percentage
+
+
+def _filter_spans_length_freq_dist(
+    freq_dist: Dict[int, float], threshold: int
+) -> Dict[int, float]:
+    """Filter frequency distribution with respect to a threshold
+
+    We're going to filter all the span lengths that fall
+    around a percentage threshold when summed.
+    """
+    total = 0.0
+    filtered_freq_dist = {}
+    for span_length, dist in freq_dist.items():
+        if total >= threshold:
+            break
+        else:
+            filtered_freq_dist[span_length] = dist
+        total += dist
+    return filtered_freq_dist

spacy/cli/debug_diff.py

@@ -0,0 +1,89 @@
+from typing import Optional
+
+import typer
+from wasabi import Printer, diff_strings, MarkdownRenderer
+from pathlib import Path
+from thinc.api import Config
+
+from ._util import debug_cli, Arg, Opt, show_validation_error, parse_config_overrides
+from ..util import load_config
+from .init_config import init_config, Optimizations
+
+
+@debug_cli.command(
+    "diff-config",
+    context_settings={"allow_extra_args": True, "ignore_unknown_options": True},
+)
+def debug_diff_cli(
+    # fmt: off
+    ctx: typer.Context,
+    config_path: Path = Arg(..., help="Path to config file", exists=True, allow_dash=True),
+    compare_to: Optional[Path] = Opt(None, help="Path to a config file to diff against, or `None` to compare against default settings", exists=True, allow_dash=True),
+    optimize: Optimizations = Opt(Optimizations.efficiency.value, "--optimize", "-o", help="Whether the user config was optimized for efficiency or accuracy. Only relevant when comparing against the default config."),
+    gpu: bool = Opt(False, "--gpu", "-G", help="Whether the original config can run on a GPU. Only relevant when comparing against the default config."),
+    pretraining: bool = Opt(False, "--pretraining", "--pt", help="Whether to compare on a config with pretraining involved. Only relevant when comparing against the default config."),
+    markdown: bool = Opt(False, "--markdown", "-md", help="Generate Markdown for GitHub issues")
+    # fmt: on
+):
+    """Show a diff of a config file with respect to spaCy's defaults or another config file. If
+    additional settings were used in the creation of the config file, then you
+    must supply these as extra parameters to the command when comparing to the default settings. The generated diff
+    can also be used when posting to the discussion forum to provide more
+    information for the maintainers.
+
+    The `optimize`, `gpu`, and `pretraining` options are only relevant when
+    comparing against the default configuration (or specifically when `compare_to` is None).
+
+    DOCS: https://spacy.io/api/cli#debug-diff
+    """
+    debug_diff(
+        config_path=config_path,
+        compare_to=compare_to,
+        gpu=gpu,
+        optimize=optimize,
+        pretraining=pretraining,
+        markdown=markdown,
+    )
+
+
+def debug_diff(
+    config_path: Path,
+    compare_to: Optional[Path],
+    gpu: bool,
+    optimize: Optimizations,
+    pretraining: bool,
+    markdown: bool,
+):
+    msg = Printer()
+    with show_validation_error(hint_fill=False):
+        user_config = load_config(config_path)
+        if compare_to:
+            other_config = load_config(compare_to)
+        else:
+            # Recreate a default config based from user's config
+            lang = user_config["nlp"]["lang"]
+            pipeline = list(user_config["nlp"]["pipeline"])
+            msg.info(f"Found user-defined language: '{lang}'")
+            msg.info(f"Found user-defined pipelines: {pipeline}")
+            other_config = init_config(
+                lang=lang,
+                pipeline=pipeline,
+                optimize=optimize.value,
+                gpu=gpu,
+                pretraining=pretraining,
+                silent=True,
+            )
+
+    user = user_config.to_str()
+    other = other_config.to_str()
+
+    if user == other:
+        msg.warn("No diff to show: configs are identical")
+    else:
+        diff_text = diff_strings(other, user, add_symbols=markdown)
+        if markdown:
+            md = MarkdownRenderer()
+            md.add(md.code_block(diff_text, "diff"))
+            print(md.text)
+        else:
+            print(diff_text)

spacy/cli/init_config.py

@@ -10,6 +10,7 @@ from jinja2 import Template
 from .. import util
 from ..language import DEFAULT_CONFIG_PRETRAIN_PATH
 from ..schemas import RecommendationSchema
+from ..util import SimpleFrozenList
 from ._util import init_cli, Arg, Opt, show_validation_error, COMMAND
 from ._util import string_to_list, import_code
 
@@ -24,16 +25,30 @@ class Optimizations(str, Enum):
     accuracy = "accuracy"
 
 
+class InitValues:
+    """
+    Default values for initialization. Dedicated class to allow synchronized default values for init_config_cli() and
+    init_config(), i.e. initialization calls via CLI respectively Python.
+    """
+
+    lang = "en"
+    pipeline = SimpleFrozenList(["tagger", "parser", "ner"])
+    optimize = Optimizations.efficiency
+    gpu = False
+    pretraining = False
+    force_overwrite = False
+
+
 @init_cli.command("config")
 def init_config_cli(
     # fmt: off
     output_file: Path = Arg(..., help="File to save the config to or - for stdout (will only output config and no additional logging info)", allow_dash=True),
-    lang: str = Opt("en", "--lang", "-l", help="Two-letter code of the language to use"),
-    pipeline: str = Opt("tagger,parser,ner", "--pipeline", "-p", help="Comma-separated names of trainable pipeline components to include (without 'tok2vec' or 'transformer')"),
-    optimize: Optimizations = Opt(Optimizations.efficiency.value, "--optimize", "-o", help="Whether to optimize for efficiency (faster inference, smaller model, lower memory consumption) or higher accuracy (potentially larger and slower model). This will impact the choice of architecture, pretrained weights and related hyperparameters."),
-    gpu: bool = Opt(False, "--gpu", "-G", help="Whether the model can run on GPU. This will impact the choice of architecture, pretrained weights and related hyperparameters."),
-    pretraining: bool = Opt(False, "--pretraining", "-pt", help="Include config for pretraining (with 'spacy pretrain')"),
-    force_overwrite: bool = Opt(False, "--force", "-F", help="Force overwriting the output file"),
+    lang: str = Opt(InitValues.lang, "--lang", "-l", help="Two-letter code of the language to use"),
+    pipeline: str = Opt(",".join(InitValues.pipeline), "--pipeline", "-p", help="Comma-separated names of trainable pipeline components to include (without 'tok2vec' or 'transformer')"),
+    optimize: Optimizations = Opt(InitValues.optimize, "--optimize", "-o", help="Whether to optimize for efficiency (faster inference, smaller model, lower memory consumption) or higher accuracy (potentially larger and slower model). This will impact the choice of architecture, pretrained weights and related hyperparameters."),
+    gpu: bool = Opt(InitValues.gpu, "--gpu", "-G", help="Whether the model can run on GPU. This will impact the choice of architecture, pretrained weights and related hyperparameters."),
+    pretraining: bool = Opt(InitValues.pretraining, "--pretraining", "-pt", help="Include config for pretraining (with 'spacy pretrain')"),
+    force_overwrite: bool = Opt(InitValues.force_overwrite, "--force", "-F", help="Force overwriting the output file"),
     # fmt: on
 ):
     """
@@ -133,11 +148,11 @@ def fill_config(
 
 def init_config(
     *,
-    lang: str,
-    pipeline: List[str],
-    optimize: str,
-    gpu: bool,
-    pretraining: bool = False,
+    lang: str = InitValues.lang,
+    pipeline: List[str] = InitValues.pipeline,
+    optimize: str = InitValues.optimize,
+    gpu: bool = InitValues.gpu,
+    pretraining: bool = InitValues.pretraining,
     silent: bool = True,
 ) -> Config:
     msg = Printer(no_print=silent)

spacy/cli/package.py

@@ -7,6 +7,7 @@ from collections import defaultdict
 from catalogue import RegistryError
 import srsly
 import sys
+import re
 
 from ._util import app, Arg, Opt, string_to_list, WHEEL_SUFFIX, SDIST_SUFFIX
 from ..schemas import validate, ModelMetaSchema
@@ -109,6 +110,24 @@ def package(
             ", ".join(meta["requirements"]),
         )
     if name is not None:
+        if not name.isidentifier():
+            msg.fail(
+                f"Model name ('{name}') is not a valid module name. "
+                "This is required so it can be imported as a module.",
+                "We recommend names that use ASCII A-Z, a-z, _ (underscore), "
+                "and 0-9. "
+                "For specific details see: https://docs.python.org/3/reference/lexical_analysis.html#identifiers",
+                exits=1,
+            )
+        if not _is_permitted_package_name(name):
+            msg.fail(
+                f"Model name ('{name}') is not a permitted package name. "
+                "This is required to correctly load the model with spacy.load.",
+                "We recommend names that use ASCII A-Z, a-z, _ (underscore), "
+                "and 0-9. "
+                "For specific details see: https://www.python.org/dev/peps/pep-0426/#name",
+                exits=1,
+            )
         meta["name"] = name
     if version is not None:
         meta["version"] = version
@@ -162,7 +181,7 @@ def package(
         imports="\n".join(f"from . import {m}" for m in imports)
     )
     create_file(package_path / "__init__.py", init_py)
-    msg.good(f"Successfully created package '{model_name_v}'", main_path)
+    msg.good(f"Successfully created package directory '{model_name_v}'", main_path)
     if create_sdist:
         with util.working_dir(main_path):
             util.run_command([sys.executable, "setup.py", "sdist"], capture=False)
@@ -171,8 +190,14 @@ def package(
     if create_wheel:
         with util.working_dir(main_path):
             util.run_command([sys.executable, "setup.py", "bdist_wheel"], capture=False)
-        wheel = main_path / "dist" / f"{model_name_v}{WHEEL_SUFFIX}"
+        wheel_name_squashed = re.sub("_+", "_", model_name_v)
+        wheel = main_path / "dist" / f"{wheel_name_squashed}{WHEEL_SUFFIX}"
         msg.good(f"Successfully created binary wheel", wheel)
+    if "__" in model_name:
+        msg.warn(
+            f"Model name ('{model_name}') contains a run of underscores. "
+            "Runs of underscores are not significant in installed package names.",
+        )
 
 
 def has_wheel() -> bool:
@@ -422,6 +447,14 @@ def _format_label_scheme(data: Dict[str, Any]) -> str:
     return md.text
 
 
+def _is_permitted_package_name(package_name: str) -> bool:
+    # regex from: https://www.python.org/dev/peps/pep-0426/#name
+    permitted_match = re.search(
+        r"^([A-Z0-9]|[A-Z0-9][A-Z0-9._-]*[A-Z0-9])$", package_name, re.IGNORECASE
+    )
+    return permitted_match is not None
+
+
 TEMPLATE_SETUP = """
 #!/usr/bin/env python
 import io

spacy/cli/project/assets.py

@@ -1,6 +1,7 @@
 from typing import Any, Dict, Optional
 from pathlib import Path
 from wasabi import msg
+import os
 import re
 import shutil
 import requests
@@ -11,6 +12,9 @@ from .._util import project_cli, Arg, Opt, PROJECT_FILE, load_project_config
 from .._util import get_checksum, download_file, git_checkout, get_git_version
 from .._util import SimpleFrozenDict, parse_config_overrides
 
+# Whether assets are extra if `extra` is not set.
+EXTRA_DEFAULT = False
+
 
 @project_cli.command(
     "assets",
@@ -20,7 +24,8 @@ def project_assets_cli(
     # fmt: off
     ctx: typer.Context,  # This is only used to read additional arguments
     project_dir: Path = Arg(Path.cwd(), help="Path to cloned project. Defaults to current working directory.", exists=True, file_okay=False),
-    sparse_checkout: bool = Opt(False, "--sparse", "-S", help="Use sparse checkout for assets provided via Git, to only check out and clone the files needed. Requires Git v22.2+.")
+    sparse_checkout: bool = Opt(False, "--sparse", "-S", help="Use sparse checkout for assets provided via Git, to only check out and clone the files needed. Requires Git v22.2+."),
+    extra: bool = Opt(False, "--extra", "-e", help="Download all assets, including those marked as 'extra'.")
     # fmt: on
 ):
     """Fetch project assets like datasets and pretrained weights. Assets are
@@ -31,7 +36,12 @@ def project_assets_cli(
     DOCS: https://spacy.io/api/cli#project-assets
     """
     overrides = parse_config_overrides(ctx.args)
-    project_assets(project_dir, overrides=overrides, sparse_checkout=sparse_checkout)
+    project_assets(
+        project_dir,
+        overrides=overrides,
+        sparse_checkout=sparse_checkout,
+        extra=extra,
+    )
 
 
 def project_assets(
@@ -39,17 +49,29 @@ def project_assets(
     *,
     overrides: Dict[str, Any] = SimpleFrozenDict(),
     sparse_checkout: bool = False,
+    extra: bool = False,
 ) -> None:
     """Fetch assets for a project using DVC if possible.
 
     project_dir (Path): Path to project directory.
+    sparse_checkout (bool): Use sparse checkout for assets provided via Git, to only check out and clone the files
+                            needed.
+    extra (bool): Whether to download all assets, including those marked as 'extra'.
     """
     project_path = ensure_path(project_dir)
     config = load_project_config(project_path, overrides=overrides)
-    assets = config.get("assets", {})
+    assets = [
+        asset
+        for asset in config.get("assets", [])
+        if extra or not asset.get("extra", EXTRA_DEFAULT)
+    ]
     if not assets:
-        msg.warn(f"No assets specified in {PROJECT_FILE}", exits=0)
+        msg.warn(
+            f"No assets specified in {PROJECT_FILE} (if assets are marked as extra, download them with --extra)",
+            exits=0,
+        )
     msg.info(f"Fetching {len(assets)} asset(s)")
+
     for asset in assets:
         dest = (project_dir / asset["dest"]).resolve()
         checksum = asset.get("checksum")
@@ -129,10 +151,17 @@ def fetch_asset(
         the asset failed.
     """
     dest_path = (project_path / dest).resolve()
-    if dest_path.exists() and checksum:
+    if dest_path.exists():
         # If there's already a file, check for checksum
-        if checksum == get_checksum(dest_path):
-            msg.good(f"Skipping download with matching checksum: {dest}")
+        if checksum:
+            if checksum == get_checksum(dest_path):
+                msg.good(f"Skipping download with matching checksum: {dest}")
+                return
+        else:
+            # If there's not a checksum, make sure the file is a possibly valid size
+            if os.path.getsize(dest_path) == 0:
+                msg.warn(f"Asset exists but with size of 0 bytes, deleting: {dest}")
+                os.remove(dest_path)
     # We might as well support the user here and create parent directories in
     # case the asset dir isn't listed as a dir to create in the project.yml
     if not dest_path.parent.exists():

spacy/cli/project/clone.py

@@ -7,11 +7,11 @@ import re
 from ... import about
 from ...util import ensure_path
 from .._util import project_cli, Arg, Opt, COMMAND, PROJECT_FILE
-from .._util import git_checkout, get_git_version
+from .._util import git_checkout, get_git_version, git_repo_branch_exists
 
 DEFAULT_REPO = about.__projects__
 DEFAULT_PROJECTS_BRANCH = about.__projects_branch__
-DEFAULT_BRANCH = "master"
+DEFAULT_BRANCHES = ["main", "master"]
 
 
 @project_cli.command("clone")
@@ -20,7 +20,7 @@ def project_clone_cli(
     name: str = Arg(..., help="The name of the template to clone"),
     dest: Optional[Path] = Arg(None, help="Where to clone the project. Defaults to current working directory", exists=False),
     repo: str = Opt(DEFAULT_REPO, "--repo", "-r", help="The repository to clone from"),
-    branch: Optional[str] = Opt(None, "--branch", "-b", help="The branch to clone from"),
+    branch: Optional[str] = Opt(None, "--branch", "-b", help=f"The branch to clone from. If not provided, will attempt {', '.join(DEFAULT_BRANCHES)}"),
     sparse_checkout: bool = Opt(False, "--sparse", "-S", help="Use sparse Git checkout to only check out and clone the files needed. Requires Git v22.2+.")
     # fmt: on
 ):
@@ -33,9 +33,25 @@ def project_clone_cli(
     """
     if dest is None:
         dest = Path.cwd() / Path(name).parts[-1]
+    if repo == DEFAULT_REPO and branch is None:
+        branch = DEFAULT_PROJECTS_BRANCH
+
     if branch is None:
-        # If it's a user repo, we want to default to other branch
-        branch = DEFAULT_PROJECTS_BRANCH if repo == DEFAULT_REPO else DEFAULT_BRANCH
+        for default_branch in DEFAULT_BRANCHES:
+            if git_repo_branch_exists(repo, default_branch):
+                branch = default_branch
+                break
+        if branch is None:
+            default_branches_msg = ", ".join(f"'{b}'" for b in DEFAULT_BRANCHES)
+            msg.fail(
+                "No branch provided and attempted default "
+                f"branches {default_branches_msg} do not exist.",
+                exits=1,
+            )
+    else:
+        if not git_repo_branch_exists(repo, branch):
+            msg.fail(f"repo: {repo} (branch: {branch}) does not exist.", exits=1)
+    assert isinstance(branch, str)
     project_clone(name, dest, repo=repo, branch=branch, sparse_checkout=sparse_checkout)
 
 
@@ -61,9 +77,9 @@ def project_clone(
     try:
         git_checkout(repo, name, dest, branch=branch, sparse=sparse_checkout)
     except subprocess.CalledProcessError:
-        err = f"Could not clone '{name}' from repo '{repo_name}'"
+        err = f"Could not clone '{name}' from repo '{repo_name}' (branch '{branch}')"
         msg.fail(err, exits=1)
-    msg.good(f"Cloned '{name}' from {repo_name}", project_dir)
+    msg.good(f"Cloned '{name}' from '{repo_name}' (branch '{branch}')", project_dir)
     if not (project_dir / PROJECT_FILE).exists():
         msg.warn(f"No {PROJECT_FILE} found in directory")
     else:

spacy/cli/templates/quickstart_training.jinja

@@ -3,9 +3,15 @@ the docs and the init config command. It encodes various best practices and
 can help generate the best possible configuration, given a user's requirements. #}
 {%- set use_transformer = hardware != "cpu" -%}
 {%- set transformer = transformer_data[optimize] if use_transformer else {} -%}
+{%- set listener_components = ["tagger", "morphologizer", "parser", "ner", "textcat", "textcat_multilabel", "entity_linker", "spancat", "trainable_lemmatizer"] -%}
 [paths]
 train = null
 dev = null
+{% if use_transformer or optimize == "efficiency" or not word_vectors -%}
+vectors = null
+{% else -%}
+vectors = "{{ word_vectors }}"
+{% endif -%}
 
 [system]
 {% if use_transformer -%}
@@ -19,10 +25,10 @@ lang = "{{ lang }}"
 {%- set has_textcat = ("textcat" in components or "textcat_multilabel" in components) -%}
 {%- set with_accuracy = optimize == "accuracy" -%}
 {%- set has_accurate_textcat = has_textcat and with_accuracy -%}
-{%- if ("tagger" in components or "morphologizer" in components or "parser" in components or "ner" in components or "entity_linker" in components or has_accurate_textcat) -%}
-{%- set full_pipeline = ["transformer" if use_transformer else "tok2vec"] + components %}
+{%- if ("tagger" in components or "morphologizer" in components or "parser" in components or "ner" in components or "spancat" in components or "trainable_lemmatizer" in components or "entity_linker" in components or has_accurate_textcat) -%}
+{%- set full_pipeline = ["transformer" if use_transformer else "tok2vec"] + components -%}
 {%- else -%}
-{%- set full_pipeline = components %}
+{%- set full_pipeline = components -%}
 {%- endif %}
 pipeline = {{ full_pipeline|pprint()|replace("'", '"')|safe }}
 batch_size = {{ 128 if hardware == "gpu" else 1000 }}
@@ -49,7 +55,7 @@ stride = 96
 factory = "morphologizer"
 
 [components.morphologizer.model]
-@architectures = "spacy.Tagger.v1"
+@architectures = "spacy.Tagger.v2"
 nO = null
 
 [components.morphologizer.model.tok2vec]
@@ -65,7 +71,7 @@ grad_factor = 1.0
 factory = "tagger"
 
 [components.tagger.model]
-@architectures = "spacy.Tagger.v1"
+@architectures = "spacy.Tagger.v2"
 nO = null
 
 [components.tagger.model.tok2vec]
@@ -118,6 +124,60 @@ grad_factor = 1.0
 @layers = "reduce_mean.v1"
 {% endif -%}
 
+{% if "spancat" in components -%}
+[components.spancat]
+factory = "spancat"
+max_positive = null
+scorer = {"@scorers":"spacy.spancat_scorer.v1"}
+spans_key = "sc"
+threshold = 0.5
+
+[components.spancat.model]
+@architectures = "spacy.SpanCategorizer.v1"
+
+[components.spancat.model.reducer]
+@layers = "spacy.mean_max_reducer.v1"
+hidden_size = 128
+
+[components.spancat.model.scorer]
+@layers = "spacy.LinearLogistic.v1"
+nO = null
+nI = null
+
+[components.spancat.model.tok2vec]
+@architectures = "spacy-transformers.TransformerListener.v1"
+grad_factor = 1.0
+
+[components.spancat.model.tok2vec.pooling]
+@layers = "reduce_mean.v1"
+
+[components.spancat.suggester]
+@misc = "spacy.ngram_suggester.v1"
+sizes = [1,2,3]
+{% endif -%}
+
+{% if "trainable_lemmatizer" in components -%}
+[components.trainable_lemmatizer]
+factory = "trainable_lemmatizer"
+backoff = "orth"
+min_tree_freq = 3
+overwrite = false
+scorer = {"@scorers":"spacy.lemmatizer_scorer.v1"}
+top_k = 1
+
+[components.trainable_lemmatizer.model]
+@architectures = "spacy.Tagger.v2"
+nO = null
+normalize = false
+
+[components.trainable_lemmatizer.model.tok2vec]
+@architectures = "spacy-transformers.TransformerListener.v1"
+grad_factor = 1.0
+
+[components.trainable_lemmatizer.model.tok2vec.pooling]
+@layers = "reduce_mean.v1"
+{% endif -%}
+
 {% if "entity_linker" in components -%}
 [components.entity_linker]
 factory = "entity_linker"
@@ -126,7 +186,7 @@ incl_context = true
 incl_prior = true
 
 [components.entity_linker.model]
-@architectures = "spacy.EntityLinker.v1"
+@architectures = "spacy.EntityLinker.v2"
 nO = null
 
 [components.entity_linker.model.tok2vec]
@@ -233,7 +293,7 @@ maxout_pieces = 3
 factory = "morphologizer"
 
 [components.morphologizer.model]
-@architectures = "spacy.Tagger.v1"
+@architectures = "spacy.Tagger.v2"
 nO = null
 
 [components.morphologizer.model.tok2vec]
@@ -246,7 +306,7 @@ width = ${components.tok2vec.model.encode.width}
 factory = "tagger"
 
 [components.tagger.model]
-@architectures = "spacy.Tagger.v1"
+@architectures = "spacy.Tagger.v2"
 nO = null
 
 [components.tagger.model.tok2vec]
@@ -290,6 +350,54 @@ nO = null
 width = ${components.tok2vec.model.encode.width}
 {% endif %}
 
+{% if "spancat" in components %}
+[components.spancat]
+factory = "spancat"
+max_positive = null
+scorer = {"@scorers":"spacy.spancat_scorer.v1"}
+spans_key = "sc"
+threshold = 0.5
+
+[components.spancat.model]
+@architectures = "spacy.SpanCategorizer.v1"
+
+[components.spancat.model.reducer]
+@layers = "spacy.mean_max_reducer.v1"
+hidden_size = 128
+
+[components.spancat.model.scorer]
+@layers = "spacy.LinearLogistic.v1"
+nO = null
+nI = null
+
+[components.spancat.model.tok2vec]
+@architectures = "spacy.Tok2VecListener.v1"
+width = ${components.tok2vec.model.encode.width}
+
+[components.spancat.suggester]
+@misc = "spacy.ngram_suggester.v1"
+sizes = [1,2,3]
+{% endif %}
+
+{% if "trainable_lemmatizer" in components -%}
+[components.trainable_lemmatizer]
+factory = "trainable_lemmatizer"
+backoff = "orth"
+min_tree_freq = 3
+overwrite = false
+scorer = {"@scorers":"spacy.lemmatizer_scorer.v1"}
+top_k = 1
+
+[components.trainable_lemmatizer.model]
+@architectures = "spacy.Tagger.v2"
+nO = null
+normalize = false
+
+[components.trainable_lemmatizer.model.tok2vec]
+@architectures = "spacy.Tok2VecListener.v1"
+width = ${components.tok2vec.model.encode.width}
+{% endif -%}
+
 {% if "entity_linker" in components -%}
 [components.entity_linker]
 factory = "entity_linker"
@@ -298,7 +406,7 @@ incl_context = true
 incl_prior = true
 
 [components.entity_linker.model]
-@architectures = "spacy.EntityLinker.v1"
+@architectures = "spacy.EntityLinker.v2"
 nO = null
 
 [components.entity_linker.model.tok2vec]
@@ -364,7 +472,7 @@ no_output_layer = false
 {% endif %}
 
 {% for pipe in components %}
-{% if pipe not in ["tagger", "morphologizer", "parser", "ner", "textcat", "textcat_multilabel", "entity_linker"] %}
+{% if pipe not in listener_components %}
 {# Other components defined by the user: we just assume they're factories #}
 [components.{{ pipe }}]
 factory = "{{ pipe }}"
@@ -421,8 +529,4 @@ compound = 1.001
 {% endif %}
 
 [initialize]
-{% if use_transformer or optimize == "efficiency" or not word_vectors -%}
 vectors = ${paths.vectors}
-{% else -%}
-vectors = "{{ word_vectors }}"
-{% endif -%}

spacy/default_config.cfg

@@ -68,12 +68,14 @@ seed = ${system.seed}
 gpu_allocator = ${system.gpu_allocator}
 dropout = 0.1
 accumulate_gradient = 1
-# Controls early-stopping. 0 disables early stopping.
+# Controls early-stopping, i.e., the number of steps to continue without
+# improvement before stopping. 0 disables early stopping.
 patience = 1600
 # Number of epochs. 0 means unlimited. If >= 0, train corpus is loaded once in
 # memory and shuffled within the training loop. -1 means stream train corpus
 # rather than loading in memory with no shuffling within the training loop.
 max_epochs = 0
+# Maximum number of update steps to train for. 0 means an unlimited number of steps.
 max_steps = 20000
 eval_frequency = 200
 # Control how scores are printed and checkpoints are evaluated.

spacy/displacy/__init__.py

@@ -7,7 +7,7 @@ USAGE: https://spacy.io/usage/visualizers
 from typing import Union, Iterable, Optional, Dict, Any, Callable
 import warnings
 
-from .render import DependencyRenderer, EntityRenderer
+from .render import DependencyRenderer, EntityRenderer, SpanRenderer
 from ..tokens import Doc, Span
 from ..errors import Errors, Warnings
 from ..util import is_in_jupyter
@@ -44,6 +44,7 @@ def render(
     factories = {
         "dep": (DependencyRenderer, parse_deps),
         "ent": (EntityRenderer, parse_ents),
+        "span": (SpanRenderer, parse_spans),
     }
     if style not in factories:
         raise ValueError(Errors.E087.format(style=style))
@@ -55,6 +56,10 @@ def render(
     renderer_func, converter = factories[style]
     renderer = renderer_func(options=options)
     parsed = [converter(doc, options) for doc in docs] if not manual else docs  # type: ignore
+    if manual:
+        for doc in docs:
+            if isinstance(doc, dict) and "ents" in doc:
+                doc["ents"] = sorted(doc["ents"], key=lambda x: (x["start"], x["end"]))
     _html["parsed"] = renderer.render(parsed, page=page, minify=minify).strip()  # type: ignore
     html = _html["parsed"]
     if RENDER_WRAPPER is not None:
@@ -203,6 +208,42 @@ def parse_ents(doc: Doc, options: Dict[str, Any] = {}) -> Dict[str, Any]:
     return {"text": doc.text, "ents": ents, "title": title, "settings": settings}
 
 
+def parse_spans(doc: Doc, options: Dict[str, Any] = {}) -> Dict[str, Any]:
+    """Generate spans in [{start: i, end: i, label: 'label'}] format.
+
+    doc (Doc): Document to parse.
+    options (Dict[str, any]): Span-specific visualisation options.
+    RETURNS (dict): Generated span types keyed by text (original text) and spans.
+    """
+    kb_url_template = options.get("kb_url_template", None)
+    spans_key = options.get("spans_key", "sc")
+    spans = [
+        {
+            "start": span.start_char,
+            "end": span.end_char,
+            "start_token": span.start,
+            "end_token": span.end,
+            "label": span.label_,
+            "kb_id": span.kb_id_ if span.kb_id_ else "",
+            "kb_url": kb_url_template.format(span.kb_id_) if kb_url_template else "#",
+        }
+        for span in doc.spans[spans_key]
+    ]
+    tokens = [token.text for token in doc]
+
+    if not spans:
+        warnings.warn(Warnings.W117.format(spans_key=spans_key))
+    title = doc.user_data.get("title", None) if hasattr(doc, "user_data") else None
+    settings = get_doc_settings(doc)
+    return {
+        "text": doc.text,
+        "spans": spans,
+        "title": title,
+        "settings": settings,
+        "tokens": tokens,
+    }
+
+
 def set_render_wrapper(func: Callable[[str], str]) -> None:
     """Set an optional wrapper function that is called around the generated
     HTML markup on displacy.render. This can be used to allow integration into

spacy/displacy/render.py

@@ -1,12 +1,15 @@
-from typing import Dict, Any, List, Optional, Union
+from typing import Any, Dict, List, Optional, Tuple, Union
 import uuid
+import itertools
 
-from .templates import TPL_DEP_SVG, TPL_DEP_WORDS, TPL_DEP_WORDS_LEMMA, TPL_DEP_ARCS
-from .templates import TPL_ENT, TPL_ENT_RTL, TPL_FIGURE, TPL_TITLE, TPL_PAGE
-from .templates import TPL_ENTS, TPL_KB_LINK
-from ..util import minify_html, escape_html, registry
 from ..errors import Errors
-
+from ..util import escape_html, minify_html, registry
+from .templates import TPL_DEP_ARCS, TPL_DEP_SVG, TPL_DEP_WORDS
+from .templates import TPL_DEP_WORDS_LEMMA, TPL_ENT, TPL_ENT_RTL, TPL_ENTS
+from .templates import TPL_FIGURE, TPL_KB_LINK, TPL_PAGE, TPL_SPAN
+from .templates import TPL_SPAN_RTL, TPL_SPAN_SLICE, TPL_SPAN_SLICE_RTL
+from .templates import TPL_SPAN_START, TPL_SPAN_START_RTL, TPL_SPANS
+from .templates import TPL_TITLE
 
 DEFAULT_LANG = "en"
 DEFAULT_DIR = "ltr"
@@ -18,7 +21,7 @@ DEFAULT_LABEL_COLORS = {
     "LOC": "#ff9561",
     "PERSON": "#aa9cfc",
     "NORP": "#c887fb",
-    "FACILITY": "#9cc9cc",
+    "FAC": "#9cc9cc",
     "EVENT": "#ffeb80",
     "LAW": "#ff8197",
     "LANGUAGE": "#ff8197",
@@ -33,6 +36,224 @@ DEFAULT_LABEL_COLORS = {
 }
 
 
+class SpanRenderer:
+    """Render Spans as SVGs."""
+
+    style = "span"
+
+    def __init__(self, options: Dict[str, Any] = {}) -> None:
+        """Initialise span renderer
+
+        options (dict): Visualiser-specific options (colors, spans)
+        """
+        # Set up the colors and overall look
+        colors = dict(DEFAULT_LABEL_COLORS)
+        user_colors = registry.displacy_colors.get_all()
+        for user_color in user_colors.values():
+            if callable(user_color):
+                # Since this comes from the function registry, we want to make
+                # sure we support functions that *return* a dict of colors
+                user_color = user_color()
+            if not isinstance(user_color, dict):
+                raise ValueError(Errors.E925.format(obj=type(user_color)))
+            colors.update(user_color)
+        colors.update(options.get("colors", {}))
+        self.default_color = DEFAULT_ENTITY_COLOR
+        self.colors = {label.upper(): color for label, color in colors.items()}
+
+        # Set up how the text and labels will be rendered
+        self.direction = DEFAULT_DIR
+        self.lang = DEFAULT_LANG
+        # These values are in px
+        self.top_offset = options.get("top_offset", 40)
+        # This is how far under the top offset the span labels appear
+        self.span_label_offset = options.get("span_label_offset", 20)
+        self.offset_step = options.get("top_offset_step", 17)
+
+        # Set up which templates will be used
+        template = options.get("template")
+        if template:
+            self.span_template = template["span"]
+            self.span_slice_template = template["slice"]
+            self.span_start_template = template["start"]
+        else:
+            if self.direction == "rtl":
+                self.span_template = TPL_SPAN_RTL
+                self.span_slice_template = TPL_SPAN_SLICE_RTL
+                self.span_start_template = TPL_SPAN_START_RTL
+            else:
+                self.span_template = TPL_SPAN
+                self.span_slice_template = TPL_SPAN_SLICE
+                self.span_start_template = TPL_SPAN_START
+
+    def render(
+        self, parsed: List[Dict[str, Any]], page: bool = False, minify: bool = False
+    ) -> str:
+        """Render complete markup.
+
+        parsed (list): Dependency parses to render.
+        page (bool): Render parses wrapped as full HTML page.
+        minify (bool): Minify HTML markup.
+        RETURNS (str): Rendered HTML markup.
+        """
+        rendered = []
+        for i, p in enumerate(parsed):
+            if i == 0:
+                settings = p.get("settings", {})
+                self.direction = settings.get("direction", DEFAULT_DIR)
+                self.lang = settings.get("lang", DEFAULT_LANG)
+            rendered.append(self.render_spans(p["tokens"], p["spans"], p.get("title")))
+
+        if page:
+            docs = "".join([TPL_FIGURE.format(content=doc) for doc in rendered])
+            markup = TPL_PAGE.format(content=docs, lang=self.lang, dir=self.direction)
+        else:
+            markup = "".join(rendered)
+        if minify:
+            return minify_html(markup)
+        return markup
+
+    def render_spans(
+        self,
+        tokens: List[str],
+        spans: List[Dict[str, Any]],
+        title: Optional[str],
+    ) -> str:
+        """Render span types in text.
+
+        Spans are rendered per-token, this means that for each token, we check if it's part
+        of a span slice (a member of a span type) or a span start (the starting token of a
+        given span type).
+
+        tokens (list): Individual tokens in the text
+        spans (list): Individual entity spans and their start, end, label, kb_id and kb_url.
+        title (str / None): Document title set in Doc.user_data['title'].
+        """
+        per_token_info = []
+        # we must sort so that we can correctly describe when spans need to "stack"
+        # which is determined by their start token, then span length (longer spans on top),
+        # then break any remaining ties with the span label
+        spans = sorted(
+            spans,
+            key=lambda s: (
+                s["start_token"],
+                -(s["end_token"] - s["start_token"]),
+                s["label"],
+            ),
+        )
+        for s in spans:
+            # this is the vertical 'slot' that the span will be rendered in
+            # vertical_position = span_label_offset + (offset_step * (slot - 1))
+            s["render_slot"] = 0
+        for idx, token in enumerate(tokens):
+            # Identify if a token belongs to a Span (and which) and if it's a
+            # start token of said Span. We'll use this for the final HTML render
+            token_markup: Dict[str, Any] = {}
+            token_markup["text"] = token
+            concurrent_spans = 0
+            entities = []
+            for span in spans:
+                ent = {}
+                if span["start_token"] <= idx < span["end_token"]:
+                    concurrent_spans += 1
+                    span_start = idx == span["start_token"]
+                    ent["label"] = span["label"]
+                    ent["is_start"] = span_start
+                    if span_start:
+                        # When the span starts, we need to know how many other
+                        # spans are on the 'span stack' and will be rendered.
+                        # This value becomes the vertical render slot for this entire span
+                        span["render_slot"] = concurrent_spans
+                    ent["render_slot"] = span["render_slot"]
+                    kb_id = span.get("kb_id", "")
+                    kb_url = span.get("kb_url", "#")
+                    ent["kb_link"] = (
+                        TPL_KB_LINK.format(kb_id=kb_id, kb_url=kb_url) if kb_id else ""
+                    )
+                    entities.append(ent)
+                else:
+                    # We don't specifically need to do this since we loop
+                    # over tokens and spans sorted by their start_token,
+                    # so we'll never use a span again after the last token it appears in,
+                    # but if we were to use these spans again we'd want to make sure
+                    # this value was reset correctly.
+                    span["render_slot"] = 0
+            token_markup["entities"] = entities
+            per_token_info.append(token_markup)
+        markup = self._render_markup(per_token_info)
+        markup = TPL_SPANS.format(content=markup, dir=self.direction)
+        if title:
+            markup = TPL_TITLE.format(title=title) + markup
+        return markup
+
+    def _render_markup(self, per_token_info: List[Dict[str, Any]]) -> str:
+        """Render the markup from per-token information"""
+        markup = ""
+        for token in per_token_info:
+            entities = sorted(token["entities"], key=lambda d: d["render_slot"])
+            # Whitespace tokens disrupt the vertical space (no line height) so that the
+            # span indicators get misaligned. We don't render them as individual
+            # tokens anyway, so we'll just not display a span indicator either.
+            is_whitespace = token["text"].strip() == ""
+            if entities and not is_whitespace:
+                slices = self._get_span_slices(token["entities"])
+                starts = self._get_span_starts(token["entities"])
+                total_height = (
+                    self.top_offset
+                    + self.span_label_offset
+                    + (self.offset_step * (len(entities) - 1))
+                )
+                markup += self.span_template.format(
+                    text=token["text"],
+                    span_slices=slices,
+                    span_starts=starts,
+                    total_height=total_height,
+                )
+            else:
+                markup += escape_html(token["text"] + " ")
+        return markup
+
+    def _get_span_slices(self, entities: List[Dict]) -> str:
+        """Get the rendered markup of all Span slices"""
+        span_slices = []
+        for entity in entities:
+            # rather than iterate over multiples of offset_step, we use entity['render_slot']
+            # to determine the vertical position, since that tells where
+            # the span starts vertically so we can extend it horizontally,
+            # past other spans that might have already ended
+            color = self.colors.get(entity["label"].upper(), self.default_color)
+            top_offset = self.top_offset + (
+                self.offset_step * (entity["render_slot"] - 1)
+            )
+            span_slice = self.span_slice_template.format(
+                bg=color,
+                top_offset=top_offset,
+            )
+            span_slices.append(span_slice)
+        return "".join(span_slices)
+
+    def _get_span_starts(self, entities: List[Dict]) -> str:
+        """Get the rendered markup of all Span start tokens"""
+        span_starts = []
+        for entity in entities:
+            color = self.colors.get(entity["label"].upper(), self.default_color)
+            top_offset = self.top_offset + (
+                self.offset_step * (entity["render_slot"] - 1)
+            )
+            span_start = (
+                self.span_start_template.format(
+                    bg=color,
+                    top_offset=top_offset,
+                    label=entity["label"],
+                    kb_link=entity["kb_link"],
+                )
+                if entity["is_start"]
+                else ""
+            )
+            span_starts.append(span_start)
+        return "".join(span_starts)
+
+
 class DependencyRenderer:
     """Render dependency parses as SVGs."""
 
@@ -105,7 +326,7 @@ class DependencyRenderer:
         RETURNS (str): Rendered SVG markup.
         """
         self.levels = self.get_levels(arcs)
-        self.highest_level = len(self.levels)
+        self.highest_level = max(self.levels.values(), default=0)
         self.offset_y = self.distance / 2 * self.highest_level + self.arrow_stroke
         self.width = self.offset_x + len(words) * self.distance
         self.height = self.offset_y + 3 * self.word_spacing
@@ -165,7 +386,7 @@ class DependencyRenderer:
         if start < 0 or end < 0:
             error_args = dict(start=start, end=end, label=label, dir=direction)
             raise ValueError(Errors.E157.format(**error_args))
-        level = self.levels.index(end - start) + 1
+        level = self.levels[(start, end, label)]
         x_start = self.offset_x + start * self.distance + self.arrow_spacing
         if self.direction == "rtl":
             x_start = self.width - x_start
@@ -181,7 +402,7 @@ class DependencyRenderer:
         y_curve = self.offset_y - level * self.distance / 2
         if self.compact:
             y_curve = self.offset_y - level * self.distance / 6
-        if y_curve == 0 and len(self.levels) > 5:
+        if y_curve == 0 and max(self.levels.values(), default=0) > 5:
             y_curve = -self.distance
         arrowhead = self.get_arrowhead(direction, x_start, y, x_end)
         arc = self.get_arc(x_start, y, y_curve, x_end)
@@ -225,15 +446,23 @@ class DependencyRenderer:
             p1, p2, p3 = (end, end + self.arrow_width - 2, end - self.arrow_width + 2)
         return f"M{p1},{y + 2} L{p2},{y - self.arrow_width} {p3},{y - self.arrow_width}"
 
-    def get_levels(self, arcs: List[Dict[str, Any]]) -> List[int]:
+    def get_levels(self, arcs: List[Dict[str, Any]]) -> Dict[Tuple[int, int, str], int]:
         """Calculate available arc height "levels".
         Used to calculate arrow heights dynamically and without wasting space.
 
         args (list): Individual arcs and their start, end, direction and label.
-        RETURNS (list): Arc levels sorted from lowest to highest.
+        RETURNS (dict): Arc levels keyed by (start, end, label).
         """
-        levels = set(map(lambda arc: arc["end"] - arc["start"], arcs))
-        return sorted(list(levels))
+        arcs = [dict(t) for t in {tuple(sorted(arc.items())) for arc in arcs}]
+        length = max([arc["end"] for arc in arcs], default=0)
+        max_level = [0] * length
+        levels = {}
+        for arc in sorted(arcs, key=lambda arc: arc["end"] - arc["start"]):
+            level = max(max_level[arc["start"] : arc["end"]]) + 1
+            for i in range(arc["start"], arc["end"]):
+                max_level[i] = level
+            levels[(arc["start"], arc["end"], arc["label"])] = level
+        return levels
 
 
 class EntityRenderer:
@@ -242,7 +471,7 @@ class EntityRenderer:
     style = "ent"
 
     def __init__(self, options: Dict[str, Any] = {}) -> None:
-        """Initialise dependency renderer.
+        """Initialise entity renderer.
 
         options (dict): Visualiser-specific options (colors, ents)
         """

spacy/displacy/templates.py

@@ -62,6 +62,55 @@ TPL_ENT_RTL = """
 </mark>
 """
 
+TPL_SPANS = """
+<div class="spans" style="line-height: 2.5; direction: {dir}">{content}</div>
+"""
+
+TPL_SPAN = """
+<span style="font-weight: bold; display: inline-block; position: relative; height: {total_height}px;">
+    {text}
+    {span_slices}
+    {span_starts}
+</span>
+"""
+
+TPL_SPAN_SLICE = """
+<span style="background: {bg}; top: {top_offset}px; height: 4px; left: -1px; width: calc(100% + 2px); position: absolute;">
+</span>
+"""
+
+
+TPL_SPAN_START = """
+<span style="background: {bg}; top: {top_offset}px; height: 4px; border-top-left-radius: 3px; border-bottom-left-radius: 3px; left: -1px; width: calc(100% + 2px); position: absolute;">
+    <span style="background: {bg}; z-index: 10; color: #000; top: -0.5em; padding: 2px 3px; position: absolute; font-size: 0.6em; font-weight: bold; line-height: 1; border-radius: 3px">
+        {label}{kb_link}
+    </span>
+</span>
+
+"""
+
+TPL_SPAN_RTL = """
+<span style="font-weight: bold; display: inline-block; position: relative;">
+    {text}
+    {span_slices}
+    {span_starts}
+</span>
+"""
+
+TPL_SPAN_SLICE_RTL = """
+<span style="background: {bg}; top: {top_offset}px; height: 4px; left: -1px; width: calc(100% + 2px); position: absolute;">
+</span>
+"""
+
+TPL_SPAN_START_RTL = """
+<span style="background: {bg}; top: {top_offset}px; height: 4px; border-top-left-radius: 3px; border-bottom-left-radius: 3px; left: -1px; width: calc(100% + 2px); position: absolute;">
+    <span style="background: {bg}; z-index: 10; color: #000; top: -0.5em; padding: 2px 3px; position: absolute; font-size: 0.6em; font-weight: bold; line-height: 1; border-radius: 3px">
+        {label}{kb_link}
+    </span>
+</span>
+"""
+
+
 # Important: this needs to start with a space!
 TPL_KB_LINK = """
  <a style="text-decoration: none; color: inherit; font-weight: normal" href="{kb_url}">{kb_id}</a>

spacy/errors.py

@@ -1,4 +1,5 @@
 import warnings
+from .compat import Literal
 
 
 class ErrorsWithCodes(type):
@@ -26,7 +27,10 @@ def setup_default_warnings():
     filter_warning("once", error_msg="[W114]")
 
 
-def filter_warning(action: str, error_msg: str):
+def filter_warning(
+    action: Literal["default", "error", "ignore", "always", "module", "once"],
+    error_msg: str,
+):
     """Customize how spaCy should handle a certain warning.
 
     error_msg (str): e.g. "W006", or a full error message
@@ -192,6 +196,22 @@ class Warnings(metaclass=ErrorsWithCodes):
     W115 = ("Skipping {method}: the floret vector table cannot be modified. "
             "Vectors are calculated from character ngrams.")
     W116 = ("Unable to clean attribute '{attr}'.")
+    W117 = ("No spans to visualize found in Doc object with spans_key: '{spans_key}'. If this is "
+            "surprising to you, make sure the Doc was processed using a model "
+            "that supports span categorization, and check the `doc.spans[spans_key]` "
+            "property manually if necessary.")
+    W118 = ("Term '{term}' not found in glossary. It may however be explained in documentation "
+            "for the corpora used to train the language. Please check "
+            "`nlp.meta[\"sources\"]` for any relevant links.")
+    W119 = ("Overriding pipe name in `config` is not supported. Ignoring override '{name_in_config}'.")
+    W120 = ("Unable to load all spans in Doc.spans: more than one span group "
+            "with the name '{group_name}' was found in the saved spans data. "
+            "Only the last span group will be loaded under "
+            "Doc.spans['{group_name}']. Skipping span group with values: "
+            "{group_values}")
+    W121 = ("Attempting to trace non-existent method '{method}' in pipe '{pipe}'")
+    W122 = ("Couldn't trace method '{method}' in pipe '{pipe}'. This can happen if the pipe class "
+            "is a Cython extension type.")
 
 
 class Errors(metaclass=ErrorsWithCodes):
@@ -437,10 +457,10 @@ class Errors(metaclass=ErrorsWithCodes):
             "same, but found '{nlp}' and '{vocab}' respectively.")
     E152 = ("The attribute {attr} is not supported for token patterns. "
             "Please use the option `validate=True` with the Matcher, PhraseMatcher, "
-            "or EntityRuler for more details.")
+            "EntityRuler or AttributeRuler for more details.")
     E153 = ("The value type {vtype} is not supported for token patterns. "
             "Please use the option validate=True with Matcher, PhraseMatcher, "
-            "or EntityRuler for more details.")
+            "EntityRuler or AttributeRuler for more details.")
     E154 = ("One of the attributes or values is not supported for token "
             "patterns. Please use the option `validate=True` with the Matcher, "
             "PhraseMatcher, or EntityRuler for more details.")
@@ -483,7 +503,7 @@ class Errors(metaclass=ErrorsWithCodes):
             "components, since spans are only views of the Doc. Use Doc and "
             "Token attributes (or custom extension attributes) only and remove "
             "the following: {attrs}")
-    E181 = ("Received invalid attributes for unkown object {obj}: {attrs}. "
+    E181 = ("Received invalid attributes for unknown object {obj}: {attrs}. "
             "Only Doc and Token attributes are supported.")
     E182 = ("Received invalid attribute declaration: {attr}\nDid you forget "
             "to define the attribute? For example: `{attr}.???`")
@@ -520,10 +540,16 @@ class Errors(metaclass=ErrorsWithCodes):
     E202 = ("Unsupported {name} mode '{mode}'. Supported modes: {modes}.")
 
     # New errors added in v3.x
+    E854 = ("Unable to set doc.ents. Check that the 'ents_filter' does not "
+            "permit overlapping spans.")
+    E855 = ("Invalid {obj}: {obj} is not from the same doc.")
+    E856 = ("Error accessing span at position {i}: out of bounds in span group "
+            "of length {length}.")
+    E857 = ("Entry '{name}' not found in edit tree lemmatizer labels.")
     E858 = ("The {mode} vector table does not support this operation. "
             "{alternative}")
     E859 = ("The floret vector table cannot be modified.")
-    E860 = ("Can't truncate fasttext-bloom vectors.")
+    E860 = ("Can't truncate floret vectors.")
     E861 = ("No 'keys' should be provided when initializing floret vectors "
             "with 'minn' and 'maxn'.")
     E862 = ("'hash_count' must be between 1-4 for floret vectors.")
@@ -566,9 +592,6 @@ class Errors(metaclass=ErrorsWithCodes):
     E879 = ("Unexpected type for 'spans' data. Provide a dictionary mapping keys to "
             "a list of spans, with each span represented by a tuple (start_char, end_char). "
             "The tuple can be optionally extended with a label and a KB ID.")
-    E880 = ("The 'wandb' library could not be found - did you install it? "
-            "Alternatively, specify the 'ConsoleLogger' in the 'training.logger' "
-            "config section, instead of the 'WandbLogger'.")
     E884 = ("The pipeline could not be initialized because the vectors "
             "could not be found at '{vectors}'. If your pipeline was already "
             "initialized/trained before, call 'resume_training' instead of 'initialize', "
@@ -888,9 +911,35 @@ class Errors(metaclass=ErrorsWithCodes):
     E1021 = ("`pos` value \"{pp}\" is not a valid Universal Dependencies tag. "
              "Non-UD tags should use the `tag` property.")
     E1022 = ("Words must be of type str or int, but input is of type '{wtype}'")
-    E1023 = ("Couldn't read EntityRuler from the {path}. This file doesn't exist.")
-    E1024 = ("A pattern with ID \"{ent_id}\" is not present in EntityRuler patterns.")
-    
+    E1023 = ("Couldn't read EntityRuler from the {path}. This file doesn't "
+             "exist.")
+    E1024 = ("A pattern with {attr_type} '{label}' is not present in "
+             "'{component}' patterns.")
+    E1025 = ("Cannot intify the value '{value}' as an IOB string. The only "
+             "supported values are: 'I', 'O', 'B' and ''")
+    E1026 = ("Edit tree has an invalid format:\n{errors}")
+    E1027 = ("AlignmentArray only supports slicing with a step of 1.")
+    E1028 = ("AlignmentArray only supports indexing using an int or a slice.")
+    E1029 = ("Edit tree cannot be applied to form.")
+    E1030 = ("Edit tree identifier out of range.")
+    E1031 = ("Could not find gold transition - see logs above.")
+    E1032 = ("`{var}` should not be {forbidden}, but received {value}.")
+    E1033 = ("Dimension {name} invalid -- only nO, nF, nP")
+    E1034 = ("Node index {i} out of bounds ({length})")
+    E1035 = ("Token index {i} out of bounds ({length})")
+    E1036 = ("Cannot index into NoneNode")
+    E1037 = ("Invalid attribute value '{attr}'.")
+    E1038 = ("Invalid JSON input: {message}")
+    E1039 = ("The {obj} start or end annotations (start: {start}, end: {end}) "
+             "could not be aligned to token boundaries.")
+    E1040 = ("Doc.from_json requires all tokens to have the same attributes. "
+             "Some tokens do not contain annotation for: {partial_attrs}")
+    E1041 = ("Expected a string, Doc, or bytes as input, but got: {type}")
+    E1042 = ("Function was called with `{arg1}`={arg1_values} and "
+             "`{arg2}`={arg2_values} but these arguments are conflicting.")
+    E1043 = ("Expected None or a value in range [{range_start}, {range_end}] for entity linker threshold, but got "
+             "{value}.")
+    E1044 = ("Misalignment in coref. Head token has no match in training doc.")
 
 
 # Deprecated model shortcuts, only used in errors and warnings

spacy/glossary.py

@@ -1,3 +1,7 @@
+import warnings
+from .errors import Warnings
+
+
 def explain(term):
     """Get a description for a given POS tag, dependency label or entity type.
 
@@ -11,6 +15,8 @@ def explain(term):
     """
     if term in GLOSSARY:
         return GLOSSARY[term]
+    else:
+        warnings.warn(Warnings.W118.format(term=term))
 
 
 GLOSSARY = {
@@ -267,6 +273,7 @@ GLOSSARY = {
     "relcl": "relative clause modifier",
     "reparandum": "overridden disfluency",
     "root": "root",
+    "ROOT": "root",
     "vocative": "vocative",
     "xcomp": "open clausal complement",
     # Dependency labels (German)
@@ -310,7 +317,6 @@ GLOSSARY = {
     "re": "repeated element",
     "rs": "reported speech",
     "sb": "subject",
-    "sb": "subject",
     "sbp": "passivized subject (PP)",
     "sp": "subject or predicate",
     "svp": "separable verb prefix",

spacy/kb.pyx

@@ -93,14 +93,14 @@ cdef class KnowledgeBase:
         self.vocab = vocab
         self._create_empty_vectors(dummy_hash=self.vocab.strings[""])
 
-    def initialize_entities(self, int64_t nr_entities):
+    def _initialize_entities(self, int64_t nr_entities):
         self._entry_index = PreshMap(nr_entities + 1)
         self._entries = entry_vec(nr_entities + 1)
 
-    def initialize_vectors(self, int64_t nr_entities):
+    def _initialize_vectors(self, int64_t nr_entities):
         self._vectors_table = float_matrix(nr_entities + 1)
 
-    def initialize_aliases(self, int64_t nr_aliases):
+    def _initialize_aliases(self, int64_t nr_aliases):
         self._alias_index = PreshMap(nr_aliases + 1)
         self._aliases_table = alias_vec(nr_aliases + 1)
 
@@ -155,8 +155,8 @@ cdef class KnowledgeBase:
             raise ValueError(Errors.E140)
 
         nr_entities = len(set(entity_list))
-        self.initialize_entities(nr_entities)
-        self.initialize_vectors(nr_entities)
+        self._initialize_entities(nr_entities)
+        self._initialize_vectors(nr_entities)
 
         i = 0
         cdef KBEntryC entry
@@ -388,9 +388,9 @@ cdef class KnowledgeBase:
             nr_entities = header[0]
             nr_aliases = header[1]
             entity_vector_length = header[2]
-            self.initialize_entities(nr_entities)
-            self.initialize_vectors(nr_entities)
-            self.initialize_aliases(nr_aliases)
+            self._initialize_entities(nr_entities)
+            self._initialize_vectors(nr_entities)
+            self._initialize_aliases(nr_aliases)
             self.entity_vector_length = entity_vector_length
 
         def deserialize_vectors(b):
@@ -512,8 +512,8 @@ cdef class KnowledgeBase:
         cdef int64_t entity_vector_length
         reader.read_header(&nr_entities, &entity_vector_length)
 
-        self.initialize_entities(nr_entities)
-        self.initialize_vectors(nr_entities)
+        self._initialize_entities(nr_entities)
+        self._initialize_vectors(nr_entities)
         self.entity_vector_length = entity_vector_length
 
         # STEP 1: load entity vectors
@@ -552,7 +552,7 @@ cdef class KnowledgeBase:
         # STEP 3: load aliases
         cdef int64_t nr_aliases
         reader.read_alias_length(&nr_aliases)
-        self.initialize_aliases(nr_aliases)
+        self._initialize_aliases(nr_aliases)
 
         cdef int64_t nr_candidates
         cdef vector[int64_t] entry_indices

spacy/lang/bg/__init__.py

@@ -2,7 +2,8 @@ from .stop_words import STOP_WORDS
 from .tokenizer_exceptions import TOKENIZER_EXCEPTIONS
 from .lex_attrs import LEX_ATTRS
 from ..tokenizer_exceptions import BASE_EXCEPTIONS
-
+from ..punctuation import COMBINING_DIACRITICS_TOKENIZER_INFIXES
+from ..punctuation import COMBINING_DIACRITICS_TOKENIZER_SUFFIXES
 from ...language import Language, BaseDefaults
 from ...attrs import LANG
 from ...util import update_exc
@@ -16,6 +17,8 @@ class BulgarianDefaults(BaseDefaults):
 
     stop_words = STOP_WORDS
     tokenizer_exceptions = update_exc(BASE_EXCEPTIONS, TOKENIZER_EXCEPTIONS)
+    suffixes = COMBINING_DIACRITICS_TOKENIZER_SUFFIXES
+    infixes = COMBINING_DIACRITICS_TOKENIZER_INFIXES
 
 
 class Bulgarian(Language):

spacy/lang/char_classes.py

@@ -45,6 +45,10 @@ _hangul_syllables = r"\uAC00-\uD7AF"
 _hangul_jamo = r"\u1100-\u11FF"
 _hangul = _hangul_syllables + _hangul_jamo
 
+_hiragana = r"\u3040-\u309F"
+_katakana = r"\u30A0-\u30FFー"
+_kana = _hiragana + _katakana
+
 # letters with diacritics - Catalan, Czech, Latin, Latvian, Lithuanian, Polish, Slovak, Turkish, Welsh
 _latin_u_extendedA = (
     r"\u0100\u0102\u0104\u0106\u0108\u010A\u010C\u010E\u0110\u0112\u0114\u0116\u0118\u011A\u011C"
@@ -244,6 +248,7 @@ _uncased = (
     + _tamil
     + _telugu
     + _hangul
+    + _kana
     + _cjk
 )
 
@@ -253,6 +258,10 @@ ALPHA = group_chars(
 ALPHA_LOWER = group_chars(_lower + _uncased)
 ALPHA_UPPER = group_chars(_upper + _uncased)
 
+_combining_diacritics = r"\u0300-\u036f"
+
+COMBINING_DIACRITICS = _combining_diacritics
+
 _units = (
     "km km² km³ m m² m³ dm dm² dm³ cm cm² cm³ mm mm² mm³ ha µm nm yd in ft "
     "kg g mg µg t lb oz m/s km/h kmh mph hPa Pa mbar mb MB kb KB gb GB tb "

spacy/lang/dsb/__init__.py

@@ -0,0 +1,16 @@
+from .lex_attrs import LEX_ATTRS
+from .stop_words import STOP_WORDS
+from ...language import Language, BaseDefaults
+
+
+class LowerSorbianDefaults(BaseDefaults):
+    lex_attr_getters = LEX_ATTRS
+    stop_words = STOP_WORDS
+
+
+class LowerSorbian(Language):
+    lang = "dsb"
+    Defaults = LowerSorbianDefaults
+
+
+__all__ = ["LowerSorbian"]

spacy/lang/dsb/examples.py

@@ -0,0 +1,15 @@
+"""
+Example sentences to test spaCy and its language models.
+
+>>> from spacy.lang.dsb.examples import sentences
+>>> docs = nlp.pipe(sentences)
+"""
+
+
+sentences = [
+    "Z tym stwori so wuměnjenje a zakład za dalše wobdźěłanje přez analyzu tekstoweje struktury a semantisku anotaciju a z tym tež za tu předstajenu digitalnu online-wersiju.",
+    "Mi so tu jara derje spodoba.",
+    "Kotre nowniny chceće měć?",
+    "Tak ako w slědnem lěśe jo teke lětosa jano doma zapustowaś móžno.",
+    "Zwóstanjo pótakem hyšći wjele źěła.",
+]

spacy/lang/dsb/lex_attrs.py

@@ -0,0 +1,113 @@
+from ...attrs import LIKE_NUM
+
+_num_words = [
+    "nul",
+    "jaden",
+    "jadna",
+    "jadno",
+    "dwa",
+    "dwě",
+    "tśi",
+    "tśo",
+    "styri",
+    "styrjo",
+    "pěś",
+    "pěśo",
+    "šesć",
+    "šesćo",
+    "sedym",
+    "sedymjo",
+    "wósym",
+    "wósymjo",
+    "źewjeś",
+    "źewjeśo",
+    "źaseś",
+    "źaseśo",
+    "jadnassćo",
+    "dwanassćo",
+    "tśinasćo",
+    "styrnasćo",
+    "pěśnasćo",
+    "šesnasćo",
+    "sedymnasćo",
+    "wósymnasćo",
+    "źewjeśnasćo",
+    "dwanasćo",
+    "dwaźasća",
+    "tśiźasća",
+    "styrźasća",
+    "pěśźaset",
+    "šesćźaset",
+    "sedymźaset",
+    "wósymźaset",
+    "źewjeśźaset",
+    "sto",
+    "tysac",
+    "milion",
+    "miliarda",
+    "bilion",
+    "biliarda",
+    "trilion",
+    "triliarda",
+]
+
+_ordinal_words = [
+    "prědny",
+    "prědna",
+    "prědne",
+    "drugi",
+    "druga",
+    "druge",
+    "tśeśi",
+    "tśeśa",
+    "tśeśe",
+    "stwórty",
+    "stwórta",
+    "stwórte",
+    "pêty",
+    "pěta",
+    "pête",
+    "šesty",
+    "šesta",
+    "šeste",
+    "sedymy",
+    "sedyma",
+    "sedyme",
+    "wósymy",
+    "wósyma",
+    "wósyme",
+    "źewjety",
+    "źewjeta",
+    "źewjete",
+    "źasety",
+    "źaseta",
+    "źasete",
+    "jadnasty",
+    "jadnasta",
+    "jadnaste",
+    "dwanasty",
+    "dwanasta",
+    "dwanaste",
+]
+
+
+def like_num(text):
+    if text.startswith(("+", "-", "±", "~")):
+        text = text[1:]
+    text = text.replace(",", "").replace(".", "")
+    if text.isdigit():
+        return True
+    if text.count("/") == 1:
+        num, denom = text.split("/")
+        if num.isdigit() and denom.isdigit():
+            return True
+    text_lower = text.lower()
+    if text_lower in _num_words:
+        return True
+    # Check ordinal number
+    if text_lower in _ordinal_words:
+        return True
+    return False
+
+
+LEX_ATTRS = {LIKE_NUM: like_num}

spacy/lang/dsb/stop_words.py

@@ -0,0 +1,15 @@
+STOP_WORDS = set(
+    """
+a abo aby ako ale až
+
+daniž dokulaž
+
+gaž
+
+jolic
+
+pak pótom
+
+teke togodla
+""".split()
+)

spacy/lang/en/tokenizer_exceptions.py

@@ -35,7 +35,7 @@ for pron in ["i"]:
 
         _exc[orth + "m"] = [
             {ORTH: orth, NORM: pron},
-            {ORTH: "m", "tenspect": 1, "number": 1},
+            {ORTH: "m"},
         ]
 
         _exc[orth + "'ma"] = [
@@ -139,26 +139,27 @@ for pron in ["he", "she", "it"]:
 
 # W-words, relative pronouns, prepositions etc.
 
-for word in [
-    "who",
-    "what",
-    "when",
-    "where",
-    "why",
-    "how",
-    "there",
-    "that",
-    "this",
-    "these",
-    "those",
+for word, morph in [
+    ("who", None),
+    ("what", None),
+    ("when", None),
+    ("where", None),
+    ("why", None),
+    ("how", None),
+    ("there", None),
+    ("that", "Number=Sing|Person=3"),
+    ("this", "Number=Sing|Person=3"),
+    ("these", "Number=Plur|Person=3"),
+    ("those", "Number=Plur|Person=3"),
 ]:
     for orth in [word, word.title()]:
-        _exc[orth + "'s"] = [
-            {ORTH: orth, NORM: word},
-            {ORTH: "'s", NORM: "'s"},
-        ]
+        if morph != "Number=Plur|Person=3":
+            _exc[orth + "'s"] = [
+                {ORTH: orth, NORM: word},
+                {ORTH: "'s", NORM: "'s"},
+            ]
 
-        _exc[orth + "s"] = [{ORTH: orth, NORM: word}, {ORTH: "s"}]
+            _exc[orth + "s"] = [{ORTH: orth, NORM: word}, {ORTH: "s"}]
 
         _exc[orth + "'ll"] = [
             {ORTH: orth, NORM: word},
@@ -182,25 +183,26 @@ for word in [
             {ORTH: "ve", NORM: "have"},
         ]
 
-        _exc[orth + "'re"] = [
-            {ORTH: orth, NORM: word},
-            {ORTH: "'re", NORM: "are"},
-        ]
+        if morph != "Number=Sing|Person=3":
+            _exc[orth + "'re"] = [
+                {ORTH: orth, NORM: word},
+                {ORTH: "'re", NORM: "are"},
+            ]
 
-        _exc[orth + "re"] = [
-            {ORTH: orth, NORM: word},
-            {ORTH: "re", NORM: "are"},
-        ]
+            _exc[orth + "re"] = [
+                {ORTH: orth, NORM: word},
+                {ORTH: "re", NORM: "are"},
+            ]
 
-        _exc[orth + "'ve"] = [
-            {ORTH: orth, NORM: word},
-            {ORTH: "'ve"},
-        ]
+            _exc[orth + "'ve"] = [
+                {ORTH: orth, NORM: word},
+                {ORTH: "'ve"},
+            ]
 
-        _exc[orth + "ve"] = [
-            {ORTH: orth},
-            {ORTH: "ve", NORM: "have"},
-        ]
+            _exc[orth + "ve"] = [
+                {ORTH: orth},
+                {ORTH: "ve", NORM: "have"},
+            ]
 
         _exc[orth + "'d"] = [
             {ORTH: orth, NORM: word},
@@ -447,7 +449,6 @@ for exc_data in [
     {ORTH: "La.", NORM: "Louisiana"},
     {ORTH: "Mar.", NORM: "March"},
     {ORTH: "Mass.", NORM: "Massachusetts"},
-    {ORTH: "May.", NORM: "May"},
     {ORTH: "Mich.", NORM: "Michigan"},
     {ORTH: "Minn.", NORM: "Minnesota"},
     {ORTH: "Miss.", NORM: "Mississippi"},

spacy/lang/es/examples.py

@@ -9,14 +9,14 @@ Example sentences to test spaCy and its language models.
 sentences = [
     "Apple está buscando comprar una startup del Reino Unido por mil millones de dólares.",
     "Los coches autónomos delegan la responsabilidad del seguro en sus fabricantes.",
-    "San Francisco analiza prohibir los robots delivery.",
+    "San Francisco analiza prohibir los robots de reparto.",
     "Londres es una gran ciudad del Reino Unido.",
     "El gato come pescado.",
     "Veo al hombre con el telescopio.",
     "La araña come moscas.",
     "El pingüino incuba en su nido sobre el hielo.",
-    "¿Dónde estais?",
-    "¿Quién es el presidente Francés?",
-    "¿Dónde está encuentra la capital de Argentina?",
+    "¿Dónde estáis?",
+    "¿Quién es el presidente francés?",
+    "¿Dónde se encuentra la capital de Argentina?",
     "¿Cuándo nació José de San Martín?",
 ]

spacy/lang/es/lex_attrs.py

@@ -47,6 +47,41 @@ _num_words = [
 ]
 
 
+_ordinal_words = [
+    "primero",
+    "segundo",
+    "tercero",
+    "cuarto",
+    "quinto",
+    "sexto",
+    "séptimo",
+    "octavo",
+    "noveno",
+    "décimo",
+    "undécimo",
+    "duodécimo",
+    "decimotercero",
+    "decimocuarto",
+    "decimoquinto",
+    "decimosexto",
+    "decimoséptimo",
+    "decimoctavo",
+    "decimonoveno",
+    "vigésimo",
+    "trigésimo",
+    "cuadragésimo",
+    "quincuagésimo",
+    "sexagésimo",
+    "septuagésimo",
+    "octogésima",
+    "nonagésima",
+    "centésima",
+    "milésima",
+    "millonésima",
+    "billonésima",
+]
+
+
 def like_num(text):
     if text.startswith(("+", "-", "±", "~")):
         text = text[1:]
@@ -57,7 +92,11 @@ def like_num(text):
         num, denom = text.split("/")
         if num.isdigit() and denom.isdigit():
             return True
-    if text.lower() in _num_words:
+    text_lower = text.lower()
+    if text_lower in _num_words:
+        return True
+    # Check ordinal number
+    if text_lower in _ordinal_words:
         return True
     return False
 

spacy/lang/es/stop_words.py

@@ -1,82 +1,80 @@
 STOP_WORDS = set(
     """
-actualmente acuerdo adelante ademas además adrede afirmó agregó ahi ahora ahí
-al algo alguna algunas alguno algunos algún alli allí alrededor ambos ampleamos
-antano antaño ante anterior antes apenas aproximadamente aquel aquella aquellas
-aquello aquellos aqui aquél aquélla aquéllas aquéllos aquí arriba arribaabajo
-aseguró asi así atras aun aunque ayer añadió aún
+a acuerdo adelante ademas además afirmó agregó ahi ahora ahí al algo alguna
+algunas alguno algunos algún alli allí alrededor ambos ante anterior antes
+apenas aproximadamente aquel aquella aquellas aquello aquellos aqui aquél
+aquélla aquéllas aquéllos aquí arriba aseguró asi así atras aun aunque añadió
+aún
 
 bajo bastante bien breve buen buena buenas bueno buenos
 
-cada casi cerca cierta ciertas cierto ciertos cinco claro comentó como con
-conmigo conocer conseguimos conseguir considera consideró consigo consigue
-consiguen consigues contigo contra cosas creo cual cuales cualquier cuando
-cuanta cuantas cuanto cuantos cuatro cuenta cuál cuáles cuándo cuánta cuántas
-cuánto cuántos cómo
+cada casi cierta ciertas cierto ciertos cinco claro comentó como con conmigo
+conocer conseguimos conseguir considera consideró consigo consigue consiguen
+consigues contigo contra creo cual cuales cualquier cuando cuanta cuantas
+cuanto cuantos cuatro cuenta cuál cuáles cuándo cuánta cuántas cuánto cuántos
+cómo
 
 da dado dan dar de debajo debe deben debido decir dejó del delante demasiado
 demás dentro deprisa desde despacio despues después detras detrás dia dias dice
-dicen dicho dieron diferente diferentes dijeron dijo dio donde dos durante día
-días dónde
+dicen dicho dieron diez diferente diferentes dijeron dijo dio doce donde dos
+durante día días dónde
 
-ejemplo el ella ellas ello ellos embargo empleais emplean emplear empleas
-empleo en encima encuentra enfrente enseguida entonces entre era eramos eran
-eras eres es esa esas ese eso esos esta estaba estaban estado estados estais
-estamos estan estar estará estas este esto estos estoy estuvo está están ex
-excepto existe existen explicó expresó él ésa ésas ése ésos ésta éstas éste
-éstos
+e el ella ellas ello ellos embargo en encima encuentra enfrente enseguida
+entonces entre era eramos eran eras eres es esa esas ese eso esos esta estaba
+estaban estado estados estais estamos estan estar estará estas este esto estos
+estoy estuvo está están excepto existe existen explicó expresó él ésa ésas ése
+ésos ésta éstas éste éstos
 
 fin final fue fuera fueron fui fuimos
 
-general gran grandes gueno
+gran grande grandes
 
 ha haber habia habla hablan habrá había habían hace haceis hacemos hacen hacer
 hacerlo haces hacia haciendo hago han hasta hay haya he hecho hemos hicieron
-hizo horas hoy hubo
+hizo hoy hubo
 
-igual incluso indicó informo informó intenta intentais intentamos intentan
-intentar intentas intento ir
+igual incluso indicó informo informó ir
 
 junto
 
-la lado largo las le lejos les llegó lleva llevar lo los luego lugar
+la lado largo las le les llegó lleva llevar lo los luego
 
 mal manera manifestó mas mayor me mediante medio mejor mencionó menos menudo mi
-mia mias mientras mio mios mis misma mismas mismo mismos modo momento mucha
-muchas mucho muchos muy más mí mía mías mío míos
+mia mias mientras mio mios mis misma mismas mismo mismos modo mucha muchas
+mucho muchos muy más mí mía mías mío míos
 
 nada nadie ni ninguna ningunas ninguno ningunos ningún no nos nosotras nosotros
-nuestra nuestras nuestro nuestros nueva nuevas nuevo nuevos nunca
+nuestra nuestras nuestro nuestros nueva nuevas nueve nuevo nuevos nunca
 
-ocho os otra otras otro otros
+o ocho once os otra otras otro otros
 
-pais para parece parte partir pasada pasado paìs peor pero pesar poca pocas
-poco pocos podeis podemos poder podria podriais podriamos podrian podrias podrá
+para parece parte partir pasada pasado paìs peor pero pesar poca pocas poco
+pocos podeis podemos poder podria podriais podriamos podrian podrias podrá
 podrán podría podrían poner por porque posible primer primera primero primeros
-principalmente pronto propia propias propio propios proximo próximo próximos
-pudo pueda puede pueden puedo pues
+pronto propia propias propio propios proximo próximo próximos pudo pueda puede
+pueden puedo pues
 
-qeu que quedó queremos quien quienes quiere quiza quizas quizá quizás quién quiénes qué
+qeu que quedó queremos quien quienes quiere quiza quizas quizá quizás quién
+quiénes qué
 
-raras realizado realizar realizó repente respecto
+realizado realizar realizó repente respecto
 
 sabe sabeis sabemos saben saber sabes salvo se sea sean segun segunda segundo
 según seis ser sera será serán sería señaló si sido siempre siendo siete sigue
-siguiente sin sino sobre sois sola solamente solas solo solos somos son soy
-soyos su supuesto sus suya suyas suyo sé sí sólo
+siguiente sin sino sobre sois sola solamente solas solo solos somos son soy su
+supuesto sus suya suyas suyo suyos sé sí sólo
 
 tal tambien también tampoco tan tanto tarde te temprano tendrá tendrán teneis
-tenemos tener tenga tengo tenido tenía tercera ti tiempo tiene tienen toda
-todas todavia todavía todo todos total trabaja trabajais trabajamos trabajan
-trabajar trabajas trabajo tras trata través tres tu tus tuvo tuya tuyas tuyo
-tuyos tú
+tenemos tener tenga tengo tenido tenía tercera tercero ti tiene tienen toda
+todas todavia todavía todo todos total tras trata través tres tu tus tuvo tuya
+tuyas tuyo tuyos tú
 
-ultimo un una unas uno unos usa usais usamos usan usar usas uso usted ustedes
+u ultimo un una unas uno unos usa usais usamos usan usar usas uso usted ustedes
 última últimas último últimos
 
-va vais valor vamos van varias varios vaya veces ver verdad verdadera verdadero
-vez vosotras vosotros voy vuestra vuestras vuestro vuestros
+va vais vamos van varias varios vaya veces ver verdad verdadera verdadero vez
+vosotras vosotros voy vuestra vuestras vuestro vuestros
 
-ya yo
+y ya yo
 """.split()
 )

spacy/lang/fi/__init__.py

@@ -2,6 +2,7 @@ from .tokenizer_exceptions import TOKENIZER_EXCEPTIONS
 from .stop_words import STOP_WORDS
 from .lex_attrs import LEX_ATTRS
 from .punctuation import TOKENIZER_INFIXES, TOKENIZER_SUFFIXES
+from .syntax_iterators import SYNTAX_ITERATORS
 from ...language import Language, BaseDefaults
 
 
@@ -11,6 +12,7 @@ class FinnishDefaults(BaseDefaults):
     tokenizer_exceptions = TOKENIZER_EXCEPTIONS
     lex_attr_getters = LEX_ATTRS
     stop_words = STOP_WORDS
+    syntax_iterators = SYNTAX_ITERATORS
 
 
 class Finnish(Language):

spacy/lang/fi/syntax_iterators.py

@@ -0,0 +1,79 @@
+from typing import Iterator, Tuple, Union
+from ...tokens import Doc, Span
+from ...symbols import NOUN, PROPN, PRON
+from ...errors import Errors
+
+
+def noun_chunks(doclike: Union[Doc, Span]) -> Iterator[Tuple[int, int, int]]:
+    """Detect base noun phrases from a dependency parse. Works on both Doc and Span."""
+    labels = [
+        "appos",
+        "nsubj",
+        "nsubj:cop",
+        "obj",
+        "obl",
+        "ROOT",
+    ]
+    extend_labels = [
+        "amod",
+        "compound",
+        "compound:nn",
+        "flat:name",
+        "nmod",
+        "nmod:gobj",
+        "nmod:gsubj",
+        "nmod:poss",
+        "nummod",
+    ]
+
+    def potential_np_head(word):
+        return word.pos in (NOUN, PROPN) and (
+            word.dep in np_deps or word.head.pos == PRON
+        )
+
+    doc = doclike.doc  # Ensure works on both Doc and Span.
+    if not doc.has_annotation("DEP"):
+        raise ValueError(Errors.E029)
+
+    np_deps = [doc.vocab.strings[label] for label in labels]
+    extend_deps = [doc.vocab.strings[label] for label in extend_labels]
+    np_label = doc.vocab.strings.add("NP")
+    conj_label = doc.vocab.strings.add("conj")
+
+    rbracket = 0
+    prev_end = -1
+    for i, word in enumerate(doclike):
+        if i < rbracket:
+            continue
+
+        # Is this a potential independent NP head or coordinated with
+        # a NOUN that is itself an independent NP head?
+        #
+        # e.g. "Terveyden ja hyvinvoinnin laitos"
+        if potential_np_head(word) or (
+            word.dep == conj_label and potential_np_head(word.head)
+        ):
+            # Try to extend to the left to include adjective/num
+            # modifiers, compound words etc.
+            lbracket = word.i
+            for ldep in word.lefts:
+                if ldep.dep in extend_deps:
+                    lbracket = ldep.left_edge.i
+                    break
+
+            # Prevent nested chunks from being produced
+            if lbracket <= prev_end:
+                continue
+
+            rbracket = word.i
+            # Try to extend the span to the right to capture
+            # appositions and noun modifiers
+            for rdep in word.rights:
+                if rdep.dep in extend_deps:
+                    rbracket = rdep.i
+            prev_end = rbracket
+
+            yield lbracket, rbracket + 1, np_label
+
+
+SYNTAX_ITERATORS = {"noun_chunks": noun_chunks}

spacy/lang/fr/lex_attrs.py

@@ -3,7 +3,7 @@ from ...attrs import LIKE_NUM
 
 _num_words = set(
     """
-zero un deux trois quatre cinq six sept huit neuf dix
+zero un une deux trois quatre cinq six sept huit neuf dix
 onze douze treize quatorze quinze seize dix-sept dix-huit dix-neuf
 vingt trente quarante cinquante soixante soixante-dix septante quatre-vingt huitante quatre-vingt-dix nonante
 cent mille mil million milliard billion quadrillion quintillion
@@ -13,7 +13,7 @@ sextillion septillion octillion nonillion decillion
 
 _ordinal_words = set(
     """
-premier deuxième second troisième quatrième cinquième sixième septième huitième neuvième dixième
+premier première deuxième second seconde troisième quatrième cinquième sixième septième huitième neuvième dixième
 onzième douzième treizième quatorzième quinzième seizième dix-septième dix-huitième dix-neuvième
 vingtième trentième quarantième cinquantième soixantième soixante-dixième septantième quatre-vingtième huitantième quatre-vingt-dixième nonantième
 centième millième millionnième milliardième billionnième quadrillionnième quintillionnième

spacy/lang/fr/syntax_iterators.py

@@ -6,16 +6,35 @@ from ...tokens import Doc, Span
 
 
 def noun_chunks(doclike: Union[Doc, Span]) -> Iterator[Tuple[int, int, int]]:
-    """Detect base noun phrases from a dependency parse. Works on Doc and Span."""
-    # fmt: off
-    labels = ["nsubj", "nsubj:pass", "obj", "iobj", "ROOT", "appos", "nmod", "nmod:poss"]
-    # fmt: on
+    """
+    Detect base noun phrases from a dependency parse. Works on both Doc and Span.
+    """
+    labels = [
+        "nsubj",
+        "nsubj:pass",
+        "obj",
+        "obl",
+        "obl:agent",
+        "obl:arg",
+        "obl:mod",
+        "nmod",
+        "pcomp",
+        "appos",
+        "ROOT",
+    ]
+    post_modifiers = ["flat", "flat:name", "flat:foreign", "fixed", "compound"]
     doc = doclike.doc  # Ensure works on both Doc and Span.
     if not doc.has_annotation("DEP"):
         raise ValueError(Errors.E029)
-    np_deps = [doc.vocab.strings[label] for label in labels]
-    conj = doc.vocab.strings.add("conj")
+    np_deps = {doc.vocab.strings.add(label) for label in labels}
+    np_modifs = {doc.vocab.strings.add(modifier) for modifier in post_modifiers}
     np_label = doc.vocab.strings.add("NP")
+    adj_label = doc.vocab.strings.add("amod")
+    det_label = doc.vocab.strings.add("det")
+    det_pos = doc.vocab.strings.add("DET")
+    adp_pos = doc.vocab.strings.add("ADP")
+    conj_label = doc.vocab.strings.add("conj")
+    conj_pos = doc.vocab.strings.add("CCONJ")
     prev_end = -1
     for i, word in enumerate(doclike):
         if word.pos not in (NOUN, PROPN, PRON):
@@ -24,16 +43,43 @@ def noun_chunks(doclike: Union[Doc, Span]) -> Iterator[Tuple[int, int, int]]:
         if word.left_edge.i <= prev_end:
             continue
         if word.dep in np_deps:
-            prev_end = word.right_edge.i
-            yield word.left_edge.i, word.right_edge.i + 1, np_label
-        elif word.dep == conj:
+            right_childs = list(word.rights)
+            right_child = right_childs[0] if right_childs else None
+
+            if right_child:
+                if (
+                    right_child.dep == adj_label
+                ):  # allow chain of adjectives by expanding to right
+                    right_end = right_child.right_edge
+                elif (
+                    right_child.dep == det_label and right_child.pos == det_pos
+                ):  # cut relative pronouns here
+                    right_end = right_child
+                elif right_child.dep in np_modifs:  # Check if we can expand to right
+                    right_end = word.right_edge
+                else:
+                    right_end = word
+            else:
+                right_end = word
+            prev_end = right_end.i
+
+            left_index = word.left_edge.i
+            left_index = left_index + 1 if word.left_edge.pos == adp_pos else left_index
+
+            yield left_index, right_end.i + 1, np_label
+        elif word.dep == conj_label:
             head = word.head
-            while head.dep == conj and head.head.i < head.i:
+            while head.dep == conj_label and head.head.i < head.i:
                 head = head.head
             # If the head is an NP, and we're coordinated to it, we're an NP
             if head.dep in np_deps:
-                prev_end = word.right_edge.i
-                yield word.left_edge.i, word.right_edge.i + 1, np_label
+                prev_end = word.i
+
+                left_index = word.left_edge.i  # eliminate left attached conjunction
+                left_index = (
+                    left_index + 1 if word.left_edge.pos == conj_pos else left_index
+                )
+                yield left_index, word.i + 1, np_label
 
 
 SYNTAX_ITERATORS = {"noun_chunks": noun_chunks}

spacy/lang/hi/lex_attrs.py

@@ -90,7 +90,7 @@ _eleven_to_beyond = [
     "अड़सठ",
     "उनहत्तर",
     "सत्तर",
-    "इकहत्तर"
+    "इकहत्तर",
     "बहत्तर",
     "तिहत्तर",
     "चौहत्तर",

spacy/lang/hsb/__init__.py

@@ -0,0 +1,18 @@
+from .lex_attrs import LEX_ATTRS
+from .stop_words import STOP_WORDS
+from .tokenizer_exceptions import TOKENIZER_EXCEPTIONS
+from ...language import Language, BaseDefaults
+
+
+class UpperSorbianDefaults(BaseDefaults):
+    lex_attr_getters = LEX_ATTRS
+    stop_words = STOP_WORDS
+    tokenizer_exceptions = TOKENIZER_EXCEPTIONS
+
+
+class UpperSorbian(Language):
+    lang = "hsb"
+    Defaults = UpperSorbianDefaults
+
+
+__all__ = ["UpperSorbian"]

spacy/lang/hsb/examples.py

@@ -0,0 +1,15 @@
+"""
+Example sentences to test spaCy and its language models.
+
+>>> from spacy.lang.hsb.examples import sentences
+>>> docs = nlp.pipe(sentences)
+"""
+
+
+sentences = [
+    "To běšo wjelgin raźone a jo se wót luźi derje pśiwzeło. Tak som dožywiła wjelgin",
+    "Jogo pśewóźowarce stej groniłej, až how w serbskich stronach njama Santa Claus nic pytaś.",
+    "A ten sobuźěłaśeŕ Statneje biblioteki w Barlinju jo pśimjeł drogotne knigły bźez rukajcowu z nagima rukoma!",
+    "Take wobchadanje z našym kulturnym derbstwom zewšym njejźo.",
+    "Wopśimjeśe drugich pśinoskow jo było na wusokem niwowje, ako pśecej.",
+]

spacy/lang/hsb/lex_attrs.py

@@ -0,0 +1,106 @@
+from ...attrs import LIKE_NUM
+
+_num_words = [
+    "nul",
+    "jedyn",
+    "jedna",
+    "jedne",
+    "dwaj",
+    "dwě",
+    "tři",
+    "třo",
+    "štyri",
+    "štyrjo",
+    "pjeć",
+    "šěsć",
+    "sydom",
+    "wosom",
+    "dźewjeć",
+    "dźesać",
+    "jědnaće",
+    "dwanaće",
+    "třinaće",
+    "štyrnaće",
+    "pjatnaće",
+    "šěsnaće",
+    "sydomnaće",
+    "wosomnaće",
+    "dźewjatnaće",
+    "dwaceći",
+    "třiceći",
+    "štyrceći",
+    "pjećdźesat",
+    "šěsćdźesat",
+    "sydomdźesat",
+    "wosomdźesat",
+    "dźewjećdźesat",
+    "sto",
+    "tysac",
+    "milion",
+    "miliarda",
+    "bilion",
+    "biliarda",
+    "trilion",
+    "triliarda",
+]
+
+_ordinal_words = [
+    "prěni",
+    "prěnja",
+    "prěnje",
+    "druhi",
+    "druha",
+    "druhe",
+    "třeći",
+    "třeća",
+    "třeće",
+    "štwórty",
+    "štwórta",
+    "štwórte",
+    "pjaty",
+    "pjata",
+    "pjate",
+    "šěsty",
+    "šěsta",
+    "šěste",
+    "sydmy",
+    "sydma",
+    "sydme",
+    "wosmy",
+    "wosma",
+    "wosme",
+    "dźewjaty",
+    "dźewjata",
+    "dźewjate",
+    "dźesaty",
+    "dźesata",
+    "dźesate",
+    "jědnaty",
+    "jědnata",
+    "jědnate",
+    "dwanaty",
+    "dwanata",
+    "dwanate",
+]
+
+
+def like_num(text):
+    if text.startswith(("+", "-", "±", "~")):
+        text = text[1:]
+    text = text.replace(",", "").replace(".", "")
+    if text.isdigit():
+        return True
+    if text.count("/") == 1:
+        num, denom = text.split("/")
+        if num.isdigit() and denom.isdigit():
+            return True
+    text_lower = text.lower()
+    if text_lower in _num_words:
+        return True
+    # Check ordinal number
+    if text_lower in _ordinal_words:
+        return True
+    return False
+
+
+LEX_ATTRS = {LIKE_NUM: like_num}

spacy/lang/hsb/stop_words.py

@@ -0,0 +1,19 @@
+STOP_WORDS = set(
+    """
+a abo ale ani
+
+dokelž
+
+hdyž
+
+jeli jelizo
+
+kaž
+
+pak potom
+
+tež tohodla
+
+zo zoby
+""".split()
+)

spacy/lang/hsb/tokenizer_exceptions.py

@@ -0,0 +1,18 @@
+from ..tokenizer_exceptions import BASE_EXCEPTIONS
+from ...symbols import ORTH, NORM
+from ...util import update_exc
+
+_exc = dict()
+for exc_data in [
+    {ORTH: "mil.", NORM: "milion"},
+    {ORTH: "wob.", NORM: "wobydler"},
+]:
+    _exc[exc_data[ORTH]] = [exc_data]
+
+for orth in [
+    "resp.",
+]:
+    _exc[orth] = [{ORTH: orth}]
+
+
+TOKENIZER_EXCEPTIONS = update_exc(BASE_EXCEPTIONS, _exc)

spacy/lang/it/__init__.py

@@ -6,13 +6,15 @@ from .tokenizer_exceptions import TOKENIZER_EXCEPTIONS
 from .punctuation import TOKENIZER_PREFIXES, TOKENIZER_INFIXES
 from ...language import Language, BaseDefaults
 from .lemmatizer import ItalianLemmatizer
+from .syntax_iterators import SYNTAX_ITERATORS
 
 
 class ItalianDefaults(BaseDefaults):
     tokenizer_exceptions = TOKENIZER_EXCEPTIONS
-    stop_words = STOP_WORDS
     prefixes = TOKENIZER_PREFIXES
     infixes = TOKENIZER_INFIXES
+    stop_words = STOP_WORDS
+    syntax_iterators = SYNTAX_ITERATORS
 
 
 class Italian(Language):

spacy/lang/it/stop_words.py

@@ -10,18 +10,18 @@ avresti avrete avrà avrò avuta avute avuti avuto
 
 basta bene benissimo brava bravo
 
-casa caso cento certa certe certi certo che chi chicchessia chiunque ci
+casa caso cento certa certe certi certo che chi chicchessia chiunque ci c'
 ciascuna ciascuno cima cio cioe circa citta città co codesta codesti codesto
 cogli coi col colei coll coloro colui come cominci comunque con concernente
 conciliarsi conclusione consiglio contro cortesia cos cosa cosi così cui
 
-da dagl dagli dai dal dall dalla dalle dallo dappertutto davanti degl degli
-dei del dell della delle dello dentro detto deve di dice dietro dire
+d' da dagl dagli dai dal dall dall' dalla dalle dallo dappertutto davanti degl degli
+dei del dell dell' della delle dello dentro detto deve di dice dietro dire
 dirimpetto diventa diventare diventato dopo dov dove dovra dovrà dovunque due
 dunque durante
 
-ebbe ebbero ebbi ecc ecco ed effettivamente egli ella entrambi eppure era
-erano eravamo eravate eri ero esempio esse essendo esser essere essi ex
+e ebbe ebbero ebbi ecc ecco ed effettivamente egli ella entrambi eppure era
+erano eravamo eravate eri ero esempio esse essendo esser essere essi ex è
 
 fa faccia facciamo facciano facciate faccio facemmo facendo facesse facessero
 facessi facessimo faceste facesti faceva facevamo facevano facevate facevi
@@ -30,21 +30,21 @@ fareste faresti farete farà farò fatto favore fece fecero feci fin finalmente
 finche fine fino forse forza fosse fossero fossi fossimo foste fosti fra
 frattempo fu fui fummo fuori furono futuro generale
 
-gia già giacche giorni giorno gli gliela gliele glieli glielo gliene governo
+gia già giacche giorni giorno gli gl' gliela gliele glieli glielo gliene governo
 grande grazie gruppo
 
 ha haha hai hanno ho
 
 ieri il improvviso in inc infatti inoltre insieme intanto intorno invece io
 
-la là lasciato lato lavoro le lei li lo lontano loro lui lungo luogo
+l' la là lasciato lato lavoro le lei li lo lontano loro lui lungo luogo
 
-ma macche magari maggior mai male malgrado malissimo mancanza marche me
+m' ma macche magari maggior mai male malgrado malissimo mancanza marche me
 medesimo mediante meglio meno mentre mesi mezzo mi mia mie miei mila miliardi
 milioni minimi ministro mio modo molti moltissimo molto momento mondo mosto
 
-nazionale ne negl negli nei nel nell nella nelle nello nemmeno neppure nessun
-nessuna nessuno niente no noi non nondimeno nonostante nonsia nostra nostre
+nazionale ne negl negli nei nel nell nella nelle nello nemmeno neppure nessun nessun'
+nessuna nessuno nient' niente no noi non nondimeno nonostante nonsia nostra nostre
 nostri nostro novanta nove nulla nuovo
 
 od oggi ogni ognuna ognuno oltre oppure ora ore osi ossia ottanta otto
@@ -56,12 +56,12 @@ potrebbe preferibilmente presa press prima primo principalmente probabilmente
 proprio puo può pure purtroppo
 
 qualche qualcosa qualcuna qualcuno quale quali qualunque quando quanta quante
-quanti quanto quantunque quasi quattro quel quella quelle quelli quello quest
+quanti quanto quantunque quasi quattro quel quel' quella quelle quelli quello quest quest'
 questa queste questi questo qui quindi
 
 realmente recente recentemente registrazione relativo riecco salvo
 
-sara sarà sarai saranno sarebbe sarebbero sarei saremmo saremo sareste
+s' sara sarà sarai saranno sarebbe sarebbero sarei saremmo saremo sareste
 saresti sarete saro sarò scola scopo scorso se secondo seguente seguito sei
 sembra sembrare sembrato sembri sempre senza sette si sia siamo siano siate
 siete sig solito solo soltanto sono sopra sotto spesso srl sta stai stando
@@ -72,12 +72,12 @@ steste stesti stette stettero stetti stia stiamo stiano stiate sto su sua
 subito successivamente successivo sue sugl sugli sui sul sull sulla sulle
 sullo suo suoi
 
-tale tali talvolta tanto te tempo ti titolo tra tranne tre trenta
+t' tale tali talvolta tanto te tempo ti titolo tra tranne tre trenta
 troppo trovato tu tua tue tuo tuoi tutta tuttavia tutte tutti tutto
 
-uguali ulteriore ultimo un una uno uomo
+uguali ulteriore ultimo un un' una uno uomo
 
-va vale vari varia varie vario verso vi via vicino visto vita voi volta volte
+v' va vale vari varia varie vario verso vi via vicino visto vita voi volta volte
 vostra vostre vostri vostro
 """.split()
 )

spacy/lang/it/syntax_iterators.py

@@ -0,0 +1,86 @@
+from typing import Union, Iterator, Tuple
+
+from ...symbols import NOUN, PROPN, PRON
+from ...errors import Errors
+from ...tokens import Doc, Span
+
+
+def noun_chunks(doclike: Union[Doc, Span]) -> Iterator[Tuple[int, int, int]]:
+    """
+    Detect base noun phrases from a dependency parse. Works on both Doc and Span.
+    """
+    labels = [
+        "nsubj",
+        "nsubj:pass",
+        "obj",
+        "obl",
+        "obl:agent",
+        "nmod",
+        "pcomp",
+        "appos",
+        "ROOT",
+    ]
+    post_modifiers = ["flat", "flat:name", "fixed", "compound"]
+    dets = ["det", "det:poss"]
+    doc = doclike.doc  # Ensure works on both Doc and Span.
+    if not doc.has_annotation("DEP"):
+        raise ValueError(Errors.E029)
+    np_deps = {doc.vocab.strings.add(label) for label in labels}
+    np_modifs = {doc.vocab.strings.add(modifier) for modifier in post_modifiers}
+    np_label = doc.vocab.strings.add("NP")
+    adj_label = doc.vocab.strings.add("amod")
+    det_labels = {doc.vocab.strings.add(det) for det in dets}
+    det_pos = doc.vocab.strings.add("DET")
+    adp_label = doc.vocab.strings.add("ADP")
+    conj = doc.vocab.strings.add("conj")
+    conj_pos = doc.vocab.strings.add("CCONJ")
+    prev_end = -1
+    for i, word in enumerate(doclike):
+        if word.pos not in (NOUN, PROPN, PRON):
+            continue
+        # Prevent nested chunks from being produced
+        if word.left_edge.i <= prev_end:
+            continue
+        if word.dep in np_deps:
+            right_childs = list(word.rights)
+            right_child = right_childs[0] if right_childs else None
+
+            if right_child:
+                if (
+                    right_child.dep == adj_label
+                ):  # allow chain of adjectives by expanding to right
+                    right_end = right_child.right_edge
+                elif (
+                    right_child.dep in det_labels and right_child.pos == det_pos
+                ):  # cut relative pronouns here
+                    right_end = right_child
+                elif right_child.dep in np_modifs:  # Check if we can expand to right
+                    right_end = word.right_edge
+                else:
+                    right_end = word
+            else:
+                right_end = word
+            prev_end = right_end.i
+
+            left_index = word.left_edge.i
+            left_index = (
+                left_index + 1 if word.left_edge.pos == adp_label else left_index
+            )
+
+            yield left_index, right_end.i + 1, np_label
+        elif word.dep == conj:
+            head = word.head
+            while head.dep == conj and head.head.i < head.i:
+                head = head.head
+            # If the head is an NP, and we're coordinated to it, we're an NP
+            if head.dep in np_deps:
+                prev_end = word.i
+
+                left_index = word.left_edge.i  # eliminate left attached conjunction
+                left_index = (
+                    left_index + 1 if word.left_edge.pos == conj_pos else left_index
+                )
+                yield left_index, word.i + 1, np_label
+
+
+SYNTAX_ITERATORS = {"noun_chunks": noun_chunks}

spacy/lang/ko/__init__.py

@@ -1,12 +1,13 @@
 from typing import Iterator, Any, Dict
 
+from .punctuation import TOKENIZER_INFIXES
 from .stop_words import STOP_WORDS
 from .tag_map import TAG_MAP
 from .lex_attrs import LEX_ATTRS
 from ...language import Language, BaseDefaults
 from ...tokens import Doc
 from ...scorer import Scorer
-from ...symbols import POS
+from ...symbols import POS, X
 from ...training import validate_examples
 from ...util import DummyTokenizer, registry, load_config_from_str
 from ...vocab import Vocab
@@ -31,15 +32,24 @@ def create_tokenizer():
 class KoreanTokenizer(DummyTokenizer):
     def __init__(self, vocab: Vocab):
         self.vocab = vocab
-        MeCab = try_mecab_import()  # type: ignore[func-returns-value]
-        self.mecab_tokenizer = MeCab("-F%f[0],%f[7]")
+        self._mecab = try_mecab_import()  # type: ignore[func-returns-value]
+        self._mecab_tokenizer = None
+
+    @property
+    def mecab_tokenizer(self):
+        # This is a property so that initializing a pipeline with blank:ko is
+        # possible without actually requiring mecab-ko, e.g. to run
+        # `spacy init vectors ko` for a pipeline that will have a different
+        # tokenizer in the end. The languages need to match for the vectors
+        # to be imported and there's no way to pass a custom config to
+        # `init vectors`.
+        if self._mecab_tokenizer is None:
+            self._mecab_tokenizer = self._mecab("-F%f[0],%f[7]")
+        return self._mecab_tokenizer
 
     def __reduce__(self):
         return KoreanTokenizer, (self.vocab,)
 
-    def __del__(self):
-        self.mecab_tokenizer.__del__()
-
     def __call__(self, text: str) -> Doc:
         dtokens = list(self.detailed_tokens(text))
         surfaces = [dt["surface"] for dt in dtokens]
@@ -47,7 +57,10 @@ class KoreanTokenizer(DummyTokenizer):
         for token, dtoken in zip(doc, dtokens):
             first_tag, sep, eomi_tags = dtoken["tag"].partition("+")
             token.tag_ = first_tag  # stem(어간) or pre-final(선어말 어미)
-            token.pos = TAG_MAP[token.tag_][POS]
+            if token.tag_ in TAG_MAP:
+                token.pos = TAG_MAP[token.tag_][POS]
+            else:
+                token.pos = X
             token.lemma_ = dtoken["lemma"]
         doc.user_data["full_tags"] = [dt["tag"] for dt in dtokens]
         return doc
@@ -76,6 +89,7 @@ class KoreanDefaults(BaseDefaults):
     lex_attr_getters = LEX_ATTRS
     stop_words = STOP_WORDS
     writing_system = {"direction": "ltr", "has_case": False, "has_letters": False}
+    infixes = TOKENIZER_INFIXES
 
 
 class Korean(Language):
@@ -90,7 +104,8 @@ def try_mecab_import() -> None:
         return MeCab
     except ImportError:
         raise ImportError(
-            "Korean support requires [mecab-ko](https://bitbucket.org/eunjeon/mecab-ko/src/master/README.md), "
+            'The Korean tokenizer ("spacy.ko.KoreanTokenizer") requires '
+            "[mecab-ko](https://bitbucket.org/eunjeon/mecab-ko/src/master/README.md), "
             "[mecab-ko-dic](https://bitbucket.org/eunjeon/mecab-ko-dic), "
             "and [natto-py](https://github.com/buruzaemon/natto-py)"
         ) from None

spacy/lang/ko/punctuation.py

@@ -0,0 +1,12 @@
+from ..char_classes import LIST_QUOTES
+from ..punctuation import TOKENIZER_INFIXES as BASE_TOKENIZER_INFIXES
+
+
+_infixes = (
+    ["·", "ㆍ", "\(", "\)"]
+    + [r"(?<=[0-9])~(?=[0-9-])"]
+    + LIST_QUOTES
+    + BASE_TOKENIZER_INFIXES
+)
+
+TOKENIZER_INFIXES = _infixes

spacy/lang/nb/stop_words.py

@@ -4,46 +4,42 @@ alle allerede alt and andre annen annet at av
 
 bak bare bedre beste blant ble bli blir blitt bris by både
 
-da dag de del dem den denne der dermed det dette disse drept du
+da dag de del dem den denne der dermed det dette disse du
 
 eller en enn er et ett etter
 
-fem fikk fire fjor flere folk for fortsatt fotball fra fram frankrike fredag
+fem fikk fire fjor flere folk for fortsatt fra fram
 funnet få får fått før først første
 
 gang gi gikk gjennom gjorde gjort gjør gjøre god godt grunn gå går
 
-ha hadde ham han hans har hele helt henne hennes her hun hva hvor hvordan
-hvorfor
+ha hadde ham han hans har hele helt henne hennes her hun
 
 i ifølge igjen ikke ingen inn
 
 ja jeg
 
 kamp kampen kan kl klart kom komme kommer kontakt kort kroner kunne kveld
-kvinner
 
-la laget land landet langt leder ligger like litt løpet lørdag
+la laget land landet langt leder ligger like litt løpet
 
-man mandag mange mannen mars med meg mellom men mener menn mennesker mens mer
-millioner minutter mot msci mye må mål måtte
+man mange med meg mellom men mener mennesker mens mer mot mye må mål måtte
 
-ned neste noe noen nok norge norsk norske ntb ny nye nå når
+ned neste noe noen nok ny nye nå når
 
-og også om onsdag opp opplyser oslo oss over
+og også om opp opplyser oss over
 
-personer plass poeng politidistrikt politiet president prosent på
+personer plass poeng på
 
-regjeringen runde rundt russland
+runde rundt
 
-sa saken samme sammen samtidig satt se seg seks selv senere september ser sett
+sa saken samme sammen samtidig satt se seg seks selv senere ser sett
 siden sier sin sine siste sitt skal skriver skulle slik som sted stedet stor
-store står sverige svært så søndag
+store står svært så
 
-ta tatt tid tidligere til tilbake tillegg tirsdag to tok torsdag tre tror
-tyskland
+ta tatt tid tidligere til tilbake tillegg tok tror
 
-under usa ut uten utenfor
+under ut uten utenfor
 
 vant var ved veldig vi videre viktig vil ville viser vår være vært
 

spacy/lang/punctuation.py

@@ -1,5 +1,5 @@
 from .char_classes import LIST_PUNCT, LIST_ELLIPSES, LIST_QUOTES, LIST_CURRENCY
-from .char_classes import LIST_ICONS, HYPHENS, CURRENCY, UNITS
+from .char_classes import LIST_ICONS, HYPHENS, CURRENCY, UNITS, COMBINING_DIACRITICS
 from .char_classes import CONCAT_QUOTES, ALPHA_LOWER, ALPHA_UPPER, ALPHA, PUNCT
 
 
@@ -44,3 +44,23 @@ TOKENIZER_INFIXES = (
         r"(?<=[{a}0-9])[:<>=/](?=[{a}])".format(a=ALPHA),
     ]
 )
+
+
+# Some languages e.g. written with the Cyrillic alphabet permit the use of diacritics
+# to mark stressed syllables in words where stress is distinctive. Such languages
+# should use the COMBINING_DIACRITICS... suffix and infix regex lists in
+# place of the standard ones.
+COMBINING_DIACRITICS_TOKENIZER_SUFFIXES = list(TOKENIZER_SUFFIXES) + [
+    r"(?<=[{a}][{d}])\.".format(a=ALPHA, d=COMBINING_DIACRITICS),
+]
+
+COMBINING_DIACRITICS_TOKENIZER_INFIXES = list(TOKENIZER_INFIXES) + [
+    r"(?<=[{al}][{d}])\.(?=[{au}{q}])".format(
+        al=ALPHA_LOWER, au=ALPHA_UPPER, q=CONCAT_QUOTES, d=COMBINING_DIACRITICS
+    ),
+    r"(?<=[{a}][{d}]),(?=[{a}])".format(a=ALPHA, d=COMBINING_DIACRITICS),
+    r"(?<=[{a}][{d}])(?:{h})(?=[{a}])".format(
+        a=ALPHA, d=COMBINING_DIACRITICS, h=HYPHENS
+    ),
+    r"(?<=[{a}][{d}])[:<>=/](?=[{a}])".format(a=ALPHA, d=COMBINING_DIACRITICS),
+]

spacy/lang/ru/__init__.py

@@ -5,6 +5,8 @@ from .stop_words import STOP_WORDS
 from .tokenizer_exceptions import TOKENIZER_EXCEPTIONS
 from .lex_attrs import LEX_ATTRS
 from .lemmatizer import RussianLemmatizer
+from ..punctuation import COMBINING_DIACRITICS_TOKENIZER_INFIXES
+from ..punctuation import COMBINING_DIACRITICS_TOKENIZER_SUFFIXES
 from ...language import Language, BaseDefaults
 
 
@@ -12,6 +14,8 @@ class RussianDefaults(BaseDefaults):
     tokenizer_exceptions = TOKENIZER_EXCEPTIONS
     lex_attr_getters = LEX_ATTRS
     stop_words = STOP_WORDS
+    suffixes = COMBINING_DIACRITICS_TOKENIZER_SUFFIXES
+    infixes = COMBINING_DIACRITICS_TOKENIZER_INFIXES
 
 
 class Russian(Language):

spacy/lang/ru/lex_attrs.py

@@ -1,56 +1,219 @@
 from ...attrs import LIKE_NUM
 
 
-_num_words = [
-    "ноль",
-    "один",
-    "два",
-    "три",
-    "четыре",
-    "пять",
-    "шесть",
-    "семь",
-    "восемь",
-    "девять",
-    "десять",
-    "одиннадцать",
-    "двенадцать",
-    "тринадцать",
-    "четырнадцать",
-    "пятнадцать",
-    "шестнадцать",
-    "семнадцать",
-    "восемнадцать",
-    "девятнадцать",
-    "двадцать",
-    "тридцать",
-    "сорок",
-    "пятьдесят",
-    "шестьдесят",
-    "семьдесят",
-    "восемьдесят",
-    "девяносто",
-    "сто",
-    "двести",
-    "триста",
-    "четыреста",
-    "пятьсот",
-    "шестьсот",
-    "семьсот",
-    "восемьсот",
-    "девятьсот",
-    "тысяча",
-    "миллион",
-    "миллиард",
-    "триллион",
-    "квадриллион",
-    "квинтиллион",
-]
+_num_words = list(
+    set(
+        """
+ноль ноля нолю нолём ноле нулевой нулевого нулевому нулевым нулевом нулевая нулевую нулевое нулевые нулевых нулевыми 
+
+четверть четверти четвертью четвертей четвертям четвертями четвертях 
+
+треть трети третью третей третям третями третях 
+
+половина половины половине половину половиной половин половинам половинами половинах половиною 
+
+один одного одному одним одном 
+первой первого первому первом первый первым первых 
+во-первых 
+единица единицы единице единицу единицей единиц единицам единицами единицах единицею 
+
+два двумя двум двух двоих двое две 
+второго второму второй втором вторым вторых 
+двойка двойки двойке двойку двойкой двоек двойкам двойками двойках двойкою  
+во-вторых 
+оба обе обеим обеими обеих обоим обоими обоих 
+
+полтора полторы полутора 
+
+три третьего третьему третьем третьим третий тремя трем трех трое троих трёх 
+тройка тройки тройке тройку тройкою троек тройкам тройками тройках тройкой 
+троечка троечки троечке троечку троечкой троечек троечкам троечками троечках троечкой 
+трешка трешки трешке трешку трешкой трешек трешкам трешками трешках трешкою 
+трёшка трёшки трёшке трёшку трёшкой трёшек трёшкам трёшками трёшках трёшкою 
+трояк трояка трояку трояком трояке трояки трояков троякам трояками трояках  
+треха треху трехой 
+трёха трёху трёхой 
+втроем втроём 
+
+четыре четвертого четвертому четвертом четвертый четвертым четверка четырьмя четырем четырех четверо четырёх четверым 
+четверых 
+вчетвером 
+
+пять пятого пятому пятом пятый пятым пятью пяти пятеро пятерых пятерыми 
+впятером 
+пятерочка пятерочки пятерочке пятерочками пятерочкой пятерочку пятерочкой пятерочками 
+пятёрочка пятёрочки пятёрочке пятёрочками пятёрочкой пятёрочку пятёрочкой пятёрочками 
+пятерка пятерки пятерке пятерками пятеркой пятерку пятерками 
+пятёрка пятёрки пятёрке пятёрками пятёркой пятёрку пятёрками 
+пятёра пятёры пятёре пятёрами пятёрой пятёру пятёрами 
+пятера пятеры пятере пятерами пятерой пятеру пятерами 
+пятак пятаки пятаке пятаками пятаком пятаку пятаками 
+
+шесть шестерка шестого шестому шестой шестом шестым шестью шести шестеро шестерых 
+вшестером 
+
+семь семерка седьмого седьмому седьмой седьмом седьмым семью семи семеро седьмых 
+всемером 
+
+восемь восьмерка восьмого восьмому восемью восьмой восьмом восьмым восеми восьмером восьми восьмью 
+восьмерых 
+ввосьмером 
+
+девять девятого девятому девятка девятом девятый девятым девятью девяти девятером вдевятером девятерых 
+вдевятером 
+
+десять десятого десятому десятка десятом десятый десятым десятью десяти десятером десятых 
+вдесятером 
+
+одиннадцать одиннадцатого одиннадцатому одиннадцатом одиннадцатый одиннадцатым одиннадцатью одиннадцати 
+одиннадцатых 
+
+двенадцать двенадцатого двенадцатому двенадцатом двенадцатый двенадцатым двенадцатью двенадцати 
+двенадцатых 
+
+тринадцать тринадцатого тринадцатому тринадцатом тринадцатый тринадцатым тринадцатью тринадцати 
+тринадцатых 
+
+четырнадцать четырнадцатого четырнадцатому четырнадцатом четырнадцатый четырнадцатым четырнадцатью четырнадцати 
+четырнадцатых 
+
+пятнадцать пятнадцатого пятнадцатому пятнадцатом пятнадцатый пятнадцатым пятнадцатью пятнадцати 
+пятнадцатых 
+пятнарик пятнарику пятнариком пятнарики 
+
+шестнадцать шестнадцатого шестнадцатому шестнадцатом шестнадцатый шестнадцатым шестнадцатью шестнадцати 
+шестнадцатых 
+
+семнадцать семнадцатого семнадцатому семнадцатом семнадцатый семнадцатым семнадцатью семнадцати семнадцатых 
+
+восемнадцать восемнадцатого восемнадцатому восемнадцатом восемнадцатый восемнадцатым восемнадцатью восемнадцати 
+восемнадцатых 
+
+девятнадцать девятнадцатого девятнадцатому девятнадцатом девятнадцатый девятнадцатым девятнадцатью девятнадцати 
+девятнадцатых 
+
+двадцать двадцатого двадцатому двадцатом двадцатый двадцатым двадцатью двадцати двадцатых 
+
+четвертак четвертака четвертаке четвертаку четвертаки четвертаком четвертаками 
+
+тридцать тридцатого тридцатому тридцатом тридцатый тридцатым тридцатью тридцати тридцатых 
+тридцадка тридцадку тридцадке тридцадки тридцадкой тридцадкою тридцадками 
+
+тридевять тридевяти тридевятью 
+
+сорок сорокового сороковому сороковом сороковым сороковой сороковых 
+сорокет сорокета сорокету сорокете сорокеты сорокетом сорокетами сорокетам 
+
+пятьдесят пятьдесятого пятьдесятому пятьюдесятью пятьдесятом пятьдесятый пятьдесятым пятидесяти пятьдесятых 
+полтинник полтинника полтиннике полтиннику полтинники полтинником полтинниками полтинникам полтинниках 
+пятидесятка пятидесятке пятидесятку пятидесятки пятидесяткой пятидесятками пятидесяткам пятидесятках 
+полтос полтоса полтосе полтосу полтосы полтосом полтосами полтосам полтосах 
+
+шестьдесят шестьдесятого шестьдесятому шестьюдесятью шестьдесятом шестьдесятый шестьдесятым шестидесятые шестидесяти 
+шестьдесятых 
+
+семьдесят семьдесятого семьдесятому семьюдесятью семьдесятом семьдесятый семьдесятым семидесяти семьдесятых 
+
+восемьдесят восемьдесятого восемьдесятому восемьюдесятью восемьдесятом восемьдесятый восемьдесятым восемидесяти 
+восьмидесяти восьмидесятых 
+
+девяносто девяностого девяностому девяностом девяностый девяностым девяноста девяностых 
+
+сто сотого сотому сотом сотен сотый сотым ста 
+стольник стольника стольнику стольнике стольники стольником стольниками 
+сотка сотки сотке соткой сотками соткам сотках 
+сотня сотни сотне сотней сотнями сотням сотнях 
+
+двести двумястами двухсотого двухсотому двухсотом двухсотый двухсотым двумстам двухстах двухсот 
+
+триста тремястами трехсотого трехсотому трехсотом трехсотый трехсотым тремстам трехстах трехсот 
+
+четыреста четырехсотого четырехсотому четырьмястами четырехсотом четырехсотый четырехсотым четыремстам четырехстах 
+четырехсот 
+
+пятьсот пятисотого пятисотому пятьюстами пятисотом пятисотый пятисотым пятистам пятистах пятисот 
+пятисотка пятисотки пятисотке пятисоткой пятисотками пятисоткам пятисоткою пятисотках 
+пятихатка пятихатки пятихатке пятихаткой пятихатками пятихаткам пятихаткою пятихатках 
+пятифан пятифаны пятифане пятифаном пятифанами пятифанах 
+
+шестьсот шестисотого шестисотому шестьюстами шестисотом шестисотый шестисотым шестистам шестистах шестисот 
+
+семьсот семисотого семисотому семьюстами семисотом семисотый семисотым семистам семистах семисот 
+
+восемьсот восемисотого восемисотому восемисотом восемисотый восемисотым восьмистами восьмистам восьмистах восьмисот 
+
+девятьсот девятисотого девятисотому девятьюстами девятисотом девятисотый девятисотым девятистам девятистах девятисот 
+
+тысяча тысячного тысячному тысячном тысячный тысячным тысячам тысячах тысячей тысяч тысячи тыс 
+косарь косаря косару косарем косарями косарях косарям косарей 
+
+десятитысячный десятитысячного десятитысячному десятитысячным десятитысячном десятитысячная десятитысячной 
+десятитысячную десятитысячною десятитысячное десятитысячные десятитысячных десятитысячными 
+
+двадцатитысячный двадцатитысячного двадцатитысячному двадцатитысячным двадцатитысячном двадцатитысячная 
+двадцатитысячной двадцатитысячную двадцатитысячною двадцатитысячное двадцатитысячные двадцатитысячных 
+двадцатитысячными 
+
+тридцатитысячный тридцатитысячного тридцатитысячному тридцатитысячным тридцатитысячном тридцатитысячная 
+тридцатитысячной тридцатитысячную тридцатитысячною тридцатитысячное тридцатитысячные тридцатитысячных 
+тридцатитысячными 
+
+сорокатысячный сорокатысячного сорокатысячному сорокатысячным сорокатысячном сорокатысячная 
+сорокатысячной сорокатысячную сорокатысячною сорокатысячное сорокатысячные сорокатысячных 
+сорокатысячными 
+
+пятидесятитысячный пятидесятитысячного пятидесятитысячному пятидесятитысячным пятидесятитысячном пятидесятитысячная 
+пятидесятитысячной пятидесятитысячную пятидесятитысячною пятидесятитысячное пятидесятитысячные пятидесятитысячных 
+пятидесятитысячными 
+
+шестидесятитысячный шестидесятитысячного шестидесятитысячному шестидесятитысячным шестидесятитысячном шестидесятитысячная 
+шестидесятитысячной шестидесятитысячную шестидесятитысячною шестидесятитысячное шестидесятитысячные шестидесятитысячных 
+шестидесятитысячными 
+
+семидесятитысячный семидесятитысячного семидесятитысячному семидесятитысячным семидесятитысячном семидесятитысячная 
+семидесятитысячной семидесятитысячную семидесятитысячною семидесятитысячное семидесятитысячные семидесятитысячных 
+семидесятитысячными 
+
+восьмидесятитысячный восьмидесятитысячного восьмидесятитысячному восьмидесятитысячным восьмидесятитысячном восьмидесятитысячная 
+восьмидесятитысячной восьмидесятитысячную восьмидесятитысячною восьмидесятитысячное восьмидесятитысячные восьмидесятитысячных 
+восьмидесятитысячными 
+
+стотысячный стотысячного стотысячному стотысячным стотысячном стотысячная стотысячной стотысячную стотысячное 
+стотысячные стотысячных стотысячными стотысячною 
+
+миллион миллионного миллионов миллионному миллионном миллионный миллионным миллионом миллиона миллионе миллиону 
+миллионов 
+лям ляма лямы лямом лямами лямах лямов 
+млн 
+
+десятимиллионная десятимиллионной десятимиллионными десятимиллионный десятимиллионным десятимиллионному 
+десятимиллионными десятимиллионную десятимиллионное  десятимиллионные десятимиллионных десятимиллионною 
+
+миллиард миллиардного миллиардному миллиардном миллиардный миллиардным миллиардом миллиарда миллиарде миллиарду 
+миллиардов 
+лярд лярда лярды лярдом лярдами лярдах лярдов 
+млрд 
+
+триллион триллионного триллионному триллионном триллионный триллионным триллионом триллиона триллионе триллиону 
+триллионов трлн 
+
+квадриллион квадриллионного квадриллионному квадриллионный квадриллионным квадриллионом квадриллиона квадриллионе 
+квадриллиону квадриллионов квадрлн 
+
+квинтиллион квинтиллионного квинтиллионному квинтиллионный квинтиллионным квинтиллионом квинтиллиона квинтиллионе 
+квинтиллиону квинтиллионов квинтлн 
+
+i ii iii iv v vi vii viii ix x xi xii xiii xiv xv xvi xvii xviii xix xx xxi xxii xxiii xxiv xxv xxvi xxvii xxvii xxix
+""".split()
+    )
+)
 
 
 def like_num(text):
     if text.startswith(("+", "-", "±", "~")):
         text = text[1:]
+    if text.endswith("%"):
+        text = text[:-1]
     text = text.replace(",", "").replace(".", "")
     if text.isdigit():
         return True

spacy/lang/ru/stop_words.py

@@ -1,52 +1,111 @@
 STOP_WORDS = set(
     """
-а
+а авось ага агу аж ай али алло ау ах ая
 
-будем будет будете будешь буду будут будучи будь будьте бы был была были было
-быть
+б будем будет будете будешь буду будут будучи будь будьте бы был была были было
+быть бац без безусловно бишь благо благодаря ближайшие близко более больше
+будто бывает бывала бывали бываю бывают бытует
 
 в вам вами вас весь во вот все всё всего всей всем всём всеми всему всех всею
-всея всю вся вы
+всея всю вся вы ваш ваша ваше ваши вдали вдобавок вдруг ведь везде вернее
+взаимно взаправду видно вишь включая вместо внакладе вначале вне вниз внизу
+вновь вовсе возможно воистину вокруг вон вообще вопреки вперекор вплоть
+вполне вправду вправе впрочем впрямь вресноту вроде вряд всегда всюду
+всякий всякого всякой всячески вчеред
 
-да для до
+г го где гораздо гав
 
-его едим едят ее её ей ел ела ем ему емъ если ест есть ешь еще ещё ею
+д да для до дабы давайте давно давным даже далее далеко дальше данная
+данного данное данной данном данному данные данный данных дану данунах
+даром де действительно довольно доколе доколь долго должен должна
+должно должны должный дополнительно другая другие другим другими
+других другое другой
 
-же
+е его едим едят ее её ей ел ела ем ему емъ если ест есть ешь еще ещё ею едва
+ежели еле
 
-за
+ж же
 
-и из или им ими имъ их
+з за затем зато зачем здесь значит зря
+
+и из или им ими имъ их ибо иль имеет имел имела имело именно иметь иначе
+иногда иным иными итак ишь
+
+й
 
 к как кем ко когда кого ком кому комья которая которого которое которой котором
-которому которою которую которые который которым которыми которых кто
+которому которою которую которые который которым которыми которых кто ка кабы
+каждая каждое каждые каждый кажется казалась казались казалось казался казаться
+какая какие каким какими каков какого какой какому какою касательно кой коли
+коль конечно короче кроме кстати ку куда
 
-меня мне мной мною мог моги могите могла могли могло могу могут мое моё моего
+л ли либо лишь любая любого любое любой любом любую любыми любых
+
+м меня мне мной мною мог моги могите могла могли могло могу могут мое моё моего
 моей моем моём моему моею можем может можете можешь мои мой моим моими моих
-мочь мою моя мы
+мочь мою моя мы мало меж между менее меньше мимо многие много многого многое
+многом многому можно мол му
 
-на нам нами нас наса наш наша наше нашего нашей нашем нашему нашею наши нашим
+н на нам нами нас наса наш наша наше нашего нашей нашем нашему нашею наши нашим
 нашими наших нашу не него нее неё ней нем нём нему нет нею ним ними них но
+наверняка наверху навряд навыворот над надо назад наиболее наизворот
+наизнанку наипаче накануне наконец наоборот наперед наперекор наподобие
+например напротив напрямую насилу настоящая настоящее настоящие настоящий
+насчет нате находиться начала начале неважно негде недавно недалеко незачем
+некем некогда некому некоторая некоторые некоторый некоторых некто некуда
+нельзя немногие немногим немного необходимо необходимости необходимые
+необходимым неоткуда непрерывно нередко несколько нету неужели нечего
+нечем нечему нечто нешто нибудь нигде ниже низко никак никакой никем
+никогда никого никому никто никуда ниоткуда нипочем ничего ничем ничему
+ничто ну нужная нужно нужного нужные нужный нужных ныне нынешнее нынешней
+нынешних нынче
 
 о об один одна одни одним одними одних одно одного одной одном одному одною
-одну он она оне они оно от
+одну он она оне они оно от оба общую обычно ого однажды однако ой около оный
+оп опять особенно особо особую особые откуда отнелижа отнелиже отовсюду
+отсюда оттого оттот оттуда отчего отчему ох очевидно очень ом
 
-по при
+п по при паче перед под подавно поди подобная подобно подобного подобные
+подобный подобным подобных поелику пожалуй пожалуйста позже поистине
+пока покамест поколе поколь покуда покудова помимо понеже поприще пор
+пора посему поскольку после посреди посредством потом потому потомушта
+похожем почему почти поэтому прежде притом причем про просто прочего
+прочее прочему прочими проще прям пусть
+
+р ради разве ранее рано раньше рядом
 
 с сам сама сами самим самими самих само самого самом самому саму свое своё
 своего своей своем своём своему своею свои свой своим своими своих свою своя
-себе себя собой собою
+себе себя собой собою самая самое самой самый самых сверх свыше се сего сей
+сейчас сие сих сквозь сколько скорее скоро следует слишком смогут сможет
+сначала снова со собственно совсем сперва спокону спустя сразу среди сродни
+стал стала стали стало стать суть сызнова
 
-та так такая такие таким такими таких такого такое такой таком такому такою
-такую те тебе тебя тем теми тех то тобой тобою того той только том томах тому
-тот тою ту ты
+та то ту ты ти так такая такие таким такими таких такого такое такой таком такому такою
+такую те тебе тебя тем теми тех тобой тобою того той только том томах тому
+тот тою также таки таков такова там твои твоим твоих твой твоя твоё
+теперь тогда тоже тотчас точно туда тут тьфу тая
 
-у уже
+у уже увы уж ура ух ую
 
-чего чем чём чему что чтобы
+ф фу
 
-эта эти этим этими этих это этого этой этом этому этот этою эту
+х ха хе хорошо хотел хотела хотелось хотеть хоть хотя хочешь хочу хуже
 
-я
+ч чего чем чём чему что чтобы часто чаще чей через чтоб чуть чхать чьим
+чьих чьё чё
+
+ш ша
+
+щ ща щас
+
+ы ых ые ый
+
+э эта эти этим этими этих это этого этой этом этому этот этою эту эдак эдакий
+эй эка экий этак этакий эх
+
+ю
+
+я явно явных яко якобы якоже
 """.split()
 )

spacy/lang/ru/tokenizer_exceptions.py

@@ -2,7 +2,6 @@ from ..tokenizer_exceptions import BASE_EXCEPTIONS
 from ...symbols import ORTH, NORM
 from ...util import update_exc
 
-
 _exc = {}
 
 _abbrev_exc = [
@@ -42,7 +41,6 @@ _abbrev_exc = [
     {ORTH: "дек", NORM: "декабрь"},
 ]
 
-
 for abbrev_desc in _abbrev_exc:
     abbrev = abbrev_desc[ORTH]
     for orth in (abbrev, abbrev.capitalize(), abbrev.upper()):
@@ -50,17 +48,354 @@ for abbrev_desc in _abbrev_exc:
         _exc[orth + "."] = [{ORTH: orth + ".", NORM: abbrev_desc[NORM]}]
 
 
-_slang_exc = [
+for abbr in [
+    # Year slang abbreviations
     {ORTH: "2к15", NORM: "2015"},
     {ORTH: "2к16", NORM: "2016"},
     {ORTH: "2к17", NORM: "2017"},
     {ORTH: "2к18", NORM: "2018"},
     {ORTH: "2к19", NORM: "2019"},
     {ORTH: "2к20", NORM: "2020"},
-]
+    {ORTH: "2к21", NORM: "2021"},
+    {ORTH: "2к22", NORM: "2022"},
+    {ORTH: "2к23", NORM: "2023"},
+    {ORTH: "2к24", NORM: "2024"},
+    {ORTH: "2к25", NORM: "2025"},
+]:
+    _exc[abbr[ORTH]] = [abbr]
 
-for slang_desc in _slang_exc:
-    _exc[slang_desc[ORTH]] = [slang_desc]
+for abbr in [
+    # Profession and academic titles abbreviations
+    {ORTH: "ак.", NORM: "академик"},
+    {ORTH: "акад.", NORM: "академик"},
+    {ORTH: "д-р архитектуры", NORM: "доктор архитектуры"},
+    {ORTH: "д-р биол. наук", NORM: "доктор биологических наук"},
+    {ORTH: "д-р ветеринар. наук", NORM: "доктор ветеринарных наук"},
+    {ORTH: "д-р воен. наук", NORM: "доктор военных наук"},
+    {ORTH: "д-р геогр. наук", NORM: "доктор географических наук"},
+    {ORTH: "д-р геол.-минерал. наук", NORM: "доктор геолого-минералогических наук"},
+    {ORTH: "д-р искусствоведения", NORM: "доктор искусствоведения"},
+    {ORTH: "д-р ист. наук", NORM: "доктор исторических наук"},
+    {ORTH: "д-р культурологии", NORM: "доктор культурологии"},
+    {ORTH: "д-р мед. наук", NORM: "доктор медицинских наук"},
+    {ORTH: "д-р пед. наук", NORM: "доктор педагогических наук"},
+    {ORTH: "д-р полит. наук", NORM: "доктор политических наук"},
+    {ORTH: "д-р психол. наук", NORM: "доктор психологических наук"},
+    {ORTH: "д-р с.-х. наук", NORM: "доктор сельскохозяйственных наук"},
+    {ORTH: "д-р социол. наук", NORM: "доктор социологических наук"},
+    {ORTH: "д-р техн. наук", NORM: "доктор технических наук"},
+    {ORTH: "д-р фармацевт. наук", NORM: "доктор фармацевтических наук"},
+    {ORTH: "д-р физ.-мат. наук", NORM: "доктор физико-математических наук"},
+    {ORTH: "д-р филол. наук", NORM: "доктор филологических наук"},
+    {ORTH: "д-р филос. наук", NORM: "доктор философских наук"},
+    {ORTH: "д-р хим. наук", NORM: "доктор химических наук"},
+    {ORTH: "д-р экон. наук", NORM: "доктор экономических наук"},
+    {ORTH: "д-р юрид. наук", NORM: "доктор юридических наук"},
+    {ORTH: "д-р", NORM: "доктор"},
+    {ORTH: "д.б.н.", NORM: "доктор биологических наук"},
+    {ORTH: "д.г.-м.н.", NORM: "доктор геолого-минералогических наук"},
+    {ORTH: "д.г.н.", NORM: "доктор географических наук"},
+    {ORTH: "д.и.н.", NORM: "доктор исторических наук"},
+    {ORTH: "д.иск.", NORM: "доктор искусствоведения"},
+    {ORTH: "д.м.н.", NORM: "доктор медицинских наук"},
+    {ORTH: "д.п.н.", NORM: "доктор психологических наук"},
+    {ORTH: "д.пед.н.", NORM: "доктор педагогических наук"},
+    {ORTH: "д.полит.н.", NORM: "доктор политических наук"},
+    {ORTH: "д.с.-х.н.", NORM: "доктор сельскохозяйственных наук"},
+    {ORTH: "д.социол.н.", NORM: "доктор социологических наук"},
+    {ORTH: "д.т.н.", NORM: "доктор технических наук"},
+    {ORTH: "д.т.н", NORM: "доктор технических наук"},
+    {ORTH: "д.ф.-м.н.", NORM: "доктор физико-математических наук"},
+    {ORTH: "д.ф.н.", NORM: "доктор филологических наук"},
+    {ORTH: "д.филос.н.", NORM: "доктор философских наук"},
+    {ORTH: "д.фил.н.", NORM: "доктор филологических наук"},
+    {ORTH: "д.х.н.", NORM: "доктор химических наук"},
+    {ORTH: "д.э.н.", NORM: "доктор экономических наук"},
+    {ORTH: "д.э.н", NORM: "доктор экономических наук"},
+    {ORTH: "д.ю.н.", NORM: "доктор юридических наук"},
+    {ORTH: "доц.", NORM: "доцент"},
+    {ORTH: "и.о.", NORM: "исполняющий обязанности"},
+    {ORTH: "к.б.н.", NORM: "кандидат биологических наук"},
+    {ORTH: "к.воен.н.", NORM: "кандидат военных наук"},
+    {ORTH: "к.г.-м.н.", NORM: "кандидат геолого-минералогических наук"},
+    {ORTH: "к.г.н.", NORM: "кандидат географических наук"},
+    {ORTH: "к.геогр.н", NORM: "кандидат географических наук"},
+    {ORTH: "к.геогр.наук", NORM: "кандидат географических наук"},
+    {ORTH: "к.и.н.", NORM: "кандидат исторических наук"},
+    {ORTH: "к.иск.", NORM: "кандидат искусствоведения"},
+    {ORTH: "к.м.н.", NORM: "кандидат медицинских наук"},
+    {ORTH: "к.п.н.", NORM: "кандидат психологических наук"},
+    {ORTH: "к.псх.н.", NORM: "кандидат психологических наук"},
+    {ORTH: "к.пед.н.", NORM: "кандидат педагогических наук"},
+    {ORTH: "канд.пед.наук", NORM: "кандидат педагогических наук"},
+    {ORTH: "к.полит.н.", NORM: "кандидат политических наук"},
+    {ORTH: "к.с.-х.н.", NORM: "кандидат сельскохозяйственных наук"},
+    {ORTH: "к.социол.н.", NORM: "кандидат социологических наук"},
+    {ORTH: "к.с.н.", NORM: "кандидат социологических наук"},
+    {ORTH: "к.т.н.", NORM: "кандидат технических наук"},
+    {ORTH: "к.ф.-м.н.", NORM: "кандидат физико-математических наук"},
+    {ORTH: "к.ф.н.", NORM: "кандидат филологических наук"},
+    {ORTH: "к.фил.н.", NORM: "кандидат филологических наук"},
+    {ORTH: "к.филол.н", NORM: "кандидат филологических наук"},
+    {ORTH: "к.фарм.наук", NORM: "кандидат фармакологических наук"},
+    {ORTH: "к.фарм.н.", NORM: "кандидат фармакологических наук"},
+    {ORTH: "к.фарм.н", NORM: "кандидат фармакологических наук"},
+    {ORTH: "к.филос.наук", NORM: "кандидат философских наук"},
+    {ORTH: "к.филос.н.", NORM: "кандидат философских наук"},
+    {ORTH: "к.филос.н", NORM: "кандидат философских наук"},
+    {ORTH: "к.х.н.", NORM: "кандидат химических наук"},
+    {ORTH: "к.х.н", NORM: "кандидат химических наук"},
+    {ORTH: "к.э.н.", NORM: "кандидат экономических наук"},
+    {ORTH: "к.э.н", NORM: "кандидат экономических наук"},
+    {ORTH: "к.ю.н.", NORM: "кандидат юридических наук"},
+    {ORTH: "к.ю.н", NORM: "кандидат юридических наук"},
+    {ORTH: "канд. архитектуры", NORM: "кандидат архитектуры"},
+    {ORTH: "канд. биол. наук", NORM: "кандидат биологических наук"},
+    {ORTH: "канд. ветеринар. наук", NORM: "кандидат ветеринарных наук"},
+    {ORTH: "канд. воен. наук", NORM: "кандидат военных наук"},
+    {ORTH: "канд. геогр. наук", NORM: "кандидат географических наук"},
+    {ORTH: "канд. геол.-минерал. наук", NORM: "кандидат геолого-минералогических наук"},
+    {ORTH: "канд. искусствоведения", NORM: "кандидат искусствоведения"},
+    {ORTH: "канд. ист. наук", NORM: "кандидат исторических наук"},
+    {ORTH: "к.ист.н.", NORM: "кандидат исторических наук"},
+    {ORTH: "канд. культурологии", NORM: "кандидат культурологии"},
+    {ORTH: "канд. мед. наук", NORM: "кандидат медицинских наук"},
+    {ORTH: "канд. пед. наук", NORM: "кандидат педагогических наук"},
+    {ORTH: "канд. полит. наук", NORM: "кандидат политических наук"},
+    {ORTH: "канд. психол. наук", NORM: "кандидат психологических наук"},
+    {ORTH: "канд. с.-х. наук", NORM: "кандидат сельскохозяйственных наук"},
+    {ORTH: "канд. социол. наук", NORM: "кандидат социологических наук"},
+    {ORTH: "к.соц.наук", NORM: "кандидат социологических наук"},
+    {ORTH: "к.соц.н.", NORM: "кандидат социологических наук"},
+    {ORTH: "к.соц.н", NORM: "кандидат социологических наук"},
+    {ORTH: "канд. техн. наук", NORM: "кандидат технических наук"},
+    {ORTH: "канд. фармацевт. наук", NORM: "кандидат фармацевтических наук"},
+    {ORTH: "канд. физ.-мат. наук", NORM: "кандидат физико-математических наук"},
+    {ORTH: "канд. филол. наук", NORM: "кандидат филологических наук"},
+    {ORTH: "канд. филос. наук", NORM: "кандидат философских наук"},
+    {ORTH: "канд. хим. наук", NORM: "кандидат химических наук"},
+    {ORTH: "канд. экон. наук", NORM: "кандидат экономических наук"},
+    {ORTH: "канд. юрид. наук", NORM: "кандидат юридических наук"},
+    {ORTH: "в.н.с.", NORM: "ведущий научный сотрудник"},
+    {ORTH: "мл. науч. сотр.", NORM: "младший научный сотрудник"},
+    {ORTH: "м.н.с.", NORM: "младший научный сотрудник"},
+    {ORTH: "проф.", NORM: "профессор"},
+    {ORTH: "профессор.кафедры", NORM: "профессор кафедры"},
+    {ORTH: "ст. науч. сотр.", NORM: "старший научный сотрудник"},
+    {ORTH: "чл.-к.", NORM: "член корреспондент"},
+    {ORTH: "чл.-корр.", NORM: "член-корреспондент"},
+    {ORTH: "чл.-кор.", NORM: "член-корреспондент"},
+    {ORTH: "дир.", NORM: "директор"},
+    {ORTH: "зам. дир.", NORM: "заместитель директора"},
+    {ORTH: "зав. каф.", NORM: "заведующий кафедрой"},
+    {ORTH: "зав.кафедрой", NORM: "заведующий кафедрой"},
+    {ORTH: "зав. кафедрой", NORM: "заведующий кафедрой"},
+    {ORTH: "асп.", NORM: "аспирант"},
+    {ORTH: "гл. науч. сотр.", NORM: "главный научный сотрудник"},
+    {ORTH: "вед. науч. сотр.", NORM: "ведущий научный сотрудник"},
+    {ORTH: "науч. сотр.", NORM: "научный сотрудник"},
+    {ORTH: "к.м.с.", NORM: "кандидат в мастера спорта"},
+]:
+    _exc[abbr[ORTH]] = [abbr]
+
+
+for abbr in [
+    # Literary phrases abbreviations
+    {ORTH: "и т.д.", NORM: "и так далее"},
+    {ORTH: "и т.п.", NORM: "и тому подобное"},
+    {ORTH: "т.д.", NORM: "так далее"},
+    {ORTH: "т.п.", NORM: "тому подобное"},
+    {ORTH: "т.е.", NORM: "то есть"},
+    {ORTH: "т.к.", NORM: "так как"},
+    {ORTH: "в т.ч.", NORM: "в том числе"},
+    {ORTH: "и пр.", NORM: "и прочие"},
+    {ORTH: "и др.", NORM: "и другие"},
+    {ORTH: "т.н.", NORM: "так называемый"},
+]:
+    _exc[abbr[ORTH]] = [abbr]
+
+
+for abbr in [
+    # Appeal to a person abbreviations
+    {ORTH: "г-н", NORM: "господин"},
+    {ORTH: "г-да", NORM: "господа"},
+    {ORTH: "г-жа", NORM: "госпожа"},
+    {ORTH: "тов.", NORM: "товарищ"},
+]:
+    _exc[abbr[ORTH]] = [abbr]
+
+
+for abbr in [
+    # Time periods abbreviations
+    {ORTH: "до н.э.", NORM: "до нашей эры"},
+    {ORTH: "по н.в.", NORM: "по настоящее время"},
+    {ORTH: "в н.в.", NORM: "в настоящее время"},
+    {ORTH: "наст.", NORM: "настоящий"},
+    {ORTH: "наст. время", NORM: "настоящее время"},
+    {ORTH: "г.г.", NORM: "годы"},
+    {ORTH: "гг.", NORM: "годы"},
+    {ORTH: "т.г.", NORM: "текущий год"},
+]:
+    _exc[abbr[ORTH]] = [abbr]
+
+
+for abbr in [
+    # Address forming elements abbreviations
+    {ORTH: "респ.", NORM: "республика"},
+    {ORTH: "обл.", NORM: "область"},
+    {ORTH: "г.ф.з.", NORM: "город федерального значения"},
+    {ORTH: "а.обл.", NORM: "автономная область"},
+    {ORTH: "а.окр.", NORM: "автономный округ"},
+    {ORTH: "м.р-н", NORM: "муниципальный район"},
+    {ORTH: "г.о.", NORM: "городской округ"},
+    {ORTH: "г.п.", NORM: "городское поселение"},
+    {ORTH: "с.п.", NORM: "сельское поселение"},
+    {ORTH: "вн.р-н", NORM: "внутригородской район"},
+    {ORTH: "вн.тер.г.", NORM: "внутригородская территория города"},
+    {ORTH: "пос.", NORM: "поселение"},
+    {ORTH: "р-н", NORM: "район"},
+    {ORTH: "с/с", NORM: "сельсовет"},
+    {ORTH: "г.", NORM: "город"},
+    {ORTH: "п.г.т.", NORM: "поселок городского типа"},
+    {ORTH: "пгт.", NORM: "поселок городского типа"},
+    {ORTH: "р.п.", NORM: "рабочий поселок"},
+    {ORTH: "рп.", NORM: "рабочий поселок"},
+    {ORTH: "кп.", NORM: "курортный поселок"},
+    {ORTH: "гп.", NORM: "городской поселок"},
+    {ORTH: "п.", NORM: "поселок"},
+    {ORTH: "в-ки", NORM: "выселки"},
+    {ORTH: "г-к", NORM: "городок"},
+    {ORTH: "з-ка", NORM: "заимка"},
+    {ORTH: "п-к", NORM: "починок"},
+    {ORTH: "киш.", NORM: "кишлак"},
+    {ORTH: "п. ст. ", NORM: "поселок станция"},
+    {ORTH: "п. ж/д ст. ", NORM: "поселок при железнодорожной станции"},
+    {ORTH: "ж/д бл-ст", NORM: "железнодорожный блокпост"},
+    {ORTH: "ж/д б-ка", NORM: "железнодорожная будка"},
+    {ORTH: "ж/д в-ка", NORM: "железнодорожная ветка"},
+    {ORTH: "ж/д к-ма", NORM: "железнодорожная казарма"},
+    {ORTH: "ж/д к-т", NORM: "железнодорожный комбинат"},
+    {ORTH: "ж/д пл-ма", NORM: "железнодорожная платформа"},
+    {ORTH: "ж/д пл-ка", NORM: "железнодорожная площадка"},
+    {ORTH: "ж/д п.п.", NORM: "железнодорожный путевой пост"},
+    {ORTH: "ж/д о.п.", NORM: "железнодорожный остановочный пункт"},
+    {ORTH: "ж/д рзд.", NORM: "железнодорожный разъезд"},
+    {ORTH: "ж/д ст. ", NORM: "железнодорожная станция"},
+    {ORTH: "м-ко", NORM: "местечко"},
+    {ORTH: "д.", NORM: "деревня"},
+    {ORTH: "с.", NORM: "село"},
+    {ORTH: "сл.", NORM: "слобода"},
+    {ORTH: "ст. ", NORM: "станция"},
+    {ORTH: "ст-ца", NORM: "станица"},
+    {ORTH: "у.", NORM: "улус"},
+    {ORTH: "х.", NORM: "хутор"},
+    {ORTH: "рзд.", NORM: "разъезд"},
+    {ORTH: "зим.", NORM: "зимовье"},
+    {ORTH: "б-г", NORM: "берег"},
+    {ORTH: "ж/р", NORM: "жилой район"},
+    {ORTH: "кв-л", NORM: "квартал"},
+    {ORTH: "мкр.", NORM: "микрорайон"},
+    {ORTH: "ост-в", NORM: "остров"},
+    {ORTH: "платф.", NORM: "платформа"},
+    {ORTH: "п/р", NORM: "промышленный район"},
+    {ORTH: "р-н", NORM: "район"},
+    {ORTH: "тер.", NORM: "территория"},
+    {
+        ORTH: "тер. СНО",
+        NORM: "территория садоводческих некоммерческих объединений граждан",
+    },
+    {
+        ORTH: "тер. ОНО",
+        NORM: "территория огороднических некоммерческих объединений граждан",
+    },
+    {ORTH: "тер. ДНО", NORM: "территория дачных некоммерческих объединений граждан"},
+    {ORTH: "тер. СНТ", NORM: "территория садоводческих некоммерческих товариществ"},
+    {ORTH: "тер. ОНТ", NORM: "территория огороднических некоммерческих товариществ"},
+    {ORTH: "тер. ДНТ", NORM: "территория дачных некоммерческих товариществ"},
+    {ORTH: "тер. СПК", NORM: "территория садоводческих потребительских кооперативов"},
+    {ORTH: "тер. ОПК", NORM: "территория огороднических потребительских кооперативов"},
+    {ORTH: "тер. ДПК", NORM: "территория дачных потребительских кооперативов"},
+    {ORTH: "тер. СНП", NORM: "территория садоводческих некоммерческих партнерств"},
+    {ORTH: "тер. ОНП", NORM: "территория огороднических некоммерческих партнерств"},
+    {ORTH: "тер. ДНП", NORM: "территория дачных некоммерческих партнерств"},
+    {ORTH: "тер. ТСН", NORM: "территория товарищества собственников недвижимости"},
+    {ORTH: "тер. ГСК", NORM: "территория гаражно-строительного кооператива"},
+    {ORTH: "ус.", NORM: "усадьба"},
+    {ORTH: "тер.ф.х.", NORM: "территория фермерского хозяйства"},
+    {ORTH: "ю.", NORM: "юрты"},
+    {ORTH: "ал.", NORM: "аллея"},
+    {ORTH: "б-р", NORM: "бульвар"},
+    {ORTH: "взв.", NORM: "взвоз"},
+    {ORTH: "взд.", NORM: "въезд"},
+    {ORTH: "дор.", NORM: "дорога"},
+    {ORTH: "ззд.", NORM: "заезд"},
+    {ORTH: "км", NORM: "километр"},
+    {ORTH: "к-цо", NORM: "кольцо"},
+    {ORTH: "лн.", NORM: "линия"},
+    {ORTH: "мгстр.", NORM: "магистраль"},
+    {ORTH: "наб.", NORM: "набережная"},
+    {ORTH: "пер-д", NORM: "переезд"},
+    {ORTH: "пер.", NORM: "переулок"},
+    {ORTH: "пл-ка", NORM: "площадка"},
+    {ORTH: "пл.", NORM: "площадь"},
+    {ORTH: "пр-д", NORM: "проезд"},
+    {ORTH: "пр-к", NORM: "просек"},
+    {ORTH: "пр-ка", NORM: "просека"},
+    {ORTH: "пр-лок", NORM: "проселок"},
+    {ORTH: "пр-кт", NORM: "проспект"},
+    {ORTH: "проул.", NORM: "проулок"},
+    {ORTH: "рзд.", NORM: "разъезд"},
+    {ORTH: "ряд", NORM: "ряд(ы)"},
+    {ORTH: "с-р", NORM: "сквер"},
+    {ORTH: "с-к", NORM: "спуск"},
+    {ORTH: "сзд.", NORM: "съезд"},
+    {ORTH: "туп.", NORM: "тупик"},
+    {ORTH: "ул.", NORM: "улица"},
+    {ORTH: "ш.", NORM: "шоссе"},
+    {ORTH: "влд.", NORM: "владение"},
+    {ORTH: "г-ж", NORM: "гараж"},
+    {ORTH: "д.", NORM: "дом"},
+    {ORTH: "двлд.", NORM: "домовладение"},
+    {ORTH: "зд.", NORM: "здание"},
+    {ORTH: "з/у", NORM: "земельный участок"},
+    {ORTH: "кв.", NORM: "квартира"},
+    {ORTH: "ком.", NORM: "комната"},
+    {ORTH: "подв.", NORM: "подвал"},
+    {ORTH: "кот.", NORM: "котельная"},
+    {ORTH: "п-б", NORM: "погреб"},
+    {ORTH: "к.", NORM: "корпус"},
+    {ORTH: "ОНС", NORM: "объект незавершенного строительства"},
+    {ORTH: "оф.", NORM: "офис"},
+    {ORTH: "пав.", NORM: "павильон"},
+    {ORTH: "помещ.", NORM: "помещение"},
+    {ORTH: "раб.уч.", NORM: "рабочий участок"},
+    {ORTH: "скл.", NORM: "склад"},
+    {ORTH: "coop.", NORM: "сооружение"},
+    {ORTH: "стр.", NORM: "строение"},
+    {ORTH: "торг.зал", NORM: "торговый зал"},
+    {ORTH: "а/п", NORM: "аэропорт"},
+    {ORTH: "им.", NORM: "имени"},
+]:
+    _exc[abbr[ORTH]] = [abbr]
+
+
+for abbr in [
+    # Others abbreviations
+    {ORTH: "тыс.руб.", NORM: "тысяч рублей"},
+    {ORTH: "тыс.", NORM: "тысяч"},
+    {ORTH: "руб.", NORM: "рубль"},
+    {ORTH: "долл.", NORM: "доллар"},
+    {ORTH: "прим.", NORM: "примечание"},
+    {ORTH: "прим.ред.", NORM: "примечание редакции"},
+    {ORTH: "см. также", NORM: "смотри также"},
+    {ORTH: "кв.м.", NORM: "квадрантный метр"},
+    {ORTH: "м2", NORM: "квадрантный метр"},
+    {ORTH: "б/у", NORM: "бывший в употреблении"},
+    {ORTH: "сокр.", NORM: "сокращение"},
+    {ORTH: "чел.", NORM: "человек"},
+    {ORTH: "б.п.", NORM: "базисный пункт"},
+]:
+    _exc[abbr[ORTH]] = [abbr]
 
 
 TOKENIZER_EXCEPTIONS = update_exc(BASE_EXCEPTIONS, _exc)

spacy/lang/sl/examples.py

@@ -0,0 +1,18 @@
+"""
+Example sentences to test spaCy and its language models.
+
+>>> from spacy.lang.sl.examples import sentences
+>>> docs = nlp.pipe(sentences)
+"""
+
+
+sentences = [
+    "Apple načrtuje nakup britanskega startupa za 1 bilijon dolarjev",
+    "France Prešeren je umrl 8. februarja 1849 v Kranju",
+    "Staro ljubljansko letališče Moste bo obnovila družba BTC",
+    "London je največje mesto v Združenem kraljestvu.",
+    "Kje se skrivaš?",
+    "Kdo je predsednik Francije?",
+    "Katero je glavno mesto Združenih držav Amerike?",
+    "Kdaj je bil rojen Milan Kučan?",
+]

spacy/lang/sl/stop_words.py

@@ -1,13 +1,10 @@
 # Source: https://github.com/stopwords-iso/stopwords-sl
-# TODO: probably needs to be tidied up – the list seems to have month names in
-# it, which shouldn't be considered stop words.
+# Removed various words that are not normally considered stop words, such as months.
 
 STOP_WORDS = set(
     """
 a
 ali
-april
-avgust
 b
 bi
 bil
@@ -19,7 +16,6 @@ biti
 blizu
 bo
 bodo
-bojo
 bolj
 bom
 bomo
@@ -37,16 +33,6 @@ da
 daleč
 dan
 danes
-datum
-december
-deset
-deseta
-deseti
-deseto
-devet
-deveta
-deveti
-deveto
 do
 dober
 dobra
@@ -54,16 +40,7 @@ dobri
 dobro
 dokler
 dol
-dolg
-dolga
-dolgi
 dovolj
-drug
-druga
-drugi
-drugo
-dva
-dve
 e
 eden
 en
@@ -74,7 +51,6 @@ enkrat
 eno
 etc.
 f
-februar
 g
 g.
 ga
@@ -93,16 +69,12 @@ iv
 ix
 iz
 j
-januar
 jaz
 je
 ji
 jih
 jim
 jo
-julij
-junij
-jutri
 k
 kadarkoli
 kaj
@@ -123,41 +95,23 @@ kje
 kjer
 kjerkoli
 ko
-koder
 koderkoli
 koga
 komu
 kot
-kratek
-kratka
-kratke
-kratki
 l
-lahka
-lahke
-lahki
-lahko
 le
 lep
 lepa
 lepe
 lepi
 lepo
-leto
 m
-maj
-majhen
-majhna
-majhni
-malce
-malo
 manj
-marec
 me
 med
 medtem
 mene
-mesec
 mi
 midva
 midve
@@ -183,7 +137,6 @@ najmanj
 naju
 največ
 nam
-narobe
 nas
 nato
 nazaj
@@ -192,7 +145,6 @@ naša
 naše
 ne
 nedavno
-nedelja
 nek
 neka
 nekaj
@@ -236,7 +188,6 @@ njuna
 njuno
 no
 nocoj
-november
 npr.
 o
 ob
@@ -244,51 +195,23 @@ oba
 obe
 oboje
 od
-odprt
-odprta
-odprti
 okoli
-oktober
 on
 onadva
 one
 oni
 onidve
-osem
-osma
-osmi
-osmo
 oz.
 p
 pa
-pet
-peta
-petek
-peti
-peto
 po
 pod
 pogosto
 poleg
-poln
-polna
-polni
-polno
 ponavadi
-ponedeljek
 ponovno
 potem
 povsod
-pozdravljen
-pozdravljeni
-prav
-prava
-prave
-pravi
-pravo
-prazen
-prazna
-prazno
 prbl.
 precej
 pred
@@ -297,19 +220,10 @@ preko
 pri
 pribl.
 približno
-primer
-pripravljen
-pripravljena
-pripravljeni
 proti
-prva
-prvi
-prvo
 r
-ravno
 redko
 res
-reč
 s
 saj
 sam
@@ -321,29 +235,17 @@ se
 sebe
 sebi
 sedaj
-sedem
-sedma
-sedmi
-sedmo
 sem
-september
 seveda
 si
 sicer
 skoraj
 skozi
-slab
 smo
 so
-sobota
 spet
-sreda
-srednja
-srednji
 sta
 ste
-stran
-stvar
 sva
 t
 ta
@@ -358,10 +260,6 @@ te
 tebe
 tebi
 tega
-težak
-težka
-težki
-težko
 ti
 tista
 tiste
@@ -371,11 +269,6 @@ tj.
 tja
 to
 toda
-torek
-tretja
-tretje
-tretji
-tri
 tu
 tudi
 tukaj
@@ -392,10 +285,6 @@ vaša
 vaše
 ve
 vedno
-velik
-velika
-veliki
-veliko
 vendar
 ves
 več
@@ -403,10 +292,6 @@ vi
 vidva
 vii
 viii
-visok
-visoka
-visoke
-visoki
 vsa
 vsaj
 vsak
@@ -420,34 +305,21 @@ vsega
 vsi
 vso
 včasih
-včeraj
 x
 z
 za
 zadaj
 zadnji
 zakaj
-zaprta
-zaprti
-zaprto
 zdaj
 zelo
 zunaj
 č
 če
 često
-četrta
-četrtek
-četrti
-četrto
 čez
 čigav
 š
-šest
-šesta
-šesti
-šesto
-štiri
 ž
 že
 """.split()

spacy/lang/tr/lex_attrs.py

@@ -53,7 +53,7 @@ _ordinal_words = [
     "doksanıncı",
     "yüzüncü",
     "bininci",
-    "mliyonuncu",
+    "milyonuncu",
     "milyarıncı",
     "trilyonuncu",
     "katrilyonuncu",

spacy/lang/uk/__init__.py

@@ -6,6 +6,8 @@ from .tokenizer_exceptions import TOKENIZER_EXCEPTIONS
 from .stop_words import STOP_WORDS
 from .lex_attrs import LEX_ATTRS
 from .lemmatizer import UkrainianLemmatizer
+from ..punctuation import COMBINING_DIACRITICS_TOKENIZER_INFIXES
+from ..punctuation import COMBINING_DIACRITICS_TOKENIZER_SUFFIXES
 from ...language import Language, BaseDefaults
 
 
@@ -13,6 +15,8 @@ class UkrainianDefaults(BaseDefaults):
     tokenizer_exceptions = TOKENIZER_EXCEPTIONS
     lex_attr_getters = LEX_ATTRS
     stop_words = STOP_WORDS
+    suffixes = COMBINING_DIACRITICS_TOKENIZER_SUFFIXES
+    infixes = COMBINING_DIACRITICS_TOKENIZER_INFIXES
 
 
 class Ukrainian(Language):

spacy/lang/uk/tokenizer_exceptions.py

@@ -6,19 +6,30 @@ from ...util import update_exc
 _exc = {}
 
 for exc_data in [
+    {ORTH: "обл.", NORM: "область"},
+    {ORTH: "р-н.", NORM: "район"},
+    {ORTH: "р-н", NORM: "район"},
+    {ORTH: "м.", NORM: "місто"},
     {ORTH: "вул.", NORM: "вулиця"},
-    {ORTH: "ім.", NORM: "імені"},
     {ORTH: "просп.", NORM: "проспект"},
+    {ORTH: "пр-кт", NORM: "проспект"},
     {ORTH: "бул.", NORM: "бульвар"},
     {ORTH: "пров.", NORM: "провулок"},
     {ORTH: "пл.", NORM: "площа"},
+    {ORTH: "майд.", NORM: "майдан"},
+    {ORTH: "мкр.", NORM: "мікрорайон"},
+    {ORTH: "ст.", NORM: "станція"},
+    {ORTH: "ж/м", NORM: "житловий масив"},
+    {ORTH: "наб.", NORM: "набережна"},
+    {ORTH: "в/ч", NORM: "військова частина"},
+    {ORTH: "в/м", NORM: "військове містечко"},
+    {ORTH: "оз.", NORM: "озеро"},
+    {ORTH: "ім.", NORM: "імені"},
     {ORTH: "г.", NORM: "гора"},
     {ORTH: "п.", NORM: "пан"},
-    {ORTH: "м.", NORM: "місто"},
     {ORTH: "проф.", NORM: "професор"},
     {ORTH: "акад.", NORM: "академік"},
     {ORTH: "доц.", NORM: "доцент"},
-    {ORTH: "оз.", NORM: "озеро"},
 ]:
     _exc[exc_data[ORTH]] = [exc_data]
 

spacy/lang/vi/lex_attrs.py

@@ -2,22 +2,29 @@ from ...attrs import LIKE_NUM
 
 
 _num_words = [
-    "không",
-    "một",
-    "hai",
-    "ba",
-    "bốn",
-    "năm",
-    "sáu",
-    "bảy",
-    "bẩy",
-    "tám",
-    "chín",
-    "mười",
-    "chục",
-    "trăm",
-    "nghìn",
-    "tỷ",
+    "không",  # Zero
+    "một",  # One
+    "mốt",  # Also one, irreplacable in niché cases for unit digit such as "51"="năm mươi mốt"
+    "hai",  # Two
+    "ba",  # Three
+    "bốn",  # Four
+    "tư",  # Also four, used in certain cases for unit digit such as "54"="năm mươi tư"
+    "năm",  # Five
+    "lăm",  # Also five, irreplacable in niché cases for unit digit such as "55"="năm mươi lăm"
+    "sáu",  # Six
+    "bảy",  # Seven
+    "bẩy",  # Also seven, old fashioned
+    "tám",  # Eight
+    "chín",  # Nine
+    "mười",  # Ten
+    "chục",  # Also ten, used for counting in tens such as "20 eggs"="hai chục trứng"
+    "trăm",  # Hundred
+    "nghìn",  # Thousand
+    "ngàn",  # Also thousand, used in the south
+    "vạn",  # Ten thousand
+    "triệu",  # Million
+    "tỷ",  # Billion
+    "tỉ",  # Also billion, used in combinatorics such as "tỉ_phú"="billionaire"
 ]
 
 

spacy/lang/xx/examples.py

@@ -59,7 +59,7 @@ sentences = [
     "Czy w ciągu ostatnich 48 godzin spożyłeś leki zawierające paracetamol?",
     "Kto ma ochotę zapoznać się z innymi niż w książkach przygodami Muminków i ich przyjaciół, temu polecam komiks Tove Jansson „Muminki i morze”.",
     "Apple está querendo comprar uma startup do Reino Unido por 100 milhões de dólares.",
-    "Carros autônomos empurram a responsabilidade do seguro para os fabricantes.."
+    "Carros autônomos empurram a responsabilidade do seguro para os fabricantes..",
     "São Francisco considera banir os robôs de entrega que andam pelas calçadas.",
     "Londres é a maior cidade do Reino Unido.",
     # Translations from English:

spacy/language.py

@@ -1,4 +1,4 @@
-from typing import Iterator, Optional, Any, Dict, Callable, Iterable
+from typing import Iterator, Optional, Any, Dict, Callable, Iterable, Collection
 from typing import Union, Tuple, List, Set, Pattern, Sequence
 from typing import NoReturn, TYPE_CHECKING, TypeVar, cast, overload
 
@@ -131,7 +131,7 @@ class Language:
         self,
         vocab: Union[Vocab, bool] = True,
         *,
-        max_length: int = 10 ** 6,
+        max_length: int = 10**6,
         meta: Dict[str, Any] = {},
         create_tokenizer: Optional[Callable[["Language"], Callable[[str], Doc]]] = None,
         batch_size: int = 1000,
@@ -354,12 +354,15 @@ class Language:
     @property
     def pipe_labels(self) -> Dict[str, List[str]]:
         """Get the labels set by the pipeline components, if available (if
-        the component exposes a labels property).
+        the component exposes a labels property and the labels are not
+        hidden).
 
         RETURNS (Dict[str, List[str]]): Labels keyed by component name.
         """
         labels = {}
         for name, pipe in self._components:
+            if hasattr(pipe, "hide_labels") and pipe.hide_labels is True:
+                continue
             if hasattr(pipe, "labels"):
                 labels[name] = list(pipe.labels)
         return SimpleFrozenDict(labels)
@@ -522,7 +525,7 @@ class Language:
         requires: Iterable[str] = SimpleFrozenList(),
         retokenizes: bool = False,
         func: Optional["Pipe"] = None,
-    ) -> Callable:
+    ) -> Callable[..., Any]:
         """Register a new pipeline component. Can be used for stateless function
         components that don't require a separate factory. Can be used as a
         decorator on a function or classmethod, or called as a function with the
@@ -771,6 +774,9 @@ class Language:
         name = name if name is not None else factory_name
         if name in self.component_names:
             raise ValueError(Errors.E007.format(name=name, opts=self.component_names))
+        # Overriding pipe name in the config is not supported and will be ignored.
+        if "name" in config:
+            warnings.warn(Warnings.W119.format(name_in_config=config.pop("name")))
         if source is not None:
             # We're loading the component from a model. After loading the
             # component, we know its real factory name
@@ -1084,16 +1090,21 @@ class Language:
             )
         return self.tokenizer(text)
 
-    def _ensure_doc(self, doc_like: Union[str, Doc]) -> Doc:
-        """Create a Doc if need be, or raise an error if the input is not a Doc or a string."""
+    def _ensure_doc(self, doc_like: Union[str, Doc, bytes]) -> Doc:
+        """Create a Doc if need be, or raise an error if the input is not
+        a Doc, string, or a byte array (generated by Doc.to_bytes())."""
         if isinstance(doc_like, Doc):
             return doc_like
         if isinstance(doc_like, str):
             return self.make_doc(doc_like)
-        raise ValueError(Errors.E866.format(type=type(doc_like)))
+        if isinstance(doc_like, bytes):
+            return Doc(self.vocab).from_bytes(doc_like)
+        raise ValueError(Errors.E1041.format(type=type(doc_like)))
 
-    def _ensure_doc_with_context(self, doc_like: Union[str, Doc], context: Any) -> Doc:
-        """Create a Doc if need be and add as_tuples context, or raise an error if the input is not a Doc or a string."""
+    def _ensure_doc_with_context(
+        self, doc_like: Union[str, Doc, bytes], context: _AnyContext
+    ) -> Doc:
+        """Call _ensure_doc to generate a Doc and set its context object."""
         doc = self._ensure_doc(doc_like)
         doc._context = context
         return doc
@@ -1219,8 +1230,9 @@ class Language:
             component_cfg = {}
         grads = {}
 
-        def get_grads(W, dW, key=None):
+        def get_grads(key, W, dW):
             grads[key] = (W, dW)
+            return W, dW
 
         get_grads.learn_rate = sgd.learn_rate  # type: ignore[attr-defined, union-attr]
         get_grads.b1 = sgd.b1  # type: ignore[attr-defined, union-attr]
@@ -1233,7 +1245,7 @@ class Language:
                 examples, sgd=get_grads, losses=losses, **component_cfg.get(name, {})
             )
         for key, (W, dW) in grads.items():
-            sgd(W, dW, key=key)  # type: ignore[call-arg, misc]
+            sgd(key, W, dW)  # type: ignore[call-arg, misc]
         return losses
 
     def begin_training(
@@ -1285,9 +1297,9 @@ class Language:
             )
         except IOError:
             raise IOError(Errors.E884.format(vectors=I["vectors"]))
-        if self.vocab.vectors.data.shape[1] >= 1:
+        if self.vocab.vectors.shape[1] >= 1:
             ops = get_current_ops()
-            self.vocab.vectors.data = ops.asarray(self.vocab.vectors.data)
+            self.vocab.vectors.to_ops(ops)
         if hasattr(self.tokenizer, "initialize"):
             tok_settings = validate_init_settings(
                 self.tokenizer.initialize,  # type: ignore[union-attr]
@@ -1332,8 +1344,8 @@ class Language:
         DOCS: https://spacy.io/api/language#resume_training
         """
         ops = get_current_ops()
-        if self.vocab.vectors.data.shape[1] >= 1:
-            self.vocab.vectors.data = ops.asarray(self.vocab.vectors.data)
+        if self.vocab.vectors.shape[1] >= 1:
+            self.vocab.vectors.to_ops(ops)
         for name, proc in self.pipeline:
             if hasattr(proc, "_rehearsal_model"):
                 proc._rehearsal_model = deepcopy(proc.model)  # type: ignore[attr-defined]
@@ -1512,7 +1524,6 @@ class Language:
 
         DOCS: https://spacy.io/api/language#pipe
         """
-        # Handle texts with context as tuples
         if as_tuples:
             texts = cast(Iterable[Tuple[Union[str, Doc], _AnyContext]], texts)
             docs_with_contexts = (
@@ -1590,8 +1601,21 @@ class Language:
         n_process: int,
         batch_size: int,
     ) -> Iterator[Doc]:
+        def prepare_input(
+            texts: Iterable[Union[str, Doc]]
+        ) -> Iterable[Tuple[Union[str, bytes], _AnyContext]]:
+            # Serialize Doc inputs to bytes to avoid incurring pickling
+            # overhead when they are passed to child processes. Also yield
+            # any context objects they might have separately (as they are not serialized).
+            for doc_like in texts:
+                if isinstance(doc_like, Doc):
+                    yield (doc_like.to_bytes(), cast(_AnyContext, doc_like._context))
+                else:
+                    yield (doc_like, cast(_AnyContext, None))
+
+        serialized_texts_with_ctx = prepare_input(texts)  # type: ignore
         # raw_texts is used later to stop iteration.
-        texts, raw_texts = itertools.tee(texts)
+        texts, raw_texts = itertools.tee(serialized_texts_with_ctx)  # type: ignore
         # for sending texts to worker
         texts_q: List[mp.Queue] = [mp.Queue() for _ in range(n_process)]
         # for receiving byte-encoded docs from worker
@@ -1611,7 +1635,13 @@ class Language:
         procs = [
             mp.Process(
                 target=_apply_pipes,
-                args=(self._ensure_doc, pipes, rch, sch, Underscore.get_state()),
+                args=(
+                    self._ensure_doc_with_context,
+                    pipes,
+                    rch,
+                    sch,
+                    Underscore.get_state(),
+                ),
             )
             for rch, sch in zip(texts_q, bytedocs_send_ch)
         ]
@@ -1624,12 +1654,12 @@ class Language:
             recv.recv() for recv in cycle(bytedocs_recv_ch)
         )
         try:
-            for i, (_, (byte_doc, byte_context, byte_error)) in enumerate(
+            for i, (_, (byte_doc, context, byte_error)) in enumerate(
                 zip(raw_texts, byte_tuples), 1
             ):
                 if byte_doc is not None:
                     doc = Doc(self.vocab).from_bytes(byte_doc)
-                    doc._context = byte_context
+                    doc._context = context
                     yield doc
                 elif byte_error is not None:
                     error = srsly.msgpack_loads(byte_error)
@@ -1664,6 +1694,7 @@ class Language:
         *,
         vocab: Union[Vocab, bool] = True,
         disable: Iterable[str] = SimpleFrozenList(),
+        enable: Iterable[str] = SimpleFrozenList(),
         exclude: Iterable[str] = SimpleFrozenList(),
         meta: Dict[str, Any] = SimpleFrozenDict(),
         auto_fill: bool = True,
@@ -1678,6 +1709,8 @@ class Language:
         disable (Iterable[str]): Names of pipeline components to disable.
             Disabled pipes will be loaded but they won't be run unless you
             explicitly enable them by calling nlp.enable_pipe.
+        enable (Iterable[str]): Names of pipeline components to enable. All other
+            pipes will be disabled (and can be enabled using `nlp.enable_pipe`).
         exclude (Iterable[str]): Names of pipeline components to exclude.
             Excluded components won't be loaded.
         meta (Dict[str, Any]): Meta overrides for nlp.meta.
@@ -1831,8 +1864,15 @@ class Language:
         # Restore the original vocab after sourcing if necessary
         if vocab_b is not None:
             nlp.vocab.from_bytes(vocab_b)
-        disabled_pipes = [*config["nlp"]["disabled"], *disable]
+
+        # Resolve disabled/enabled settings.
+        disabled_pipes = cls._resolve_component_status(
+            [*config["nlp"]["disabled"], *disable],
+            [*config["nlp"].get("enabled", []), *enable],
+            config["nlp"]["pipeline"],
+        )
         nlp._disabled = set(p for p in disabled_pipes if p not in exclude)
+
         nlp.batch_size = config["nlp"]["batch_size"]
         nlp.config = filled if auto_fill else config
         if after_pipeline_creation is not None:
@@ -1984,6 +2024,42 @@ class Language:
         serializers["vocab"] = lambda p: self.vocab.to_disk(p, exclude=exclude)
         util.to_disk(path, serializers, exclude)
 
+    @staticmethod
+    def _resolve_component_status(
+        disable: Iterable[str], enable: Iterable[str], pipe_names: Collection[str]
+    ) -> Tuple[str, ...]:
+        """Derives whether (1) `disable` and `enable` values are consistent and (2)
+        resolves those to a single set of disabled components. Raises an error in
+        case of inconsistency.
+
+        disable (Iterable[str]): Names of components or serialization fields to disable.
+        enable (Iterable[str]): Names of pipeline components to enable.
+        pipe_names (Iterable[str]): Names of all pipeline components.
+
+        RETURNS (Tuple[str, ...]): Names of components to exclude from pipeline w.r.t.
+                                   specified includes and excludes.
+        """
+
+        if disable is not None and isinstance(disable, str):
+            disable = [disable]
+        to_disable = disable
+
+        if enable:
+            to_disable = [
+                pipe_name for pipe_name in pipe_names if pipe_name not in enable
+            ]
+            if disable and disable != to_disable:
+                raise ValueError(
+                    Errors.E1042.format(
+                        arg1="enable",
+                        arg2="disable",
+                        arg1_values=enable,
+                        arg2_values=disable,
+                    )
+                )
+
+        return tuple(to_disable)
+
     def from_disk(
         self,
         path: Union[str, Path],
@@ -2156,7 +2232,7 @@ def _copy_examples(examples: Iterable[Example]) -> List[Example]:
 
 
 def _apply_pipes(
-    ensure_doc: Callable[[Union[str, Doc]], Doc],
+    ensure_doc: Callable[[Union[str, Doc, bytes], _AnyContext], Doc],
     pipes: Iterable[Callable[..., Iterator[Doc]]],
     receiver,
     sender,
@@ -2177,17 +2253,19 @@ def _apply_pipes(
     Underscore.load_state(underscore_state)
     while True:
         try:
-            texts = receiver.get()
-            docs = (ensure_doc(text) for text in texts)
+            texts_with_ctx = receiver.get()
+            docs = (
+                ensure_doc(doc_like, context) for doc_like, context in texts_with_ctx
+            )
             for pipe in pipes:
                 docs = pipe(docs)  # type: ignore[arg-type, assignment]
             # Connection does not accept unpickable objects, so send list.
             byte_docs = [(doc.to_bytes(), doc._context, None) for doc in docs]
-            padding = [(None, None, None)] * (len(texts) - len(byte_docs))
+            padding = [(None, None, None)] * (len(texts_with_ctx) - len(byte_docs))
             sender.send(byte_docs + padding)  # type: ignore[operator]
         except Exception:
             error_msg = [(None, None, srsly.msgpack_dumps(traceback.format_exc()))]
-            padding = [(None, None, None)] * (len(texts) - 1)
+            padding = [(None, None, None)] * (len(texts_with_ctx) - 1)
             sender.send(error_msg + padding)
 
 

spacy/lexeme.pyi

@@ -19,7 +19,7 @@ class Lexeme:
     @property
     def vector_norm(self) -> float: ...
     vector: Floats1d
-    rank: str
+    rank: int
     sentiment: float
     @property
     def orth_(self) -> str: ...

spacy/lexeme.pyx

@@ -130,8 +130,10 @@ cdef class Lexeme:
             return 0.0
         vector = self.vector
         xp = get_array_module(vector)
-        return (xp.dot(vector, other.vector) / (self.vector_norm * other.vector_norm))
-
+        result = xp.dot(vector, other.vector) / (self.vector_norm * other.vector_norm)
+        # ensure we get a scalar back (numpy does this automatically but cupy doesn't)
+        return result.item()
+    
     @property
     def has_vector(self):
         """RETURNS (bool): Whether a word vector is associated with the object.

spacy/lookups.py

@@ -85,7 +85,7 @@ class Table(OrderedDict):
         value: The value to set.
         """
         key = get_string_id(key)
-        OrderedDict.__setitem__(self, key, value)
+        OrderedDict.__setitem__(self, key, value)  # type:ignore[assignment]
         self.bloom.add(key)
 
     def set(self, key: Union[str, int], value: Any) -> None:
@@ -104,7 +104,7 @@ class Table(OrderedDict):
         RETURNS: The value.
         """
         key = get_string_id(key)
-        return OrderedDict.__getitem__(self, key)
+        return OrderedDict.__getitem__(self, key)  # type:ignore[index]
 
     def get(self, key: Union[str, int], default: Optional[Any] = None) -> Any:
         """Get the value for a given key. String keys will be hashed.
@@ -114,7 +114,7 @@ class Table(OrderedDict):
         RETURNS: The value.
         """
         key = get_string_id(key)
-        return OrderedDict.get(self, key, default)
+        return OrderedDict.get(self, key, default)  # type:ignore[arg-type]
 
     def __contains__(self, key: Union[str, int]) -> bool:  # type: ignore[override]
         """Check whether a key is in the table. String keys will be hashed.

spacy/matcher/dependencymatcher.pyi

@@ -0,0 +1,66 @@
+from typing import Any, Callable, Dict, List, Optional, Tuple, Union
+from .matcher import Matcher
+from ..vocab import Vocab
+from ..tokens.doc import Doc
+from ..tokens.span import Span
+
+class DependencyMatcher:
+    """Match dependency parse tree based on pattern rules."""
+
+    _patterns: Dict[str, List[Any]]
+    _raw_patterns: Dict[str, List[Any]]
+    _tokens_to_key: Dict[str, List[Any]]
+    _root: Dict[str, List[Any]]
+    _tree: Dict[str, List[Any]]
+    _callbacks: Dict[
+        Any, Callable[[DependencyMatcher, Doc, int, List[Tuple[int, List[int]]]], Any]
+    ]
+    _ops: Dict[str, Any]
+    vocab: Vocab
+    _matcher: Matcher
+    def __init__(self, vocab: Vocab, *, validate: bool = ...) -> None: ...
+    def __reduce__(
+        self,
+    ) -> Tuple[
+        Callable[
+            [Vocab, Dict[str, Any], Dict[str, Callable[..., Any]]], DependencyMatcher
+        ],
+        Tuple[
+            Vocab,
+            Dict[str, List[Any]],
+            Dict[
+                str,
+                Callable[
+                    [DependencyMatcher, Doc, int, List[Tuple[int, List[int]]]], Any
+                ],
+            ],
+        ],
+        None,
+        None,
+    ]: ...
+    def __len__(self) -> int: ...
+    def __contains__(self, key: Union[str, int]) -> bool: ...
+    def add(
+        self,
+        key: Union[str, int],
+        patterns: List[List[Dict[str, Any]]],
+        *,
+        on_match: Optional[
+            Callable[[DependencyMatcher, Doc, int, List[Tuple[int, List[int]]]], Any]
+        ] = ...
+    ) -> None: ...
+    def has_key(self, key: Union[str, int]) -> bool: ...
+    def get(
+        self, key: Union[str, int], default: Optional[Any] = ...
+    ) -> Tuple[
+        Optional[
+            Callable[[DependencyMatcher, Doc, int, List[Tuple[int, List[int]]]], Any]
+        ],
+        List[List[Dict[str, Any]]],
+    ]: ...
+    def remove(self, key: Union[str, int]) -> None: ...
+    def __call__(self, doclike: Union[Doc, Span]) -> List[Tuple[int, List[int]]]: ...
+
+def unpickle_matcher(
+    vocab: Vocab, patterns: Dict[str, Any], callbacks: Dict[str, Callable[..., Any]]
+) -> DependencyMatcher: ...

spacy/matcher/matcher.pyi

@@ -1,4 +1,6 @@
-from typing import Any, List, Dict, Tuple, Optional, Callable, Union, Iterator, Iterable
+from typing import Any, List, Dict, Tuple, Optional, Callable, Union
+from typing import Iterator, Iterable, overload
+from ..compat import Literal
 from ..vocab import Vocab
 from ..tokens import Doc, Span
 
@@ -31,12 +33,22 @@ class Matcher:
     ) -> Union[
         Iterator[Tuple[Tuple[Doc, Any], Any]], Iterator[Tuple[Doc, Any]], Iterator[Doc]
     ]: ...
+    @overload
     def __call__(
         self,
         doclike: Union[Doc, Span],
         *,
-        as_spans: bool = ...,
+        as_spans: Literal[False] = ...,
         allow_missing: bool = ...,
         with_alignments: bool = ...
-    ) -> Union[List[Tuple[int, int, int]], List[Span]]: ...
+    ) -> List[Tuple[int, int, int]]: ...
+    @overload
+    def __call__(
+        self,
+        doclike: Union[Doc, Span],
+        *,
+        as_spans: Literal[True],
+        allow_missing: bool = ...,
+        with_alignments: bool = ...
+    ) -> List[Span]: ...
     def _normalize_key(self, key: Any) -> Any: ...

spacy/matcher/matcher.pyx

@@ -18,7 +18,7 @@ from ..tokens.doc cimport Doc, get_token_attr_for_matcher
 from ..tokens.span cimport Span
 from ..tokens.token cimport Token
 from ..tokens.morphanalysis cimport MorphAnalysis
-from ..attrs cimport ID, attr_id_t, NULL_ATTR, ORTH, POS, TAG, DEP, LEMMA, MORPH
+from ..attrs cimport ID, attr_id_t, NULL_ATTR, ORTH, POS, TAG, DEP, LEMMA, MORPH, ENT_IOB
 
 from ..schemas import validate_token_pattern
 from ..errors import Errors, MatchPatternError, Warnings
@@ -86,10 +86,14 @@ cdef class Matcher:
         is a dictionary mapping attribute IDs to values, and optionally a
         quantifier operator under the key "op". The available quantifiers are:
 
-        '!': Negate the pattern, by requiring it to match exactly 0 times.
-        '?': Make the pattern optional, by allowing it to match 0 or 1 times.
-        '+': Require the pattern to match 1 or more times.
-        '*': Allow the pattern to zero or more times.
+        '!':      Negate the pattern, by requiring it to match exactly 0 times.
+        '?':      Make the pattern optional, by allowing it to match 0 or 1 times.
+        '+':      Require the pattern to match 1 or more times.
+        '*':      Allow the pattern to zero or more times.
+        '{n}':    Require the pattern to match exactly _n_ times.
+        '{n,m}':  Require the pattern to match at least _n_ but not more than _m_ times.
+        '{n,}':   Require the pattern to match at least _n_ times.
+        '{,m}':   Require the pattern to match at most _m_ times.
 
         The + and * operators return all possible matches (not just the greedy
         ones). However, the "greedy" argument can filter the final matches
@@ -244,8 +248,12 @@ cdef class Matcher:
                         pipe = "parser"
                     error_msg = Errors.E155.format(pipe=pipe, attr=self.vocab.strings.as_string(attr))
                     raise ValueError(error_msg)
-        matches = find_matches(&self.patterns[0], self.patterns.size(), doclike, length,
-                                extensions=self._extensions, predicates=self._extra_predicates, with_alignments=with_alignments)
+
+        if self.patterns.empty():
+            matches = []
+        else:
+            matches = find_matches(&self.patterns[0], self.patterns.size(), doclike, length,
+                                    extensions=self._extensions, predicates=self._extra_predicates, with_alignments=with_alignments)
         final_matches = []
         pairs_by_id = {}
         # For each key, either add all matches, or only the filtered,
@@ -686,18 +694,14 @@ cdef int8_t get_is_match(PatternStateC state,
     return True
 
 
-cdef int8_t get_is_final(PatternStateC state) nogil:
+cdef inline int8_t get_is_final(PatternStateC state) nogil:
     if state.pattern[1].quantifier == FINAL_ID:
-        id_attr = state.pattern[1].attrs[0]
-        if id_attr.attr != ID:
-            with gil:
-                raise ValueError(Errors.E074.format(attr=ID, bad_attr=id_attr.attr))
         return 1
     else:
         return 0
 
 
-cdef int8_t get_quantifier(PatternStateC state) nogil:
+cdef inline int8_t get_quantifier(PatternStateC state) nogil:
     return state.pattern.quantifier
 
 
@@ -786,6 +790,7 @@ def _preprocess_pattern(token_specs, vocab, extensions_table, extra_predicates):
 def _get_attr_values(spec, string_store):
     attr_values = []
     for attr, value in spec.items():
+        input_attr = attr
         if isinstance(attr, str):
             attr = attr.upper()
             if attr == '_':
@@ -798,7 +803,10 @@ def _get_attr_values(spec, string_store):
                 attr = "SENT_START"
             attr = IDS.get(attr)
         if isinstance(value, str):
-            value = string_store.add(value)
+            if attr == ENT_IOB and value in Token.iob_strings():
+                value = Token.iob_strings().index(value)
+            else:
+                value = string_store.add(value)
         elif isinstance(value, bool):
             value = int(value)
         elif isinstance(value, int):
@@ -811,7 +819,7 @@ def _get_attr_values(spec, string_store):
             attr_values.append((attr, value))
         else:
             # should be caught in validation
-            raise ValueError(Errors.E152.format(attr=attr))
+            raise ValueError(Errors.E152.format(attr=input_attr))
     return attr_values
 
 
@@ -1000,8 +1008,29 @@ def _get_operators(spec):
         return (ONE,)
     elif spec["OP"] in lookup:
         return lookup[spec["OP"]]
+    #Min_max {n,m}
+    elif spec["OP"].startswith("{") and spec["OP"].endswith("}"):
+        # {n}  --> {n,n}  exactly n                 ONE,(n)
+        # {n,m}--> {n,m}  min of n, max of m        ONE,(n),ZERO_ONE,(m)
+        # {,m} --> {0,m}  min of zero, max of m     ZERO_ONE,(m)
+        # {n,} --> {n,∞} min of n, max of inf       ONE,(n),ZERO_PLUS
+
+        min_max = spec["OP"][1:-1]
+        min_max = min_max if "," in min_max else f"{min_max},{min_max}"
+        n, m = min_max.split(",")
+
+        #1. Either n or m is a blank string and the other is numeric -->isdigit
+        #2. Both are numeric and n <= m
+        if (not n.isdecimal() and not m.isdecimal()) or (n.isdecimal() and m.isdecimal() and int(n) > int(m)):
+            keys = ", ".join(lookup.keys()) + ", {n}, {n,m}, {n,}, {,m} where n and m are integers and n <= m "
+            raise ValueError(Errors.E011.format(op=spec["OP"], opts=keys))
+
+        # if n is empty string, zero would be used
+        head = tuple(ONE for __ in range(int(n or 0)))
+        tail = tuple(ZERO_ONE for __ in range(int(m) - int(n or 0))) if m else (ZERO_PLUS,)
+        return head + tail
     else:
-        keys = ", ".join(lookup.keys())
+        keys = ", ".join(lookup.keys()) + ", {n}, {n,m}, {n,}, {,m} where n and m are integers and n <= m "
         raise ValueError(Errors.E011.format(op=spec["OP"], opts=keys))
 
 

spacy/matcher/phrasematcher.pyi

@@ -1,6 +1,6 @@
-from typing import List, Tuple, Union, Optional, Callable, Any, Dict
-
-from . import Matcher
+from typing import List, Tuple, Union, Optional, Callable, Any, Dict, overload
+from ..compat import Literal
+from .matcher import Matcher
 from ..vocab import Vocab
 from ..tokens import Doc, Span
 
@@ -14,16 +14,24 @@ class PhraseMatcher:
     def add(
         self,
         key: str,
-        docs: List[List[Dict[str, Any]]],
+        docs: List[Doc],
         *,
         on_match: Optional[
             Callable[[Matcher, Doc, int, List[Tuple[Any, ...]]], Any]
         ] = ...,
     ) -> None: ...
     def remove(self, key: str) -> None: ...
+    @overload
     def __call__(
         self,
         doclike: Union[Doc, Span],
         *,
-        as_spans: bool = ...,
-    ) -> Union[List[Tuple[int, int, int]], List[Span]]: ...
+        as_spans: Literal[False] = ...,
+    ) -> List[Tuple[int, int, int]]: ...
+    @overload
+    def __call__(
+        self,
+        doclike: Union[Doc, Span],
+        *,
+        as_spans: Literal[True],
+    ) -> List[Span]: ...

spacy/matcher/phrasematcher.pyx

@@ -118,6 +118,8 @@ cdef class PhraseMatcher:
                     # if token is not found, break out of the loop
                     current_node = NULL
                     break
+            path_nodes.push_back(current_node)
+            path_keys.push_back(self._terminal_hash)
             # remove the tokens from trie node if there are no other
             # keywords with them
             result = map_get(current_node, self._terminal_hash)

spacy/ml/_precomputable_affine.py

@@ -22,9 +22,11 @@ def forward(model, X, is_train):
     nP = model.get_dim("nP")
     nI = model.get_dim("nI")
     W = model.get_param("W")
-    Yf = model.ops.gemm(X, W.reshape((nF * nO * nP, nI)), trans2=True)
+    # Preallocate array for layer output, including padding.
+    Yf = model.ops.alloc2f(X.shape[0] + 1, nF * nO * nP, zeros=False)
+    model.ops.gemm(X, W.reshape((nF * nO * nP, nI)), trans2=True, out=Yf[1:])
     Yf = Yf.reshape((Yf.shape[0], nF, nO, nP))
-    Yf = model.ops.xp.vstack((model.get_param("pad"), Yf))
+    Yf[0] = model.get_param("pad")
 
     def backward(dY_ids):
         # This backprop is particularly tricky, because we get back a different

spacy/ml/callbacks.py

@@ -1,9 +1,14 @@
-from functools import partial
-from typing import Type, Callable, TYPE_CHECKING
+from typing import Type, Callable, Dict, TYPE_CHECKING, List, Optional, Set
+import functools
+import inspect
+import types
+import warnings
 
 from thinc.layers import with_nvtx_range
 from thinc.model import Model, wrap_model_recursive
+from thinc.util import use_nvtx_range
 
+from ..errors import Warnings
 from ..util import registry
 
 if TYPE_CHECKING:
@@ -11,29 +16,106 @@ if TYPE_CHECKING:
     from ..language import Language  # noqa: F401
 
 
-@registry.callbacks("spacy.models_with_nvtx_range.v1")
-def create_models_with_nvtx_range(
-    forward_color: int = -1, backprop_color: int = -1
-) -> Callable[["Language"], "Language"]:
-    def models_with_nvtx_range(nlp):
-        pipes = [
-            pipe
-            for _, pipe in nlp.components
-            if hasattr(pipe, "is_trainable") and pipe.is_trainable
-        ]
+DEFAULT_NVTX_ANNOTATABLE_PIPE_METHODS = [
+    "pipe",
+    "predict",
+    "set_annotations",
+    "update",
+    "rehearse",
+    "get_loss",
+    "initialize",
+    "begin_update",
+    "finish_update",
+    "update",
+]
 
-        # We need process all models jointly to avoid wrapping callbacks twice.
-        models = Model(
-            "wrap_with_nvtx_range",
-            forward=lambda model, X, is_train: ...,
-            layers=[pipe.model for pipe in pipes],
-        )
 
-        for node in models.walk():
+def models_with_nvtx_range(nlp, forward_color: int, backprop_color: int):
+    pipes = [
+        pipe
+        for _, pipe in nlp.components
+        if hasattr(pipe, "is_trainable") and pipe.is_trainable
+    ]
+
+    seen_models: Set[int] = set()
+    for pipe in pipes:
+        for node in pipe.model.walk():
+            if id(node) in seen_models:
+                continue
+            seen_models.add(id(node))
             with_nvtx_range(
                 node, forward_color=forward_color, backprop_color=backprop_color
             )
 
+    return nlp
+
+
+@registry.callbacks("spacy.models_with_nvtx_range.v1")
+def create_models_with_nvtx_range(
+    forward_color: int = -1, backprop_color: int = -1
+) -> Callable[["Language"], "Language"]:
+    return functools.partial(
+        models_with_nvtx_range,
+        forward_color=forward_color,
+        backprop_color=backprop_color,
+    )
+
+
+def nvtx_range_wrapper_for_pipe_method(self, func, *args, **kwargs):
+    if isinstance(func, functools.partial):
+        return func(*args, **kwargs)
+    else:
+        with use_nvtx_range(f"{self.name} {func.__name__}"):
+            return func(*args, **kwargs)
+
+
+def pipes_with_nvtx_range(
+    nlp, additional_pipe_functions: Optional[Dict[str, List[str]]]
+):
+    for _, pipe in nlp.components:
+        if additional_pipe_functions:
+            extra_funcs = additional_pipe_functions.get(pipe.name, [])
+        else:
+            extra_funcs = []
+
+        for name in DEFAULT_NVTX_ANNOTATABLE_PIPE_METHODS + extra_funcs:
+            func = getattr(pipe, name, None)
+            if func is None:
+                if name in extra_funcs:
+                    warnings.warn(Warnings.W121.format(method=name, pipe=pipe.name))
+                continue
+
+            wrapped_func = functools.partial(
+                types.MethodType(nvtx_range_wrapper_for_pipe_method, pipe), func
+            )
+
+            # Try to preserve the original function signature.
+            try:
+                wrapped_func.__signature__ = inspect.signature(func)  # type: ignore
+            except:
+                pass
+
+            try:
+                setattr(
+                    pipe,
+                    name,
+                    wrapped_func,
+                )
+            except AttributeError:
+                warnings.warn(Warnings.W122.format(method=name, pipe=pipe.name))
+
+    return nlp
+
+
+@registry.callbacks("spacy.models_and_pipes_with_nvtx_range.v1")
+def create_models_and_pipes_with_nvtx_range(
+    forward_color: int = -1,
+    backprop_color: int = -1,
+    additional_pipe_functions: Optional[Dict[str, List[str]]] = None,
+) -> Callable[["Language"], "Language"]:
+    def inner(nlp):
+        nlp = models_with_nvtx_range(nlp, forward_color, backprop_color)
+        nlp = pipes_with_nvtx_range(nlp, additional_pipe_functions)
         return nlp
 
-    return models_with_nvtx_range
+    return inner

spacy/ml/extract_spans.py

@@ -63,4 +63,4 @@ def _get_span_indices(ops, spans: Ragged, lengths: Ints1d) -> Ints1d:
 
 
 def _ensure_cpu(spans: Ragged, lengths: Ints1d) -> Tuple[Ragged, Ints1d]:
-    return (Ragged(to_numpy(spans.dataXd), to_numpy(spans.lengths)), to_numpy(lengths))
+    return Ragged(to_numpy(spans.dataXd), to_numpy(spans.lengths)), to_numpy(lengths)

spacy/ml/models/__init__.py

@@ -5,3 +5,10 @@ from .spancat import *  # noqa
 from .tagger import *  # noqa
 from .textcat import *  # noqa
 from .tok2vec import *  # noqa
+
+# some models require Torch
+from thinc.util import has_torch
+if has_torch:
+    from .coref import * #noqa
+    from .span_predictor import * #noqa
+

spacy/ml/models/coref.py

@@ -0,0 +1,383 @@
+from typing import List, Tuple, Callable, cast
+
+from thinc.api import Model, chain, get_width
+from thinc.api import PyTorchWrapper, ArgsKwargs
+from thinc.types import Floats2d, Ints2d
+from thinc.util import torch, xp2torch, torch2xp
+
+from ...tokens import Doc
+from ...util import registry
+
+
+EPSILON = 1e-7
+
+
+@registry.architectures("spacy.Coref.v1")
+def build_wl_coref_model(
+    tok2vec: Model[List[Doc], List[Floats2d]],
+    distance_embedding_size: int = 20,
+    hidden_size: int = 1024,
+    depth: int = 1,
+    dropout: float = 0.3,
+    # pairs to keep per mention after rough scoring
+    antecedent_limit: int = 50,
+    antecedent_batch_size: int = 512,
+    nI=None,
+) -> Model[List[Doc], Tuple[Floats2d, Ints2d]]:
+
+    with Model.define_operators({">>": chain}):
+        coref_clusterer: Model[List[Floats2d], Tuple[Floats2d, Ints2d]] = Model(
+            "coref_clusterer",
+            forward=coref_forward,
+            init=coref_init,
+            dims={"nI": nI},
+            attrs={
+                "distance_embedding_size": distance_embedding_size,
+                "hidden_size": hidden_size,
+                "depth": depth,
+                "dropout": dropout,
+                "antecedent_limit": antecedent_limit,
+                "antecedent_batch_size": antecedent_batch_size,
+            },
+        )
+
+        model = tok2vec >> coref_clusterer
+        model.set_ref("coref_clusterer", coref_clusterer)
+    return model
+
+
+def coref_init(model: Model, X=None, Y=None):
+    if model.layers:
+        return
+
+    if X is not None and model.has_dim("nI") is None:
+        model.set_dim("nI", get_width(X))
+
+    hidden_size = model.attrs["hidden_size"]
+    depth = model.attrs["depth"]
+    dropout = model.attrs["dropout"]
+    antecedent_limit = model.attrs["antecedent_limit"]
+    antecedent_batch_size = model.attrs["antecedent_batch_size"]
+    distance_embedding_size = model.attrs["distance_embedding_size"]
+
+    model._layers = [
+        PyTorchWrapper(
+            CorefClusterer(
+                model.get_dim("nI"),
+                distance_embedding_size,
+                hidden_size,
+                depth,
+                dropout,
+                antecedent_limit,
+                antecedent_batch_size,
+            ),
+            convert_inputs=convert_coref_clusterer_inputs,
+            convert_outputs=convert_coref_clusterer_outputs,
+        )
+        # TODO maybe we need mixed precision and grad scaling?
+    ]
+
+
+def coref_forward(model: Model, X, is_train: bool):
+    return model.layers[0](X, is_train)
+
+def convert_coref_clusterer_inputs(model: Model, X: List[Floats2d], is_train: bool):
+    # The input here is List[Floats2d], one for each doc
+    # just use the first
+    # TODO real batching
+    X = X[0]
+    word_features = xp2torch(X, requires_grad=is_train)
+
+    def backprop(args: ArgsKwargs) -> List[Floats2d]:
+        # convert to xp and wrap in list
+        gradients = cast(Floats2d, torch2xp(args.args[0]))
+        return [gradients]
+
+    return ArgsKwargs(args=(word_features,), kwargs={}), backprop
+
+
+def convert_coref_clusterer_outputs(
+    model: Model, inputs_outputs, is_train: bool
+) -> Tuple[Tuple[Floats2d, Ints2d], Callable]:
+    _, outputs = inputs_outputs
+    scores, indices = outputs
+
+    def convert_for_torch_backward(dY: Floats2d) -> ArgsKwargs:
+        dY_t = xp2torch(dY[0])
+        return ArgsKwargs(
+            args=([scores],),
+            kwargs={"grad_tensors": [dY_t]},
+        )
+
+    scores_xp = cast(Floats2d, torch2xp(scores))
+    indices_xp = cast(Ints2d, torch2xp(indices))
+    return (scores_xp, indices_xp), convert_for_torch_backward
+
+
+class CorefClusterer(torch.nn.Module):
+    """
+    Combines all coref modules together to find coreferent token pairs.
+    Submodules (in the order of their usage in the pipeline):
+        - rough_scorer (RoughScorer) that prunes candidate pairs
+        - pw (DistancePairwiseEncoder) that computes pairwise features
+        - a_scorer (AnaphoricityScorer) produces the final scores
+    """
+
+    def __init__(
+        self,
+        dim: int,
+        dist_emb_size: int,
+        hidden_size: int,
+        n_layers: int,
+        dropout: float,
+        roughk: int,
+        batch_size: int,
+    ):
+        super().__init__()
+        """
+        dim: Size of the input features.
+        dist_emb_size: Size of the distance embeddings.
+        hidden_size: Size of the coreference candidate embeddings.
+        n_layers: Numbers of layers in the AnaphoricityScorer.
+        dropout: Dropout probability to apply across all modules.
+        roughk: Number of candidates the RoughScorer returns.
+        batch_size: Internal batch-size for the more expensive scorer.
+        """
+        self.dropout = torch.nn.Dropout(dropout)
+        self.batch_size = batch_size
+        self.pw = DistancePairwiseEncoder(dist_emb_size, dropout)
+
+        pair_emb = dim * 3 + self.pw.shape
+        self.a_scorer = AnaphoricityScorer(pair_emb, hidden_size, n_layers, dropout)
+        self.lstm = torch.nn.LSTM(
+            input_size=dim,
+            hidden_size=dim,
+            batch_first=True,
+        )
+
+        self.rough_scorer = RoughScorer(dim, dropout, roughk)
+
+    def forward(self, word_features: torch.Tensor) -> Tuple[torch.Tensor, torch.Tensor]:
+        """
+        1. LSTM encodes the incoming word_features.
+        2. The RoughScorer scores and prunes the candidates.
+        3. The DistancePairwiseEncoder embeds the distances between pairs.
+        4. The AnaphoricityScorer scores all pairs in mini-batches.
+
+        word_features: torch.Tensor containing word encodings
+
+        returns:
+            coref_scores: n_words x roughk floats.
+            top_indices: n_words x roughk integers.
+        """
+        self.lstm.flatten_parameters()  # XXX without this there's a warning
+        word_features = torch.unsqueeze(word_features, dim=0)
+        words, _ = self.lstm(word_features)
+        words = words.squeeze()
+        # words: n_words x dim
+        words = self.dropout(words)
+        # Obtain bilinear scores and leave only top-k antecedents for each word
+        # top_rough_scores: (n_words x roughk)
+        # top_indices: (n_words x roughk)
+        top_rough_scores, top_indices = self.rough_scorer(words)
+        # Get pairwise features
+        # (n_words x roughk x n_pw_features)
+        pw = self.pw(top_indices)
+        batch_size = self.batch_size
+        a_scores_lst: List[torch.Tensor] = []
+
+        for i in range(0, len(words), batch_size):
+            pw_batch = pw[i : i + batch_size]
+            words_batch = words[i : i + batch_size]
+            top_indices_batch = top_indices[i : i + batch_size]
+            top_rough_scores_batch = top_rough_scores[i : i + batch_size]
+
+            # a_scores_batch    [batch_size, n_ants]
+            a_scores_batch = self.a_scorer(
+                all_mentions=words,
+                mentions_batch=words_batch,
+                pw_batch=pw_batch,
+                top_indices_batch=top_indices_batch,
+                top_rough_scores_batch=top_rough_scores_batch,
+            )
+            a_scores_lst.append(a_scores_batch)
+
+        coref_scores = torch.cat(a_scores_lst, dim=0)
+        return coref_scores, top_indices
+
+
+# Note this function is kept here to keep a torch dep out of coref_util.
+def add_dummy(tensor: torch.Tensor, eps: bool = False):
+    """Prepends zeros (or a very small value if eps is True)
+    to the first (not zeroth) dimension of tensor.
+    """
+    kwargs = dict(device=tensor.device, dtype=tensor.dtype)
+    shape: List[int] = list(tensor.shape)
+    shape[1] = 1
+    if not eps:
+        dummy = torch.zeros(shape, **kwargs)  # type: ignore
+    else:
+        dummy = torch.full(shape, EPSILON, **kwargs)  # type: ignore
+    output = torch.cat((dummy, tensor), dim=1)
+    return output
+
+
+class AnaphoricityScorer(torch.nn.Module):
+    """Calculates anaphoricity scores by passing the inputs into a FFNN"""
+
+    def __init__(self, in_features: int, hidden_size, depth, dropout):
+        super().__init__()
+        hidden_size = hidden_size
+        if not depth:
+            hidden_size = in_features
+        layers = []
+        for i in range(depth):
+            layers.extend(
+                [
+                    torch.nn.Linear(hidden_size if i else in_features, hidden_size),
+                    torch.nn.LeakyReLU(),
+                    torch.nn.Dropout(dropout),
+                ]
+            )
+        self.hidden = torch.nn.Sequential(*layers)
+        self.out = torch.nn.Linear(hidden_size, out_features=1)
+
+    def forward(
+        self,
+        *,  # type: ignore  # pylint: disable=arguments-differ  #35566 in pytorch
+        all_mentions: torch.Tensor,
+        mentions_batch: torch.Tensor,
+        pw_batch: torch.Tensor,
+        top_indices_batch: torch.Tensor,
+        top_rough_scores_batch: torch.Tensor,
+    ) -> torch.Tensor:
+        """Builds a pairwise matrix, scores the pairs and returns the scores.
+
+        Args:
+            all_mentions (torch.Tensor): [n_mentions, mention_emb]
+            mentions_batch (torch.Tensor): [batch_size, mention_emb]
+            pw_batch (torch.Tensor): [batch_size, n_ants, pw_emb]
+            top_indices_batch (torch.Tensor): [batch_size, n_ants]
+            top_rough_scores_batch (torch.Tensor): [batch_size, n_ants]
+
+        Returns:
+            torch.Tensor [batch_size, n_ants + 1]
+                anaphoricity scores for the pairs + a dummy column
+        """
+        # [batch_size, n_ants, pair_emb]
+        pair_matrix = self._get_pair_matrix(
+            all_mentions, mentions_batch, pw_batch, top_indices_batch
+        )
+
+        # [batch_size, n_ants]
+        scores = top_rough_scores_batch + self._ffnn(pair_matrix)
+        scores = add_dummy(scores, eps=True)
+
+        return scores
+
+    def _ffnn(self, x: torch.Tensor) -> torch.Tensor:
+        """
+        x: tensor of shape (batch_size x roughk x n_features
+        returns: tensor of shape (batch_size x antecedent_limit)
+        """
+        x = self.out(self.hidden(x))
+        return x.squeeze(2)
+
+    @staticmethod
+    def _get_pair_matrix(
+        all_mentions: torch.Tensor,
+        mentions_batch: torch.Tensor,
+        pw_batch: torch.Tensor,
+        top_indices_batch: torch.Tensor,
+    ) -> torch.Tensor:
+        """
+        Builds the matrix used as input for AnaphoricityScorer.
+
+        all_mentions: (n_mentions x mention_emb),
+            all the valid mentions of the document,
+            can be on a different device
+        mentions_batch: (batch_size x mention_emb),
+            the mentions of the current batch.
+        pw_batch: (batch_size x roughk x pw_emb),
+            pairwise distance features of the current batch.
+        top_indices_batch: (batch_size x n_ants),
+            indices of antecedents of each mention
+
+        Returns:
+            out: pairwise features (batch_size x n_ants x pair_emb)
+        """
+        emb_size = mentions_batch.shape[1]
+        n_ants = pw_batch.shape[1]
+
+        a_mentions = mentions_batch.unsqueeze(1).expand(-1, n_ants, emb_size)
+        b_mentions = all_mentions[top_indices_batch]
+        similarity = a_mentions * b_mentions
+
+        out = torch.cat((a_mentions, b_mentions, similarity, pw_batch), dim=2)
+        return out
+
+
+class RoughScorer(torch.nn.Module):
+    """
+    Cheaper module that gives a rough estimate of the anaphoricity of two
+    candidates, only top scoring candidates are considered on later
+    steps to reduce computational cost.
+    """
+
+    def __init__(self, features: int, dropout: float, antecedent_limit: int):
+        super().__init__()
+        self.dropout = torch.nn.Dropout(dropout)
+        self.bilinear = torch.nn.Linear(features, features)
+        self.k = antecedent_limit
+
+    def forward(
+        self,  # type: ignore
+        mentions: torch.Tensor,
+    ) -> Tuple[torch.Tensor, torch.Tensor]:
+        """
+        Returns rough anaphoricity scores for candidates, which consist of
+        the bilinear output of the current model summed with mention scores.
+        """
+        # [n_mentions, n_mentions]
+        pair_mask = torch.arange(mentions.shape[0])
+        pair_mask = pair_mask.unsqueeze(1) - pair_mask.unsqueeze(0)
+        pair_mask = torch.log((pair_mask > 0).to(torch.float))
+        pair_mask = pair_mask.to(mentions.device)
+        bilinear_scores = self.dropout(self.bilinear(mentions)).mm(mentions.T)
+        rough_scores = pair_mask + bilinear_scores
+        top_scores, indices = torch.topk(
+            rough_scores, k=min(self.k, len(rough_scores)), dim=1, sorted=False
+        )
+
+        return top_scores, indices
+
+
+class DistancePairwiseEncoder(torch.nn.Module):
+    def __init__(self, distance_embedding_size, dropout):
+        """
+        Takes the top_indices indicating, which is a ranked
+        list for each word and its most likely corresponding
+        anaphora candidates. For each of these pairs it looks
+        up a distance embedding from a table, where the distance
+        corresponds to the log-distance.
+
+        distance_embedding_size: int,
+            Dimensionality of the distance-embeddings table.
+        dropout: float,
+            Dropout probability.
+        """
+        super().__init__()
+        emb_size = distance_embedding_size
+        self.distance_emb = torch.nn.Embedding(9, emb_size)
+        self.dropout = torch.nn.Dropout(dropout)
+        self.shape = emb_size
+
+    def forward(self, top_indices: torch.Tensor) -> torch.Tensor:
+        word_ids = torch.arange(0, top_indices.size(0))
+        distance = (word_ids.unsqueeze(1) - word_ids[top_indices]).clamp_min_(min=1)
+        log_distance = distance.to(torch.float).log2().floor_()
+        log_distance = log_distance.clamp_max_(max=6).to(torch.long)
+        distance = torch.where(distance < 5, distance - 1, log_distance + 2)
+        distance = distance.to(top_indices.device)
+        distance = self.distance_emb(distance)
+        return self.dropout(distance)

spacy/ml/models/coref_util.py

@@ -0,0 +1,207 @@
+from typing import List, Tuple, Set, Dict, cast
+from thinc.types import Ints2d
+from spacy.tokens import Doc
+
+# type alias to make writing this less tedious
+MentionClusters = List[List[Tuple[int, int]]]
+
+DEFAULT_CLUSTER_PREFIX = "coref_clusters"
+
+
+class GraphNode:
+    def __init__(self, node_id: int):
+        self.id = node_id
+        self.links: Set[GraphNode] = set()
+        self.visited = False
+
+    def link(self, another: "GraphNode"):
+        self.links.add(another)
+        another.links.add(self)
+
+    def __repr__(self) -> str:
+        return str(self.id)
+
+
+def get_sentence_ids(doc):
+    out = []
+    sent_id = -1
+    for tok in doc:
+        if tok.is_sent_start:
+            sent_id += 1
+        out.append(sent_id)
+    return out
+
+
+# from model.py, refactored to be non-member
+def get_predicted_antecedents(xp, antecedent_idx, antecedent_scores):
+    """Get the ID of the antecedent for each span. -1 if no antecedent."""
+    predicted_antecedents = []
+    for i, idx in enumerate(xp.argmax(antecedent_scores, axis=1) - 1):
+        if idx < 0:
+            predicted_antecedents.append(-1)
+        else:
+            predicted_antecedents.append(antecedent_idx[i][idx])
+    return predicted_antecedents
+
+
+# from model.py, refactored to be non-member
+def get_predicted_clusters(
+    xp, span_starts, span_ends, antecedent_idx, antecedent_scores
+):
+    """Convert predictions to usable cluster data.
+
+    return values:
+
+    clusters: a list of spans (i, j) that are a cluster
+
+    Note that not all spans will be in the final output; spans with no
+    antecedent or referrent are omitted from clusters and mention2cluster.
+    """
+    # Get predicted antecedents
+    predicted_antecedents = get_predicted_antecedents(
+        xp, antecedent_idx, antecedent_scores
+    )
+
+    # Get predicted clusters
+    mention_to_cluster_id = {}
+    predicted_clusters = []
+    for i, predicted_idx in enumerate(predicted_antecedents):
+        if predicted_idx < 0:
+            continue
+        assert i > predicted_idx, f"span idx: {i}; antecedent idx: {predicted_idx}"
+        # Check antecedent's cluster
+        antecedent = (int(span_starts[predicted_idx]), int(span_ends[predicted_idx]))
+        antecedent_cluster_id = mention_to_cluster_id.get(antecedent, -1)
+        if antecedent_cluster_id == -1:
+            antecedent_cluster_id = len(predicted_clusters)
+            predicted_clusters.append([antecedent])
+            mention_to_cluster_id[antecedent] = antecedent_cluster_id
+        # Add mention to cluster
+        mention = (int(span_starts[i]), int(span_ends[i]))
+        predicted_clusters[antecedent_cluster_id].append(mention)
+        mention_to_cluster_id[mention] = antecedent_cluster_id
+
+    predicted_clusters = [tuple(c) for c in predicted_clusters]
+    return predicted_clusters
+
+
+def select_non_crossing_spans(
+    idxs: List[int], starts: List[int], ends: List[int], limit: int
+) -> List[int]:
+    """Given a list of spans sorted in descending order, return the indexes of
+    spans to keep, discarding spans that cross.
+
+    Nested spans are allowed.
+    """
+    # ported from Model._extract_top_spans
+    selected: List[int] = []
+    start_to_max_end: Dict[int, int] = {}
+    end_to_min_start: Dict[int, int] = {}
+
+    for idx in idxs:
+        if len(selected) >= limit or idx > len(starts):
+            break
+
+        start, end = starts[idx], ends[idx]
+        cross = False
+
+        for ti in range(start, end):
+            max_end = start_to_max_end.get(ti, -1)
+            if ti > start and max_end > end:
+                cross = True
+                break
+
+            min_start = end_to_min_start.get(ti, -1)
+            if ti < end and 0 <= min_start < start:
+                cross = True
+                break
+
+        if not cross:
+            # this index will be kept
+            # record it so we can exclude anything that crosses it
+            selected.append(idx)
+            max_end = start_to_max_end.get(start, -1)
+            if end > max_end:
+                start_to_max_end[start] = end
+            min_start = end_to_min_start.get(end, -1)
+            if min_start == -1 or start < min_start:
+                end_to_min_start[end] = start
+
+    # sort idxs by order in doc
+    selected = sorted(selected, key=lambda idx: (starts[idx], ends[idx]))
+    # This was causing many repetitive entities in the output - removed for now
+    # while len(selected) < limit:
+    #     selected.append(selected[0])  # this seems a bit weird?
+    return selected
+
+
+def create_head_span_idxs(ops, doclen: int):
+    """Helper function to create single-token span indices."""
+    aa = ops.xp.arange(0, doclen)
+    bb = ops.xp.arange(0, doclen) + 1
+    return ops.asarray2i([aa, bb]).T
+
+
+def get_clusters_from_doc(doc) -> List[List[Tuple[int, int]]]:
+    """Convert the span clusters in a Doc to simple integer tuple lists. The
+    ints are char spans, to be tokenization independent.
+    """
+    out = []
+    keys = sorted(list(doc.spans.keys()))
+    for key in keys:
+        val = doc.spans[key]
+        cluster = []
+        for span in val:
+
+            head_i = span.root.i
+            head = doc[head_i]
+            char_span = (head.idx, head.idx + len(head))
+            cluster.append(char_span)
+
+        # don't want duplicates
+        cluster = list(set(cluster))
+        out.append(cluster)
+    return out
+
+
+def create_gold_scores(
+    ments: Ints2d, clusters: List[List[Tuple[int, int]]]
+) -> List[List[bool]]:
+    """Given mentions considered for antecedents and gold clusters,
+    construct a gold score matrix. This does not include the placeholder.
+
+    In the gold matrix, the value of a true antecedent is True, and otherwise
+    it is False. These will be converted to 1/0 values later.
+    """
+    # make a mapping of mentions to cluster id
+    # id is not important but equality will be
+    ment2cid: Dict[Tuple[int, int], int] = {}
+    for cid, cluster in enumerate(clusters):
+        for ment in cluster:
+            ment2cid[ment] = cid
+
+    ll = len(ments)
+    out = []
+    # The .tolist() call is necessary with cupy but not numpy
+    mentuples = [cast(Tuple[int, int], tuple(mm.tolist())) for mm in ments]
+    for ii, ment in enumerate(mentuples):
+        if ment not in ment2cid:
+            # this is not in a cluster so it has no antecedent
+            out.append([False] * ll)
+            continue
+
+        # this might change if no real antecedent is a candidate
+        row = []
+        cid = ment2cid[ment]
+        for jj, ante in enumerate(mentuples):
+            # antecedents must come first
+            if jj >= ii:
+                row.append(False)
+                continue
+
+            row.append(cid == ment2cid.get(ante, -1))
+
+        out.append(row)
+
+    # caller needs to convert to array, and add placeholder
+    return out

spacy/ml/models/entity_linker.py

@@ -1,34 +1,82 @@
 from pathlib import Path
-from typing import Optional, Callable, Iterable, List
+from typing import Optional, Callable, Iterable, List, Tuple
 from thinc.types import Floats2d
 from thinc.api import chain, clone, list2ragged, reduce_mean, residual
-from thinc.api import Model, Maxout, Linear
+from thinc.api import Model, Maxout, Linear, noop, tuplify, Ragged
 
 from ...util import registry
 from ...kb import KnowledgeBase, Candidate, get_candidates
 from ...vocab import Vocab
 from ...tokens import Span, Doc
+from ..extract_spans import extract_spans
+from ...errors import Errors
 
 
-@registry.architectures("spacy.EntityLinker.v1")
+@registry.architectures("spacy.EntityLinker.v2")
 def build_nel_encoder(
     tok2vec: Model, nO: Optional[int] = None
 ) -> Model[List[Doc], Floats2d]:
-    with Model.define_operators({">>": chain, "**": clone}):
+    with Model.define_operators({">>": chain, "&": tuplify}):
         token_width = tok2vec.maybe_get_dim("nO")
         output_layer = Linear(nO=nO, nI=token_width)
         model = (
-            tok2vec
-            >> list2ragged()
+            ((tok2vec >> list2ragged()) & build_span_maker())
+            >> extract_spans()
             >> reduce_mean()
-            >> residual(Maxout(nO=token_width, nI=token_width, nP=2, dropout=0.0))  # type: ignore[arg-type]
+            >> residual(Maxout(nO=token_width, nI=token_width, nP=2, dropout=0.0))  # type: ignore
             >> output_layer
         )
         model.set_ref("output_layer", output_layer)
         model.set_ref("tok2vec", tok2vec)
+    # flag to show this isn't legacy
+    model.attrs["include_span_maker"] = True
     return model
 
 
+def build_span_maker(n_sents: int = 0) -> Model:
+    model: Model = Model("span_maker", forward=span_maker_forward)
+    model.attrs["n_sents"] = n_sents
+    return model
+
+
+def span_maker_forward(model, docs: List[Doc], is_train) -> Tuple[Ragged, Callable]:
+    ops = model.ops
+    n_sents = model.attrs["n_sents"]
+    candidates = []
+    for doc in docs:
+        cands = []
+        try:
+            sentences = [s for s in doc.sents]
+        except ValueError:
+            # no sentence info, normal in initialization
+            for tok in doc:
+                tok.is_sent_start = tok.i == 0
+            sentences = [doc[:]]
+        for ent in doc.ents:
+            try:
+                # find the sentence in the list of sentences.
+                sent_index = sentences.index(ent.sent)
+            except AttributeError:
+                # Catch the exception when ent.sent is None and provide a user-friendly warning
+                raise RuntimeError(Errors.E030) from None
+            # get n previous sentences, if there are any
+            start_sentence = max(0, sent_index - n_sents)
+            # get n posterior sentences, or as many < n as there are
+            end_sentence = min(len(sentences) - 1, sent_index + n_sents)
+            # get token positions
+            start_token = sentences[start_sentence].start
+            end_token = sentences[end_sentence].end
+            # save positions for extraction
+            cands.append((start_token, end_token))
+
+        candidates.append(ops.asarray2i(cands))
+    candlens = ops.asarray1i([len(cands) for cands in candidates])
+    candidates = ops.xp.concatenate(candidates)
+    outputs = Ragged(candidates, candlens)
+    # because this is just rearranging docs, the backprop does nothing
+    return outputs, lambda x: []
+
+
 @registry.misc("spacy.KBFromFile.v1")
 def load_kb(kb_path: Path) -> Callable[[Vocab], KnowledgeBase]:
     def kb_from_file(vocab):

spacy/ml/models/multi_task.py

@@ -23,7 +23,7 @@ def create_pretrain_vectors(
     maxout_pieces: int, hidden_size: int, loss: str
 ) -> Callable[["Vocab", Model], Model]:
     def create_vectors_objective(vocab: "Vocab", tok2vec: Model) -> Model:
-        if vocab.vectors.data.shape[1] == 0:
+        if vocab.vectors.shape[1] == 0:
             raise ValueError(Errors.E875)
         model = build_cloze_multi_task_model(
             vocab, tok2vec, hidden_size=hidden_size, maxout_pieces=maxout_pieces
@@ -85,7 +85,7 @@ def get_characters_loss(ops, docs, prediction, nr_char):
     target = ops.asarray(to_categorical(target_ids, n_classes=256), dtype="f")
     target = target.reshape((-1, 256 * nr_char))
     diff = prediction - target
-    loss = (diff ** 2).sum()
+    loss = (diff**2).sum()
     d_target = diff / float(prediction.shape[0])
     return loss, d_target
 
@@ -116,7 +116,7 @@ def build_multi_task_model(
 def build_cloze_multi_task_model(
     vocab: "Vocab", tok2vec: Model, maxout_pieces: int, hidden_size: int
 ) -> Model:
-    nO = vocab.vectors.data.shape[1]
+    nO = vocab.vectors.shape[1]
     output_layer = chain(
         cast(Model[List["Floats2d"], Floats2d], list2array()),
         Maxout(

spacy/ml/models/parser.py

@@ -72,7 +72,7 @@ def build_tb_parser_model(
     t2v_width = tok2vec.get_dim("nO") if tok2vec.has_dim("nO") else None
     tok2vec = chain(
         tok2vec,
-        cast(Model[List["Floats2d"], Floats2d], list2array()),
+        list2array(),
         Linear(hidden_width, t2v_width),
     )
     tok2vec.set_dim("nO", hidden_width)

spacy/ml/models/span_predictor.py

@@ -0,0 +1,245 @@
+from typing import List, Tuple, cast
+
+from thinc.api import Model, chain, tuplify, get_width
+from thinc.api import PyTorchWrapper, ArgsKwargs
+from thinc.types import Floats2d, Ints1d
+from thinc.util import torch, xp2torch, torch2xp
+
+from ...tokens import Doc
+from ...util import registry
+from .coref_util import get_sentence_ids
+
+
+@registry.architectures("spacy.SpanPredictor.v1")
+def build_span_predictor(
+    tok2vec: Model[List[Doc], List[Floats2d]],
+    hidden_size: int = 1024,
+    distance_embedding_size: int = 64,
+    conv_channels: int = 4,
+    window_size: int = 1,
+    max_distance: int = 128,
+    prefix: str = "coref_head_clusters",
+):
+    # TODO add model return types
+
+    nI = None
+
+    with Model.define_operators({">>": chain, "&": tuplify}):
+        span_predictor: Model[List[Floats2d], List[Floats2d]] = Model(
+            "span_predictor",
+            forward=span_predictor_forward,
+            init=span_predictor_init,
+            dims={"nI": nI},
+            attrs={
+                "distance_embedding_size": distance_embedding_size,
+                "hidden_size": hidden_size,
+                "conv_channels": conv_channels,
+                "window_size": window_size,
+                "max_distance": max_distance,
+            },
+        )
+        head_info = build_get_head_metadata(prefix)
+        model = (tok2vec & head_info) >> span_predictor
+        model.set_ref("span_predictor", span_predictor)
+
+    return model
+
+
+def span_predictor_init(model: Model, X=None, Y=None):
+    if model.layers:
+        return
+
+    if X is not None and model.has_dim("nI") is None:
+        model.set_dim("nI", get_width(X))
+
+    hidden_size = model.attrs["hidden_size"]
+    distance_embedding_size = model.attrs["distance_embedding_size"]
+    conv_channels = model.attrs["conv_channels"]
+    window_size = model.attrs["window_size"]
+    max_distance = model.attrs["max_distance"]
+
+    model._layers = [
+        PyTorchWrapper(
+            SpanPredictor(
+                model.get_dim("nI"),
+                hidden_size,
+                distance_embedding_size,
+                conv_channels,
+                window_size,
+                max_distance,
+            ),
+            convert_inputs=convert_span_predictor_inputs,
+        )
+        # TODO maybe we need mixed precision and grad scaling?
+    ]
+
+
+def span_predictor_forward(model: Model, X, is_train: bool):
+    return model.layers[0](X, is_train)
+
+
+def convert_span_predictor_inputs(
+    model: Model,
+    X: Tuple[List[Floats2d], Tuple[List[Ints1d], List[Ints1d]]],
+    is_train: bool,
+):
+    tok2vec, (sent_ids, head_ids) = X
+    # Normally we should use the input is_train, but for these two it's not relevant
+    # TODO fix the type here, or remove it
+    def backprop(args: ArgsKwargs) -> Tuple[List[Floats2d], None]:
+        gradients = cast(Floats2d, torch2xp(args.args[1]))
+        # The sent_ids and head_ids are None because no gradients
+        return ([gradients], None)
+
+    word_features = xp2torch(tok2vec[0], requires_grad=is_train)
+    sent_ids_tensor = xp2torch(sent_ids[0], requires_grad=False)
+    if not head_ids[0].size:
+        head_ids_tensor = torch.empty(size=(0,))
+    else:
+        head_ids_tensor = xp2torch(head_ids[0], requires_grad=False)
+
+    argskwargs = ArgsKwargs(
+        args=(sent_ids_tensor, word_features, head_ids_tensor), kwargs={}
+    )
+    return argskwargs, backprop
+
+
+def build_get_head_metadata(prefix):
+    model = Model(
+        "HeadDataProvider", attrs={"prefix": prefix}, forward=head_data_forward
+    )
+    return model
+
+
+def head_data_forward(model, docs, is_train):
+    """A layer to generate the extra data needed for the span predictor."""
+    sent_ids = []
+    head_ids = []
+    prefix = model.attrs["prefix"]
+    for doc in docs:
+        sids = model.ops.asarray2i(get_sentence_ids(doc))
+        sent_ids.append(sids)
+        heads = []
+        for key, sg in doc.spans.items():
+            if not key.startswith(prefix):
+                continue
+            for span in sg:
+                # TODO warn if spans are more than one token
+                heads.append(span[0].i)
+        heads = model.ops.asarray2i(heads)
+        head_ids.append(heads)
+    # each of these is a list with one entry per doc
+    # backprop is just a placeholder
+    # TODO it would probably be better to have a list of tuples than two lists of arrays
+    return (sent_ids, head_ids), lambda x: []
+
+
+# TODO this should maybe have a different name from the component
+class SpanPredictor(torch.nn.Module):
+    def __init__(
+        self,
+        input_size: int,
+        hidden_size: int,
+        dist_emb_size: int,
+        conv_channels: int,
+        window_size: int,
+        max_distance: int,
+    ):
+        super().__init__()
+        if max_distance % 2 != 0:
+            raise ValueError("max_distance has to be an even number")
+        # input size = single token size
+        # 64 = probably distance emb size
+        self.ffnn = torch.nn.Sequential(
+            torch.nn.Linear(input_size * 2 + dist_emb_size, hidden_size),
+            torch.nn.ReLU(),
+            torch.nn.Dropout(0.3),
+            # TODO seems weird the 256 isn't a parameter???
+            torch.nn.Linear(hidden_size, 256),
+            torch.nn.ReLU(),
+            torch.nn.Dropout(0.3),
+            # this use of dist_emb_size looks wrong but it was 64...?
+            torch.nn.Linear(256, dist_emb_size),
+        )
+        kernel_size = window_size * 2 + 1
+        self.conv = torch.nn.Sequential(
+            torch.nn.Conv1d(dist_emb_size, conv_channels, kernel_size, 1, 1),
+            torch.nn.Conv1d(conv_channels, 2, kernel_size, 1, 1),
+        )
+        self.max_distance = max_distance
+        # handle distances between +-(max_distance - 2 / 2)
+        self.emb = torch.nn.Embedding(max_distance, dist_emb_size)
+
+    def forward(
+        self,
+        sent_id,
+        words: torch.Tensor,
+        heads_ids: torch.Tensor,
+    ) -> torch.Tensor:
+        """
+        Calculates span start/end scores of words for each span
+        for each head.
+
+        sent_id: Sentence id of each word.
+        words: features for each word in the document.
+        heads_ids: word indices of span heads
+
+        Returns:
+            torch.Tensor: span start/end scores, (n_heads x n_words x 2)
+        """
+        # If we don't receive heads, return empty
+        device = heads_ids.device
+        if heads_ids.nelement() == 0:
+            return torch.empty(size=(0,))
+        # Obtain distance embedding indices, [n_heads, n_words]
+        relative_positions = heads_ids.unsqueeze(1) - torch.arange(
+            words.shape[0], device=device
+        ).unsqueeze(0)
+        md = self.max_distance
+        # make all valid distances positive
+        emb_ids = relative_positions + (md - 2) // 2
+        # "too_far"
+        emb_ids[(emb_ids < 0) + (emb_ids > md - 2)] = md - 1
+        # Obtain "same sentence" boolean mask: (n_heads x n_words)
+        heads_ids = heads_ids.long()
+        same_sent = sent_id[heads_ids].unsqueeze(1) == sent_id.unsqueeze(0)
+        # To save memory, only pass candidates from one sentence for each head
+        # pair_matrix contains concatenated span_head_emb + candidate_emb + distance_emb
+        # for each candidate among the words in the same sentence as span_head
+        # (n_heads x input_size * 2 x distance_emb_size)
+        rows, cols = same_sent.nonzero(as_tuple=True)
+        pair_matrix = torch.cat(
+            (
+                words[heads_ids[rows]],
+                words[cols],
+                self.emb(emb_ids[rows, cols]),
+            ),
+            dim=1,
+        )
+        lengths = same_sent.sum(dim=1)
+        padding_mask = torch.arange(0, lengths.max().item(), device=device).unsqueeze(0)
+        # (n_heads x max_sent_len)
+        padding_mask = padding_mask < lengths.unsqueeze(1)
+        # (n_heads x max_sent_len x input_size * 2 + distance_emb_size)
+        # This is necessary to allow the convolution layer to look at several
+        # word scores
+        padded_pairs = torch.zeros(
+            *padding_mask.shape, pair_matrix.shape[-1], device=device
+        )
+        padded_pairs[padding_mask] = pair_matrix
+        res = self.ffnn(padded_pairs)  # (n_heads x n_candidates x last_layer_output)
+        res = self.conv(res.permute(0, 2, 1)).permute(
+            0, 2, 1
+        )  # (n_heads x n_candidates, 2)
+
+        scores = torch.full(
+            (heads_ids.shape[0], words.shape[0], 2), float("-inf"), device=device
+        )
+        scores[rows, cols] = res[padding_mask]
+        # Make sure that start <= head <= end during inference
+        if not self.training:
+            valid_starts = torch.log((relative_positions >= 0).to(torch.float))
+            valid_ends = torch.log((relative_positions <= 0).to(torch.float))
+            valid_positions = torch.stack((valid_starts, valid_ends), dim=2)
+            return scores + valid_positions
+        return scores