Skip to content

CHANGELOG.md

Latest commit

 

History

History
88 lines (70 loc) · 3.65 KB

CHANGELOG.md

File metadata and controls

88 lines (70 loc) · 3.65 KB
Raw

Change Log

1.3.0 — 2026-04-17

Breaking changes. Python 3.10+ required. WordNet backend rewritten against modern wn>=1.0 (@goodmami) with the Open English WordNet 2024 lexicon; synset identifiers are now oewn-<offset>-<pos> rather than PWN-3.0 bank.n.01 form. Use synset.id, synset.lemmas(), synset.definition(), synset.pos on the returned objects.

Bugs fixed:

  • utils.remove_tags — reduce was undefined, .iteritems() was Py2-only.
  • utils.synset_properties — double eval('synset.' + parameter) replaced with getattr.
  • utils.penn2morphy — bare except: narrowed to except KeyError:.
  • semeval.py — from BeautifulSoup import BeautifulSoup (Py2 package) → bs4; broken three-line yield / stray unicode() calls fixed.
  • allwords_wsd.disambiguate — surface_words / lemmas / morphy_poss were unbound when context_is_lemmatized=True.
  • similarity.max_similarity — sorted((score, synset), …) crashed on tied scores (modern wn.Synset is not < comparable).
  • pywsd/__init__.py version string (was 1.2.4 vs setup.py 1.2.5).

Import-time behaviour:

  • __builtins__['wn'] = WordNet(...) pollution removed; wn is now a lazy proxy imported explicitly by each module.
  • Eager print('Warming up PyWSD …') + simple_lesk('This is a foo bar sentence', 'bar') warmup deleted.
  • 80 MB pywsd/data/signatures/signatures.pkl no longer bundled. Signatures are computed on demand and memoised in-process with functools.lru_cache.
  • import pywsd dropped from ~11 s to ~1.3 s.

Information-content similarity:

  • Upstream wn.ic.compute v1.1.0 double-counts ancestors reachable via multiple inheritance, producing P(c) > 1 and IC(c) < 0 (e.g. Brown res(dog, cat) = −0.29). pywsd._ic.build_ic implements Resnik 1995 §3.2 correctly with a global visited set per source synset.
  • pywsd/data/ic_oewn-2024_brown.pkl (3.3 MB, ~1 M tokens) and pywsd/data/ic_oewn-2024_wikipedia.pkl (3.3 MB, ~103 M tokens, via HuggingFace datasets / wikitext-103-raw-v1) are bundled. Wikipedia is the default. scripts/build_ic.py regenerates them.
  • pywsd._wordnet.set_ic(freq_dict) overrides the bundled IC.

Packaging:

  • setup.py (distutils.core — removed in Python 3.12) → pyproject.toml (PEP 621). setup.py is now a 3-line shim.
  • Runtime deps trimmed: dropped six, pandas, numpy, and the wn==0.0.23 pin; now nltk>=3.8, wn>=1.0,<2. beautifulsoup4 moved to the [semeval] extra.
  • GitHub Actions CI (3.10 / 3.11 / 3.12 / 3.13 / 3.14, pytest -W error) replaces .travis.yml + tox.ini (which still listed py27).
  • Tests consolidated into tests/; 33 tests covering lesk, baselines, similarity, all-words, and IC correctness.

17.01.15

  • Official pyWSD version 1.0 announced.

14.01.15

  • Added all-words WSD functionality
  • Added disambiguate() for all-words WSD

05.01.15:

  • Removed old svm.py.
  • Added Merlin Machine Learning library to support SVM.

06.11.14:

  • Fixed bugs in original_lesk() and POS checks in simple_lesk()

25.10.14:

  • Added corpus readers for SemEval-2007 coarse-grain WSD data and SemCor

29.05.14 (user request):

  • Automatically lemmatize ambiguous_word parameter because without it being a lemma, wn.synsets(ambiguous_word) breaks the system.

15.05.14 (bugfix):

  • Added try-excepts to allow old NLTK Synset properties.

15.04.14:

  • Changed Synset properties to methods adhering to new NLTK version (see see http://goo.gl/hO79KO)
  • Added N-best results, pywsd now outputs ranked list of synsets (possible with normalized scores too), see test_wsd.py

03.04.14:

  • Added similarity based WSD methods.