State of the art Lexical Substitution in Context


Keywords
deep-learning, language-model, lexical-analysis, lexical-features, lexical-parser, machine-learning, natural-language-processing, nlp, transformer
License
MIT
Install
pip install lexsub==1.1.0

Documentation

Setup

pip3 install lexsub
 
Release: https://pypi.org/project/lexsub/   
Version: 1.0.5 

Background

The Lexical Substitution task involves selecting and ranking lexical paraphrases for a target word in a given sentential context. In the task, annotators and systems find an alternative substitute word or phrase for a target word in context. The task involves both finding the synonyms and disambiguating the context. To give an example:

Context: The wine was too strong to drink.
Target Word: strong

Predicted Candidates (Ranked): 
powerful 
potent 
warm
hot 
solid 
hard 

Powerful and potent are much better replacements as indicated by the score next to them as well. Whereas, all are viable replacement candidates given the context from a language modeling perspective.
Now notice two important things that is interesting:

  • Not all synonyms fit in the context.
    (direct lexical substitutions aka synonyms won’t always work)
  • Not all words that fit in the context preserve the meaning of the sentence. (LM score doesn’t always correlate)

Datasets:

A. Lexical Substitution:

B. Word Sense Disambiguation:

References:

  1. SOTA-BERT
  2. Pre-BERT-SOTA,Melamud
  3. PIC- Katrin
  4. SemBERT, AAAI 2020
  5. LIBERT
  6. Morgifier LSTM, ICLR 2020