BabelEnconding at SemEval-2020 Task 3: Contextual Similarity as a Combination of Multilingualism and Language Models

TL;DR

This paper presents BabelEnconding, a multilingual language model-based system for predicting contextual word similarity, leveraging translation and multilingual models to improve correlation with human judgments across multiple languages.

Contribution

The paper introduces a novel approach combining translation and multilingual language models to enhance contextual similarity prediction in multilingual settings.

Findings

Ranked top-3 in six out of eight language combinations

Achieved highest scores in three subtask categories

Demonstrated effectiveness of multilingual evidence in similarity prediction

Abstract

This paper describes the system submitted by our team (BabelEnconding) to SemEval-2020 Task 3: Predicting the Graded Effect of Context in Word Similarity. We propose an approach that relies on translation and multilingual language models in order to compute the contextual similarity between pairs of words. Our hypothesis is that evidence from additional languages can leverage the correlation with the human generated scores. BabelEnconding was applied to both subtasks and ranked among the top-3 in six out of eight task/language combinations and was the highest scoring system three times.