[Paper Review] A Computational Approach to Measuring the Semantic Divergence of Cognates

Ana Sabina Uban, Alina-Maria Ciobanu+1 more|arXiv (Cornell University)|Dec 2, 2020

Natural Language Processing Techniques4 citations

TL;DR

This paper proposes a cross-lingual word embedding-based method to quantitatively measure semantic divergence among cognates in related languages, particularly Romance languages and Latin. It introduces a framework for detecting false friends—cognates with divergent meanings—distinguishing between 'hard' and 'soft' false friends, and providing correction suggestions, achieving up to 88.46% recall on manually validated data sets.

ABSTRACT

Meaning is the foundation stone of intercultural communication. Languages are continuously changing, and words shift their meanings for various reasons. Semantic divergence in related languages is a key concern of historical linguistics. In this paper we investigate semantic divergence across languages by measuring the semantic similarity of cognate sets in multiple languages. The method that we propose is based on cross-lingual word embeddings. In this paper we implement and evaluate our method on English and five Romance languages, but it can be extended easily to any language pair, requiring only large monolingual corpora for the involved languages and a small bilingual dictionary for the pair. This language-agnostic method facilitates a quantitative analysis of cognates divergence -- by computing degrees of semantic similarity between cognate pairs -- and provides insights for identifying false friends. As a second contribution, we formulate a straightforward method for detecting false friends, and introduce the notion of "soft false friend" and "hard false friend", as well as a measure of the degree of "falseness" of a false friends pair. Additionally, we propose an algorithm that can output suggestions for correcting false friends, which could result in a very helpful tool for language learning or translation.

Motivation & Objective

• To develop a language-agnostic method for quantifying semantic divergence between cognate word pairs across related languages.
• To automatically detect false friends—cognates with divergent meanings—using cross-lingual word embeddings.
• To introduce a distinction between 'hard' and 'soft' false friends and define a measure of 'falseness' for improved detection.
• To provide a correction suggestion algorithm for false friends to support language learning and translation systems.
• To evaluate the method on multiple Romance languages and English, using both manual and WordNet-based gold standards.

Proposed method

• The method uses cross-lingual word embeddings trained on large monolingual corpora and a small bilingual dictionary to compute semantic similarity between cognate pairs.
• Semantic divergence is measured using a pairwise similarity metric between word embeddings across language pairs.
• A classifier uses embedding similarity scores as features to distinguish true cognates from false friends.
• The framework introduces a 'falseness' measure to quantify how misleading a false friend pair is, enabling distinction between hard and soft false friends.
• For correction suggestions, the model uses embedding space proximity to recommend the most semantically appropriate word in context.
• Evaluation uses both manually curated test sets and WordNet-based gold standards to validate performance across multiple language pairs.

Experimental results

Research questions

• RQ1 How can semantic divergence between cognates in related languages be quantitatively measured using cross-lingual embeddings?
• RQ2 To what extent can cross-lingual word embeddings detect false friends among cognate pairs in Romance languages?
• RQ3 Can a distinction between 'hard' and 'soft' false friends be meaningfully defined and measured using embedding similarity?
• RQ4 How effective is the proposed method in identifying and suggesting corrections for false friends compared to existing approaches?
• RQ5 Does the semantic divergence measure align with established linguistic classifications of language relationships?

Key findings

• Spanish and Portuguese showed the highest semantic similarity among the Romance language pairs, indicating the closest semantic divergence.
• Romanian exhibited the greatest semantic divergence from Latin, likely due to its geographical and historical distance from the Latin linguistic core.
• The method achieved 88.46% recall on the manually validated English-Spanish false friend detection task, indicating strong performance in identifying false friends.
• For the WordNet-based evaluation, the method achieved 77.25% accuracy on English-Portuguese pairs, with precision and recall values reflecting moderate to high performance.
• The clustering of languages based on semantic divergence produced a hierarchy consistent with the accepted historical tree of Romance languages.
• Even though English largely inherited Latin words through borrowing, its semantic similarity to Latin remained close to that of modern Romance languages, suggesting shared semantic evolution.