Novel metrics for computing semantic similarity with sense embeddings

Abstract

In the last years many efforts have been spent to build word embeddings, a representational device in which word meanings are described through dense unit vectors of real numbers over a continuous, high-dimensional Euclidean space, where similarity can be interpreted as a metric. Afterwards, senselevel embeddings have been proposed to describe the meaning of senses, rather than terms. More recently, additional intermediate representations have been designed, providing a vector description for pairs ⟨term, sense⟩, and mapping both term and sense descriptions onto a shared semantic space. However, surprisingly enough, this wealth of approaches and resources has not been supported by a parallel refinement in the metrics used to compute semantic similarity: to date, the semantic similarity featuring two input entities is mostly computed as the maximization of some angular distance intervening between vector pairs, typically cosine similarity. In this work we introduce two novel similarity metrics to compare sense-level representations, and show that by exploiting the features of sense-embeddings it is possible to substantially improve on existing strategies, by obtaining enhanced correlation with human similarity ratings. Additionally, we argue that semantic similarity needs to be complemented by another task, involving the identification of the senses at the base of the similarity rating. We experimentally verified that the proposed metrics are beneficial when dealing with both semantic similarity task and sense identification task. The experimentation also provides a detailed how-to illustrating how six important sets of sense embeddings can be used to implement the proposed similarity metrics.

LMMS Reloaded: Transformer-based Sense Embeddings for Disambiguation and Beyond

“…As with GWCS, LMMS-SP XLNet-L stands out with clearly best results. While LMMS-SP also outperforms most non-contextual embeddings, it still underperforms LessLex (Colla et al, 2020b) embeddings, which are based on ensembles and learned using BabelNet. We also note that LMMS-SP performs particularly well on the 'Observed' set corresponding to senses learned from annotated corpora.…”

Enhancing service excellence: analyzing natural language question answering with advanced cosine similarity

“…Because precision and accuracy will affect the appropriate size and produce the right size results [56]. In addition, a measuring instrument must have high accuracy and precision to reduce the occurrence of errors in measurement [57].…”

CONcreTEXT norms: Concreteness ratings for Italian and English words in context

“…Contextual annotation thus involves a hidden task, that is, the identification of the meanings at the base of the concreteness rating (more on the meaning-identification task can be found in [49]). All targets were then meaning-annotated according to WordNet and are part of the released data.…”

“…One main assumption underlying the whole work is that each term may have more associated meanings, and meaning selection is determined based on the context, which, in turn, corresponds to some sort of word meaning disambiguation. That is, the concreteness rating involves a hidden task in which meanings are identified [49], and concreteness is supposed to be a property of word meanings rather than of word forms/terms. While we fully acknowledge the importance of addressing word ambiguity, encompassing both homonymy (where a word form's meanings have distinct historical origins and are not related) and polysemy (where a word form's meanings stem from the same lexical source and are related), practical considerations challenge the rigid application of the etymological criterion.…”