A Multilingual Evaluation Dataset for Monolingual Word Sense Alignment (original) (raw)

Monolingual Word Sense Alignment as a Classification Problem

2021

Words are defined based on their meanings in various ways in different resources. Aligning word senses across monolingual lexicographic resources increases domain coverage and enables integration and incorporation of data. In this paper, we explore the application of classification methods using manually-extracted features along with representation learning techniques in the task of word sense alignment and semantic relationship detection. We demonstrate that the performance of classification methods dramatically varies based on the type of semantic relationships due to the nature of the task but outperforms the previous experiments.

MultiMirror: Neural Cross-lingual Word Alignment for Multilingual Word Sense Disambiguation

2021

Word Sense Disambiguation (WSD), i.e., the task of assigning senses to words in context, has seen a surge of interest with the advent of neural models and a considerable increase in performance up to 80% F1 in English. However, when considering other languages, the availability of training data is limited, which hampers scaling WSD to many languages. To address this issue, we put forward MULTIMIRROR, a sense projection approach for multilingual WSD based on a novel neural discriminative model for word alignment: given as input a pair of parallel sentences, our model – trained with a low number of instances – is capable of jointly aligning, at the same time, all source and target tokens with each other, surpassing its competitors across several language combinations. We demonstrate that projecting senses from English by leveraging the alignments produced by our model leads a simple mBERT-powered classifier to achieve a new state of the art on established WSD datasets in French, Germa...

From Word Alignment to Word Senses, via Multilingual

Most of the successful commercial applications in language processing (text and/or speech) dispense of any explicit concern on semantics, with the usual motivations stemming from the computational high costs required, in case of large volumes of data, for dealing with semantics. With recent advances in corpus linguistics and statistical-based methods in NLP, revealing useful semantic features of linguistic data is becoming cheaper and cheaper and the accuracy of this process is steadily improving. Lately, there seems to be a growing acceptance of the idea that multilingual lexical ontologies might be the key towards aligning different views on the semantic atomic units to be used in characterizing the general meaning of various and multilingual documents. Depending on the granularity at which semantic distinctions are necessary, the accuracy of the basic semantic processing (such as word sense disambiguation) can be very high with relatively low complexity computing. The paper substantiates this statement by presenting a statistical/based system for word alignment and word sense disambiguation in parallel corpora. We describe a word alignment platform which ensures text pre-processing (tokenization, POS-tagging, lemmatization, chunking, sentence and word alignment) as required by an accurate word sense disambiguation.

The ELEXIS system for monolingual sense linking in dictionaries

CERN European Organization for Nuclear Research - Zenodo, 2021

Sense linking is the task of inferring any potential relationships between senses stored in two dictionaries. This is a challenging task and in this paper we present our system that combines Natural Language Processing (NLP) and non-textual approaches to solve this task. We formalise linking as inferring links between pairs of senses as exact equivalents, partial equivalents (broader/narrower) or a looser relation or no relation between the two senses. This formulates the problem as a five-class classification for each pair of senses between the two dictionary entries. The work is limited to the case where the dictionaries are in the same language and thus we are only matching senses whose headword matches exactly; we call this task Monolingual Word Sense Alignment (MWSA). We have built tools for this task into an existing framework called Naisc and we describe the architecture of this system as part of the ELEXIS infrastructure, which covers all parts of the lexicographic process including dictionary drafting. Next, we look at methods of linking that rely on the text of the definitions to link, firstly looking at some basic methodologies and then implementing methods that use deep learning models such as BERT. We then look at methods that can exploit non-textual information about the senses in a meaningful way. Afterwards, we describe the challenge of inferring links holistically, taking into account that the links inferred by direct comparison of the definitions may lead to logical contradictions, e.g., multiple senses being equivalent to a single target sense. Finally, we document the creation of a test set for this MWSA task that covers 17 dictionary pairs in 15 languages and some results for our systems on this benchmark. The combination of these tools provides a highly flexible implementation that can link senses between a wide variety of input dictionaries and we demonstrate how linking can be done as part of the ELEXIS toolchain.

From Word Alignment to Word Senses, via Multilingual Wordnets

2006

Lexical Sense Alignment using Weighted Bipartite b-Matching

2019

Lexical resources are important components of natural language processing (NLP) applications providing linguistic information about the vocabulary of a language and the semantic relationships between the words. While there is an increasing number of lexical resources, particularly expert-made ones such as WordNet or FrameNet as well as collaboratively- curated ones such as Wikipedia1 or Wiktionary2 , manual construction and maintenance of such resources is a cumbersome task. This can be efficiently addressed by NLP techniques. Aligned resources have shown to improve word, knowledge and domain coverage and increase multilingualism by creating new lexical resources such as Yago , BabelNet and ConceptNet In addition, they can improve the performance of NLP tasks such as word sense disambiguation semantic role tagging and semantic relations extraction.

Designing the ELEXIS Parallel Sense-Annotated Dataset in 10 European Languages

Over the course of the last few years, lexicography has witnessed the burgeoning of increasingly reliable automatic approaches supporting the creation of lexicographic resources such as dictionaries, lexical knowledge bases and annotated datasets. In fact, recent achievements in the field of Natural Language Processing and particularly in Word Sense Disambiguation have widely demonstrated their effectiveness not only for the creation of lexicographic resources, but also for enabling a deeper analysis of lexical-semantic data both within and across languages. Nevertheless, we argue that the potential derived from the connections between the two fields is far from exhausted. In this work, we address a serious limitation affecting both lexicography and Word Sense Disambiguation, i.e. the lack of high-quality sense-annotated data and describe our efforts aimed at constructing a novel entirely manually annotated parallel dataset in 10 European languages. For the purposes of the present p...

MSD-1030: A Well-built Multi-Sense Evaluation Dataset for Sense Representation Models

2020

Sense embedding models handle polysemy by giving each distinct meaning of a word form a separate representation. They are considered improvements over word models, and their effectiveness is usually judged with benchmarks such as semantic similarity datasets. However, most of these datasets are not designed for evaluating sense embeddings. In this research, we show that there are at least six concerns about evaluating sense embeddings with existing benchmark datasets, including the large proportions of single-sense words and the unexpected inferior performance of several multi-sense models to their single-sense counterparts. These observations call into serious question whether evaluations based on these datasets can reflect the sense model’s ability to capture different meanings. To address the issues, we propose the Multi-Sense Dataset (MSD-1030), which contains a high ratio of multi-sense word pairs. A series of analyses and experiments show that MSD-1030 serves as a more reliabl...

A Multilingual Evaluation Dataset for Monolingual Word Sense Alignment (original) (raw)

Related papers