Improving Word Sense Disambiguation with Translations

Abstract

It has been conjectured that multilingual information can help monolingual word sense disambiguation (WSD). However, existing WSD systems rarely consider multilingual information, and no effective method has been proposed for improving WSD by generating translations. In this paper, we present a novel approach that improves the performance of a base WSD system using machine translation. Since our approach is language independent, we perform WSD experiments on several languages. The results demonstrate that our methods can consistently improve the performance of WSD systems, and obtain state-ofthe-art results in both English and multilingual WSD. To facilitate the use of lexical translation information, we also propose BABALIGN, an precise bitext alignment algorithm which is guided by multilingual lexical correspondences from BabelNet.

Semi-Supervised and Unsupervised Sense Annotation via Translations

“…We apply LABELPROP to a sense-annotated English corpus comprised of SemCor (Miller et al, 1994) and the WordNet Gloss Corpus (WNG) (Langone et al, 2004). Following Luan et al (2020), we translate each sentence of our English corpus with Google Translate independently into Italian, Spanish, French, and German. As described in Section 3.1 we use BABALIGN (Luan et al, 2020) for word alignment, with FASTALIGN (Dyer et al, 2013) as its base word aligner.…”

“…For alignment, we use BABALIGN (Luan et al, 2020), a high-precision alignment tool which leverages translation information from BabelNet to improve on a base alignment system. In particular, BABALIGN augments the input corpus with lexical translation pairs to bias the aligner towards aligning words which are mutual translations.…”

“…Now that we have annotated both sides of the bitext independently, we leverage the lexical translation information inherent in the bitext to increase the accuracy of the sense annotations. To improve the performance of our base WSD system, we employ the SOFTCONSTRAINT method of Luan et al (2020). This method is applicable to any base WSD system which assigns a numerical score, such as a probability, to each sense of a disambiguated word.…”

“…In our case, translation information is readily available from the bitext. As with our LABELPROP method, we use BABALIGN (Luan et al, 2020) to word align the bitext. For each sense-annotated token, the aligned word or phrase is treated as its translation.…”

FII_CROSS at SemEval-2021 Task 2: Multilingual and Cross-lingual Word-in-Context Disambiguation

“…We notice that evaluating such approaches is not easy. For instance, traditional Word Sense Disambiguation (WSD) fails to test latent representations unless these are linked to explicit sense inventories such as WordNet (Matusevych, 2016) or BabelNet (Navigli and Pozetto, 2012;Luan et al, 2020). To resolve the problem of disambiguation for both lingual dimensions, we tried to use a combination of well-known algorithms to provide an optimal system.…”