Semi-Supervised and Unsupervised Sense Annotation via Translations

Abstract

Acquisition of multilingual training data continues to be a challenge in word sense disambiguation (WSD). To address this problem, unsupervised approaches have been developed in recent years that automatically generate sense annotations suitable for training supervised WSD systems. We present three new methods to creating sense-annotated corpora, which leverage translations, parallel corpora, lexical resources, and contextual and synset embeddings. Our semi-supervised method applies machine translation to transfer existing sense annotations to other languages. Our two unsupervised methods use a knowledge-based WSD system to annotate a parallel corpus, and refine the resulting sense annotations by identifying lexical translations. We obtain state-ofthe-art results on standard WSD benchmarks. Our annotated corpora are made available at: GitHub Link.

Multi-Head Self-Attention Gated-Dilated Convolutional Neural Network for Word Sense Disambiguation

“…He experimented with 9 different contexts based on language models including ELMo, BERT, and RoBERTa, and investigated their impacts on WSD [18]. Hauer combines semi-supervised and unsupervised methods to construct new corpus, which has been experimentally demonstrated to achieve advanced results on standard WSD model [19]. Semi-supervised method provides an effective way to improve WSD model's accuracy using labeled and unlabeled data.…”

One Sense per Translation

“…Barba et al (2020) use a pre-trained language model to identify semantically-equivalent translations of manually sense-annotated tokens. Most recently, Hauer et al (2021) propose a family of pipeline approaches employing WSD methods, machine translation, lexical resources, and various filtering techniques.…”