Can we tackle the semantic of text in a hierarchical way?
This is my last project for word sense disambiguation coarse and fine grained in the MNLP course held by professor Roberto Navigli in Sapienza, Rome.

Abstract

Word Sense Disambiguation (WSD) poses a significant challenge within the field of Natural Language Processing (NLP), requiring the accurate determination of a word’s intended meaning in a given context. This document aims to explore diverse strategies documented in the literature for solving this issue. These strategies are then applied to the provided dataset, evaluating their effectiveness.

Additionally, the document analyzes the connection between identifying coarse-grained senses and enhancing the precise classification of fine-grained senses, and vice versa. By investigating this relationship, the document aims to uncover the synergistic effects of these seemingly disparate processes.

Introduction

WSD aims to solve the ambiguity of word meaning in context; when a word has multiple meaning (polysemy) is fundamental to disambiguate the given target word in order to fully understand the complete meaning of a sentence. In order to accomplish this task two big group of technique have been applied in literature during time: fully neural approaches ([1][2] and many others) and knowledge-based approaches [3][4]. Here in this document, we will explore neural approaches applied on the given dataset. Moreover, a couple of experiment have been devoted to study how might be possible to combine in a single model the power to extrapolate in hierarchical way deeper semantic in form of word senses.

Read all the paper at this link: https://www.emanuelerucci.it/wp-content/uploads/2023/09/Report.pdf