LyS_ACoruña at SemEval-2022 Task 10: Repurposing Off-the-Shelf Tools for Sentiment Analysis as Semantic Dependency Parsing

Proceedings of the 16th International Workshop on Semantic Evaluation (SemEval-2022), pages 1389 - 1400 July 14-15, 2022 ©2022 Association for Computational Linguistics LyS_ACoruña at SemEval-2022 Task 10: Repurposing Off-the-Shelf Tools for Sentiment Analysis as Semantic Dependency Parsing Iago Alonso-Alonso, David Vilares and Carlos Gómez-Rodríguez Universidade da Coruña, CITIC Departamento de Ciencias de la Computación y Tecnologías de la Información Campus de Elviña s/n, 15071 A Coruña, Spain {iago.alonso,david.vilares,carlos.gomez}@udc.es Abstract This paper addressed the problem of structured sentiment analysis using a bi-affine semantic dependency parser, large pre-trained language models, and publicly available translation mod- els. For the monolingual setup, we considered: (i) training on a single treebank, and (ii) relax- ing the setup by training on treebanks coming from different languages that can be adequately processed by cross-lingual language models. For the zero-shot setup and a given target tree- bank, we relied on: (i) a word-level translation of available treebanks in other languages to get noisy, unlikely-grammatical, but annotated data (we release as much of it as licenses allow), and (ii) merging those translated treebanks to obtain training data. In the post-evaluation phase, we also trained cross-lingual models that simply merged all the English treebanks and did not use word-level translations, and yet obtained better results. According to the official results, we ranked 8th and 9th in the monolingual and cross-lingual setups. 1 Introduction Sentiment Analysis (SA, Pang and Lee, 2008) deals with the automatic processing of subjective infor- mation in natural language texts. Early work on SA focused on conceptually simpler tasks, such as polarity classification at the sentence or docu- ment level. With the advances in natural language processing (NLP), more fine-grained and complex tasks have been proposed, such as detecting the entity that expresses an opinionated chunk of text, or the entity that was targeted. More particularly, Barnes et al. (2021) consider sentiment analysis as a (graph) structured task, and discuss up to five subtasks: (i) sentiment expression extraction, (ii) sentiment target extraction, (iii) sentiment holder extraction, (iv) defining the relationship between these elements, and (v) assigning a polarity label. They discuss that although these tasks have been extensively studied by different authors (Turney, 2002; Pontiki et al., 2015; Zhang et al., 2019, in- ter alia), they are not addressed all together. They also discuss that such subdivision into subtasks might have a negative impact in the general analy- sis of the sentence, and that a joint analysis could translate into a holistic approach. To do so, they propose to encapsulate all these tasks in the form of a sentiment graph. Formally, the goal is to find the set of opinion tuples {O1, . . . , Oi, . . . , On} in a given text, where each opinion Oi is a tuple of the form (h, t, e, p) where h is a holder who expresses a polarity p towards a target t through a sentiment expression e, implicitly defining pairwise relation- ships between elements of the same tuple. We illustrate an example in Figure 1. …

Content truncated to 500 words