PINGAN_AI at SemEval-2022 Task 9: Recipe knowledge enhanced model applied in Competence-based Multimodal Question Answering

Proceedings of the 16th International Workshop on Semantic Evaluation (SemEval-2022), pages 151 - 157 July 14-15, 2022 ©2022 Association for Computational Linguistics HYU at SemEval-2022 Task 2: Effective Idiomaticity Detection with Consideration at Different Levels of Contextualization Youngju Joung∗ Dept. of Industrial Security Dept. of Applied Statistics Chung Ang University ojoo.yj@gmail.com Taeuk Kim∗∗ Dept. of Computer Science Dept. of Artificial Intelligence Hanyang University kimtaeuk@hanyang.ac.kr Abstract We propose a unified framework that enables us to consider various aspects of contextualiza- tion at different levels to better identify the id- iomaticity of multi-word expressions. Through extensive experiments, we demonstrate that our approach based on the inter- and inner-sentence context of a target MWE is effective in improv- ing the performance of related models. We also share our experience in detail on the task of SemEval-2022 Task 2 such that future work on the same task can be benefited from this. 1 Introduction Multi-word expressions (MWEs) are a group of lin- guistic components containing two or more words with outstanding collocation (Baldwin and Kim, 2010; Constant et al., 2017). MWEs are valuable in that they contribute to enriching the expressive- ness of a language, allowing diverse interpretations of their meaning according to the context in which they are located. That is, the semantics of an MWE can be originated from either (i) the direct composi- tion of the literal definitions of its constituents (i.e., compositional meaning) or (ii) its conventional us- age in the language (i.e., idiomatic meaning). For instance, given an expression called wet blanket, its compositional meaning is ‘a piece of cloth soaked in liquid’, whereas its idiomatic meaning is ‘a per- son who spoils the mood’ (see Table 1). While MWEs function as an effective means of improving the abundance of a language, they are also one of the main obstacles that compli- cate natural language processing (NLP), from the perspective that an NLP model should be able to precisely identify their mode. In addition, the cur- rent trend where the goal of most NLP models is chiefly focused on capturing compositionality raises the question of how properly to deal with *Work done while Youngju was an intern at HYU. *∗Corresponding author. Category Meaning Example Compositional (Non-idiomatic) A piece of cloth soaked in liquid. And finally, the snow falls again, this time in a thick, wet blanket that encapsulates everything. Idiomatic A person who spoils the mood. When Marie brings him down to earth, it’s not clear if she’s being a passive-aggressive wet blanket or if she might have a point. Table 1: Comparison between the compositional and idiomatic meanings of the expression wet blanket. idiomatic aspects of linguistic expressions (Garcia et al., 2021a,b; Zeng and Bhat, 2021). An intuitive solution for mitigating the afore- mentioned problem is an introduction of a sophisti- cated method designed to estimate the idiomaticity of a given expression, which enables the separate processing of the expression according to its cat- egory. In a similar vein, we propose a series of techniques for better detecting …

Content truncated to 500 words