Abstract: Metaphoric expressions are regular in ordinary language. Metaphor identification is important in natural language processing since it comes in several common tasks. Conventional methodologies, like phrase-level metaphor identification, identify metaphors with word pairs, where an objective word whose metaphoricity is to be distinguished is given ahead of time. However, such objective words are not featured in genuine content information; a more up-to-date approach is sequential metaphor identification. Also, most of the conventional methodologies use restricted linguistic context to identify metaphors like by considering a single verbs argument or the sentence containing a phrase. Since context has an inevitable role in identifying metaphors, the wider context is critical in metaphor identification tasks. In this work, a novel neural sequential metaphor identification system, constrained to semantically correct input and considers a wider context area, has been proposed. The system is tested on two widely used metaphor datasets: VUA and MOH-X and outperforms the previous approaches.
Keywords: BERT, BiLSTM, Context- dependent, Metaphor.
|
DOI:
10.17148/IJARCCE.2020.9641
