Introducing PerMet 1.0: A Metaphor-Annotated Corpus for Persian

Abstract

Metaphor plays a central role in human language and thought, and corpus-linguistic approaches enable its systematic investigation. Such research requires large, representative collections of metaphor-annotated linguistic data from diverse contexts. Despite the increasing availability of metaphor corpora in various languages, Persian remains underrepresented, with few publicly available resources and no large-scale register-diverse metaphor corpus. This paper introduces PerMet 1.0, a metaphor-annotated corpus for Persian. The corpus consists of approximately 120,000 tokens (about 99,000 lexical units) drawn from five registers: academic, news, fiction, social media, and spoken discourse. Five independent annotators labeled the corpus using Metaphor Identification Procedure Vrije Universiteit (MIPVU), with adaptations for Persian. Inter-annotator agreement showed a high level of consistency (κ = 0.952), confirming the reliability of the annotation. Preliminary analysis shows that 13.1% of the lexical units are related to metaphor, with the academic register showing the highest proportion, followed by news, social media, spoken, and fiction. PerMet 1.0 offers a foundational resource for research on metaphor in Persian, cross-linguistic comparative studies, and the development and fine-tuning of machine learning or large language models for automatic metaphor identification.

Resources

Details