A Corpus Construction-Based Approach to Word Sense Classification in Persian

<p dir="ltr">In this research, using a corpus-based approach, the study focuses on Persian compound verbs excluding the verb "khordan" (to eat). In compound verbs, apart from the verb "khordan," which is also referred to as a light verb, there are multiple non-verbal elements such as "zarb" (strike), "ghaza" (food), "zamin" (land), and others that co-occur. This co-occurrence in different compound verbs leads to the light verb "khordan" having various meanings alongside its non-verbal components. The aim of this research is to utilize a data-driven method for categorizing the different meanings of the light verb "khordan." For this purpose, a corpus of 50 million words, randomly sampled from the Hamshahri corpus, was used. Using the AntConc corpus analysis software, the non-verbal elements of the light verb "shodan" (to become) were extracted. Subsequently, 35 verbs were selected based on their frequency in the corpus, and for each verb, 20 usage examples were extracted from the corpus. Using a behavioral profile analysis method, the syntactic, morphological, and semantic characteristics of these verbs were annotated. Then, the behavioral profiles of the verbs were analyzed using hierarchical cluster analysis in the R software. The semantic clusters obtained from the cluster analysis indicate that meanings are categorized into distinct clusters based on the contextual usage of the verbs. Compound verbs, whose non-verbal components share a common semantic field, tend to group together in similar and closely related clusters.</p>