조사 ‘-의’의 발음과 음성 인식 출력형에 관한 문법적 접근

Abstract

In this study, we examine the phonetic realizations of the Korean particle ‘-ui’ and investigate how this particle is output by automatic speech recognition programs. Through this investigation, we also infer the linguistic knowledge embedded in the language models of such programs. In standard Korean, the particle ‘-ui’ is typically realized as [e]. When this is transcribed directly as ‘-e’ without correction, it can cause confusion between the genitive particle ‘-ui’ and the locative particle ‘-e.’ Such misrecognition alters the grammatical case of sentence constituents, leading to an incorrect understanding of the sentence structure produced by automatic speech recognition. In this research, we use the automatic speech recognition program Naver Clover Speech to observe the output forms that appear when the particle ‘-ui’ is input in various phonetic forms under the language model’s correction. When the correction process is effective and recognition accuracy is high, we can hypothesize morphological, syntactic, and semantic reasons from a linguistic standpoint.