Parallel Corpus for Japanese Spoken-to-Written Style Conversion

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 6346–6353 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 6346 Parallel Corpus for Japanese Spoken-to-Written Style Conversion Mana Ihori, Akihiko Takashima, Ryo Masumura NTT Media Intelligence Laboratories, NTT Corporation 1-1 Hikarinooka, Yokosuka-Shi, Kanagawa 239-0847, Japan {mana.ihori.kx, akihiko.takashima.dg, ryou.masumura.ba}@hco.ntt.co.jp Abstract With the increase of automatic speech recognition (ASR) applications, spoken-to-written style conversion that transforms spoken-style text into written-style text is becoming an important technology to increase the readability of ASR transcriptions. To establish such conversion technology, a parallel corpus of spoken-style text and written-style text is beneficial because it can be utilized for building end-to-end neural sequence transformation models. Spoken-to-written style conversion involves multiple conversion problems including punctuation restoration, disfluency detection, and simplification. However, most existing corpora tend to be made for just one of these conversion problems. In addition, in Japanese, we have to consider not only general spoken-to-written style conversion problems but also Japanese-specific ones, such as language style unification (e.g., polite, frank, and direct styles) and omitted postpositional particle expressions restoration. Therefore, we created a new Japanese parallel corpus of spoken-style text and written-style text that can simultaneously handle general problems and Japanese-specific ones. To make this corpus, we prepared four types of spoken-style text and utilized a crowdsourcing service for manually converting them into written-style text. This paper describes the building setup of this corpus and reports the baseline results of spoken-to-written style conversion using the latest neural sequence transformation models. Keywords: Spoken-to-written style conversion, Japanese text style conversion, crowdsourcing, ASR transcription 1. Introduction It has become increasingly important to precisely under- stand spoken language because various automatic speech recognition (ASR) applications such as artificial intelli- gence speakers (Li et al., 2017; Purington et al., 2017) and automatic dictation systems (Shang et al., 2018; Li et al., 2019) have been growing recently. Spoken languages are typically transcribed as spoken-style text that includes dis- fluencies and redundant expressions because ASR systems convert speech into text in a literal manner. However, it is difficult for humans to read such spoken-style text because they are more familiar with reading written-style text that does not include these expressions. In addition, spoken- style text has an adverse effect on subsequent processing (e.g., machine translation, summarization), because these technologies are often developed to handle written-style text. Therefore, spoken-style text needs to be converted into written-style text. Spoken-to-written style conversion is considered as mono- lingual translation (Wubben et al., 2010) regarded as se- quence to sequence mapping from text to text. So far, var- ious methods such as noisy channel models and hidden Markov models (Johnson and Charniak, 2004; Ferguson et al., 2015; Matusov et al., 2006) have been introduced to handle these kinds of monolingual translation problems. In recent studies, neural sequence transformation models (Sutskever et al., 2014) have been utilized for the monolin- gual translation problems and demonstrated superior per- formance (See et al., 2017). However, such models require a large parallel corpus of …

Content truncated to 500 words