1 Introduction

This section introduces the development of generative artificial intelligence technology and its impact on linguistic research, emphasizing the importance of spoken language corpora in comparative analysis between human language and AI-generated language. It points out the existing genre bias in current corpora, with a relative shortage of spoken language data. The necessity of constructing high-quality spoken language corpora is elaborated, including their role in showcasing natural human speech, providing research data, as well as key considerations in the design and construction process, and the importance of openness and transparency.

2 The Necessity of Modern Chinese Spoken Language Corpus Construction

This section discusses the necessity of constructing a modern Chinese spoken language corpus. Chinese researchers value empirical studies and have built several Chinese spoken language corpora, such as the "Beijing Spoken Language Survey" and the corpus query system from Beijing Language and Culture University, as well as the Beijing Spontaneous Speech Corpus from the Chinese Academy of Social Sciences, providing rich examples for research. However, existing corpora face various issues, such as being outdated and unable to reflect contemporary spoken language usage; most are not publicly accessible, making resource sharing difficult. Therefore, this study began collecting natural spoken dialogues from native Chinese speakers in 2018 to fill gaps in this field both domestically and internationally.

3 Corpus Design and Spoken Language Collection

This section details the design and collection process of the modern Chinese spoken language corpus. The corpus aims to collect daily natural spoken dialogues, targeting Chinese residents aged 18 and above, regardless of gender, education level, etc. The corpus scale is set at one million characters, focusing on quality and representativeness despite its small size, and is designed as a dynamic corpus, expanding annually. A feasibility analysis was conducted before collection, setting a goal of 250 dialogues, each lasting at least 20 minutes. For ethical preparation, an informed consent form was developed to ensure participants' voluntary and informed participation, with recordings used solely for research. Recruitment was conducted via social media, and participants were compensated for providing high-quality recordings. The spoken language collection was conducted in two phases: semi-interviews were used in 2018, and participants were allowed to self-record starting in 2022. As of 2024, spoken dialogues from 28 provinces and municipalities, totaling approximately 55 hours, have been collected. During the collection process, recordings were checked and filtered to ensure clear audio quality and that dialogues met research requirements.

4 Spoken Language Transcription

This section discusses the key aspects of spoken language transcription in the construction of the modern Chinese spoken language corpus. Transcription is the process of converting dynamic speech into static text, which is selective and subjective. Since fully faithful, all-feature transcription faces challenges in efficiency and feasibility in practice, it is necessary to establish a hierarchical transcription system based on research objectives, prioritizing the transcription of core features. This study developed transcription rules for the modern Chinese spoken language corpus, adopting a simple orthographic transcription method, using standard Chinese characters to record spoken features. For sounds without corresponding Chinese characters, a combination of characters and Pinyin is used for transcription; for the transcription of "他" and "她," a unified "TA" is used; additionally, anonymization is applied to information that may reveal speakers' privacy. Regarding consistency and reliability in transcription, this study primarily employs manual transcription, using InqScribe and Praat tools for transcription, inspection, and correction, thereby improving the quality of transcribed texts. Cross-checking ensures the consistency and reliability of transcriptions.

5 Corpus Processing and Release

This section discusses the processing and release strategy of the modern Chinese spoken language corpus. The 240,000-character corpus collected in 2018 has been released via GitHub, as it is free and allows immediate uploading of new data, although it does not support online retrieval. Subsequent plans include developing corpus retrieval and analysis software. Data collected from 2022 to 2024 will be uploaded by the end of 2025. The currently released materials include transcribed raw data, speaker information, and the transcription system, with word segmentation and part-of-speech tagging applied.

6 Conclusion

This section summarizes the design and construction of the modern Chinese spoken language corpus, noting that it is a dynamic corpus that has been expanding annually since 2018 and partially released for free research use. Additionally, a modern Chinese interlanguage spoken corpus has been constructed for comparison. However, challenges such as privacy protection limitations, gender imbalance, and the absence of specific age groups persist. Future efforts will focus on attracting more diverse groups to participate and improving information collection to enhance corpus quality.