Linguistics And CHILDES Research Paper

Child language transcript sharing had been initiated by Roger Brown in the first large longitudinal study of child language in the audiorecording era. Brown’s study of children he called Adam, Eve, and Sarah, which established procedures and basic measures that would become standard in the field, had been started in the early 1960s at Harvard. In addition to weekly or biweekly recordings of the children at home, Brown established a seminar of graduate students and col-leagues who met to study the newly produced transcripts. To facilitate this effort, the Brown corpus was produced using mimeograph masters. One full (though somewhat faded) set of these transcript copies was still available in 1983, and indeed formed one of the first large corpora entered into the CHILDES system (after being photocopied to improve contrast, scanned, then checked and reformatted extensively to match the CHILDES-prescribed format).

The history of Brown’s generous sharing of the Adam, Eve, and Sarah transcripts was crucial to the ready acceptance of the CHILDES proposal by the field of child linguists as well, since it had demonstrated the potential for multiple waves of analysis of a wide variety of phenomena using a single corpus. Dozens of articles and books had been published using the Brown corpus, focusing on topics as diverse as the development of reference to the past (Cromer 1968), maternal teaching techniques (Moerk 1977), conversations about kinship (Goldfield and Snow 1992), and semantic and morphological development (Brown himself 1973). These provided evidence of the utility, efficiency, and feasibility of data sharing.

2. Design Of CHILDES

Initial funding from MacArthur provided for the development of the basic components of the system. These included a set of transcription conventions, known collectively as CHAT (Codes for the Human Analysis of Transcripts) (see MacWhinney 2000a). The CHAT guidelines were imposed ex post facto on important corpora, such as Brown’s, that had already been collected, but of course much more importantly came to define one set of expectations for how transcripts should be produced. While considerable discussion concerning the advantages of various transcript formats occurred during the 1980s, and while limitations of the CHAT system for representing conversational features easily are clear, nonetheless the convergence by a large portion of the child language community on a single transcription system had important advantages. The CHAT system was elaborated with suggested procedures for phonetic, morphosyntactic, speech act, and error coding. It is important to note, though, that the vast majority of the CHAT guidelines are merely advisory; a transcript that conforms in certain minimal ways to the CHAT format (minCHAT) can be incorporated into the CHILDES database and analyzed using the CLAN programs. The major inducement for adopting CHAT guidelines was that transcripts that conformed to CHAT were susceptible to analysis by CLAN (Child Language Analysis) programs, a set of programs designed to automate various analytic procedures that had become standard in the field of child language. The earliest CLAN programs produced automatic counts of morphemes, words, utterances, and turns, to generate measures such as Mean Length of Utterance, Mean Length of Turn, and Type-token Ratio. Key word searches were also made possible (KWAL), as were procedures for Boolean searches (Combo). Subsequent expansions added programs to select certain marked portions of the transcript for further analysis (GEM), to calculate contingencies between codes in adjacent turns (Keymap), to code transcript lines for overlap with preceding lines (CHIP), and so forth. In addition, morphological tagging programs have been made available for a number of languages. These various programs are designed in most cases to analyze many files simultaneously and to generate output in a format directly usable by statistical analysis programs.

The CHILDES database is a constantly growing electronic repository of transcript data (see MacWhinney 2000b). Corpora included in the data-base as of January 2000 represented the acquisition of 26 languages as well as a number of language pairs and triplets (in the Bilingual section). In addition, data from children with language disorders and from narrative elicitation procedures are included in separate sections. Access to files in the database is free to anyone who agrees to use the data in accordance with the guidelines specified, primarily to cite the sources and the CHILDES manual appropriately. Files can be accessed through the website or via CD-ROMs distributed in conjunction with the CHILDES manuals.

3. Funding History

After the initial period of funding from the MacArthur Foundation, support for the CHILDES system was provided to Brian MacWhinney and Catherine Snow through the National Institute of Child Health and Human Development (NICHD) and subsequently to MacWhinney from NICHD and the National Science Foundation Linguistics Program.

4. Beyond CHILDES

Since the CHILDES enterprise was designed to promote collection, use, sharing, and analysis of transcript data, it did not optimally accommodate the inclusion of audio or video data. As researchers increasingly turn to such data, and seek methods for annotating, analyzing, and sharing them, a successor system to CHILDES called TalkBank has been founded, with support from the National Science Foundation. TalkBank (http: //www.talkbank.org/) will provide standards, tools, and support for creating, searching, and publishing video-based language inter-actions using electronic networks.

Bibliography:

1. Brown R 1973 A First Language: The Early Stages. George Allen & Unwin Ltd, London
2. Cromer R 1968 The development of temporal reference during the acquisition of language. Ph.D. thesis, Harvard University.
3. Goldfield B, Snow C E 1992 What’s cousin Arthur’s daddy’s name?: The acquisition of knowledge about kinship. First Language 12: 187–205
4. Moerk E 1977 Pragmatic and Semantic Aspects of Early Language Development. University Park Press, Baltimor
5. MacWhinney B 2000a The CHILDES Project: Tools for Analyzing Talk, Volume I: Transcription Format and Pro-grams, 3rd edn. Lawrence Erlbaum Associates, Mahwah, NJ
6. MacWhinney B 2000b The CHILDES Project: Tools for Analyzing Talk, Volume II: The Database, 3rd edn. Lawrence Erlbaum Associates, Mahwah, NJ