The transcription of speech into a graphic code is a basic procedure in both dialectology and sociolinguistics, and typically, it is this graphic analog of speech that forms the data for analysis in both disciplines. The act of transcription, however, affects linguistic data in subtle,difficult-to-detect ways. This paper demonstrates some of those effects by examining impressionistic phonetic data in American linguistic atlases. The paper identifies three sources of transcriber effects in the data: (1)conceptual differences regarding the phonetic status of particular sounds(e.g., offglides of diphthongs) and how they should be transcribed, (2)normative differences regarding the phonetic values of particular symbols, and(3) changing scribal practices as transcribers discover the importance of phonetic details that they had previously overlooked. When transcriber effects such as these are not identified and accounted for in the analysis of the data, they can create misleading results.
