Corpus of Regional American English

Corpus of Regional American English
Name	Corpus of Regional American English
Abbreviation	C-R-A-E
Country	United States
Established	2003
Developers	Harvard University; University of Georgia; University of Michigan; University of Minnesota
Languages	English (Regional varieties)
Size	~1.7 million words (spoken and written)
Media	Audio recordings; transcripts; survey responses

Corpus of Regional American English is a large, publicly referenced corpus capturing spoken and written varieties across the United States. It was developed to document regional lexical, phonological, and syntactic variation among speakers sampled from distinct locales such as Boston, New York City, Philadelphia, Baltimore, Washington, D.C., Charlotte, Atlanta, New Orleans, Jackson, Houston, Dallas, San Antonio, Phoenix, Los Angeles, and San Francisco. The corpus has informed comparative work involving datasets from institutions like Harvard University, University of Georgia, University of Michigan, Linguistic Society of America, National Science Foundation, and American Dialect Society.

Overview

The corpus maps regional distribution of lexical items, phonetic realizations, and syntactic patterns across metropolitan and rural sites including Chicago, Detroit, Cleveland, Pittsburgh, Cincinnati, St. Louis, Columbus (Ohio), Minneapolis, St. Paul, Milwaukee, Indianapolis, Nashville, Memphis, Birmingham, Charleston (South Carolina), Savannah, Jacksonville, Tampa, Orlando, Sacramento, San Diego, Seattle, and Portland (Oregon). Contributors include fieldworkers affiliated with Yale University, Princeton University, Columbia University, University of Pennsylvania, Brown University, Dartmouth College, Cornell University, and regional colleges. The project aligns with earlier surveys such as the Linguistic Atlas of the United States and Canada and the Dictionary of American Regional English.

Design and Data Collection

Data collection combined interview techniques from sociolinguistic fieldwork pioneered by researchers connected to William Labov and projects at University of Pennsylvania and Columbia University. Sampling frames targeted age cohorts, gender groups, and local residence histories in sites like Raleigh, Greensboro, Burlington (Vermont), Providence, Manchester (New Hampshire), Baton Rouge, Little Rock, Omaha, Sioux Falls, Boise, Albuquerque, Santa Fe, Las Vegas, and Honolulu. Materials include recorded interviews, elicitation tasks used in studies at University of California, Berkeley, and written surveys similar to instruments from the American National Election Studies and the General Social Survey. Annotation protocols drew on standards from International Phonetic Association transcriptions and tagging conventions used by the Penn Treebank and corpora curated by Linguistic Data Consortium.

Linguistic Features and Findings

Analyses of the corpus have documented well-known features such as the Northern Cities Vowel Shift found around Buffalo, Rochester, Syracuse, and Albany (New York); the Southern Vowel Shift in areas like Charleston (South Carolina), Savannah, and Columbia (South Carolina); and the cot–caught merger patterns prevalent in Denver, Salt Lake City, Phoenix, and Las Vegas. Studies revealed lexical regionalisms—terms for footwear, carbonated beverages, and communal spaces—parallel to entries in the Dictionary of American Regional English and to patterns reported by scholars at University of Texas at Austin, Texas A&M University, Rice University, Vanderbilt University, and University of North Carolina at Chapel Hill. Syntactic variation such as multiple negation and the use of completive markers have been traced in corpora aligned with demographic data from United States Census Bureau tracts. Phonetic analyses used tools developed at Max Planck Institute for Psycholinguistics and software like Praat, with comparative references to acoustic datasets from University College London and McGill University.

Research Applications and Studies

Researchers have employed the corpus in studies bridging sociolinguistics, historical linguistics, computational modeling, and forensic linguistics. Projects at MIT, Stanford University, Carnegie Mellon University, University of California, Los Angeles, University of Illinois Urbana-Champaign, Pennsylvania State University, University of Florida, University of Kentucky, University of Missouri, University of Wisconsin–Madison, and Rutgers University have used it to model dialect leveling, sound change progression, and age-grading patterns. Comparative work has been published in journals associated with American Association for Applied Linguistics, Modern Language Association, Cambridge University Press, and Oxford University Press. The corpus has also informed speech-recognition adaptation efforts at companies collaborating with labs at Google, Microsoft, and Apple.

Access and Data Formats

Access to the corpus is typically mediated through academic partnerships and data-use agreements with institutions such as Harvard University and University of Georgia. Data formats include orthographic transcripts, time-aligned audio files (WAV), Praat TextGrids, and annotation tiers compatible with tools from the Linguistic Data Consortium and formats used in the Penn Treebank community. Metadata often references place identifiers consistent with United States Geological Survey GNIS entries and demographic variables linked to United States Census Bureau data. Licensing models have varied from open-access subsets for classroom use to restricted-access datasets for research under Institutional Review Board oversight at universities like University of Michigan and University of Minnesota.

Limitations and Criticism

Critiques address sampling biases, representativeness, and corpus size relative to large-scale corpora like those maintained by Linguistic Data Consortium or commercial datasets hosted by Google and Microsoft. Scholars at Yale University, Princeton University, Stanford University, University of California, Berkeley, and University of Pennsylvania have noted constraints in longitudinal coverage, uneven rural representation in states such as Montana, Wyoming, North Dakota, and South Dakota, and underrepresentation of some ethnic varieties prominent in Los Angeles, Houston, Miami, and New York City. Methodological debates compare elicitation versus naturally occurring speech, referencing protocols from the International Phonetic Association and standards discussed at meetings of the Linguistic Society of America. Data privacy and consent procedures have been scrutinized with reference to Institutional Review Boards at Harvard University and University of Georgia.

Category:Linguistic corpora