This is a report on efforts to determine the cognitive reflexes of the oral readings of diagnostic texts by struggling readers. The problem arises in connection with a diagnostic text constructed to obtain a linguistic profile of the decoding abilities of struggling readers. It has long been recognized that some oral productions that differ from the printed text may be differences in pronunciation due to dialect differences and do not signal mistakes in reading. Thus the word FIND read as /fayn/ is a potential error, which may reflect either a colloquial pronunciation of the word "find" with consonant cluster simplification, or the incorrect selection of the word "fine." Speakers of African American English do not pronounce the possessive 's in RAY'S COAT, but always use the possessive /s/ in final position, as in "This is Ray's." It is not clear whether or not the oral reading /rey kowt/ reflects the correct interpretation of a relationship of possession between the two nouns. To date, no program has been put forward to resolve such ambiguities. It is therefore difficult to say whether instruction on the full forms of words truncated in speech will improve reading.

It is proposed that an error in reading a given word, in the sense of selection of the wrong lexical item, will raise the probability of an error in the text following that is semantically and syntactically related to the word in question. The probability of a potential error of a given type being a clear error can then be estimated form the frequency of errors in the following text, compared to the frequency of following errors for clear errors and the frequency of following errors for correct readings.

The RX program, constructed to determine knowledge of phoneme-to-grapheme correspondences from oral reading errors, was modified to measure the probabilities of errors in the text following 112 potential items in a dialect reading. Some potential error types were found to match the frequencies of following errors for clear errors, others for correct reading. For some errors, significant differences between Latinos, African Americans and Whites reflect differences in the cognitive status of the grammatical forms involved.

Comparisons with speech patterns of the same populations showed high correlation between the realization of a form in speech and the likelihood of its omission in reading being a reading error. Such correlations showed opposite signs for some grammatical forms. Thus a high probability of omission of verbal /s/ in speech is correlated with a high probability that its omission in reading is a correct error, while a high probability of omission of possessive /s/ in speech is correlated with a high probability that its omission is a clear error. An effort will be made to explain this inverse relationship.

William Labov, Professor of Linguistics, Director of the Linguistics Laboratory, University of Pennsylvania. Major studies include The Social Stratification of English in New York City (1966), Sociolinguistic Patterns (1972), Language in the Inner City (1972), Principles of Linguistic Change (1994, 2000). He is the director of the Atlas of North American English and the Urban Minorities Reading Project. Labov is co-editor of Language Variation and Change, served as president of the Linguistic Society of America (1979), and is a member the National Academy of Science.
http://www.ling.upenn.edu/~labov/