Revisiting the Observer's Paradox

Almost four decades ago William Labov mapped out an approach to the study of linguistic variation and change that has become the cornerstone for sociolinguistic research up to the present. Labov's approach was based on the observation and recording of vernacular speech, the unmonitored speech that people use every day to communicate with each other in a variety of social situations, as the most systematic source of data for the analysis of linguistic structure (Labov 1966). However, the act of observation itself militates against obtaining the most casual speech styles, creating what Labov calls the "observer's paradox." Labov first confronted the problem of the observer's paradox by devising a series of interview contexts that varied the amount of attention informants paid to speech; in later studies he included peer group interviews as a way to reduce the effects of the field-worker. However, linguistic interviews, even those with peer groups, typically include a field-worker at least as part of the audience, and we cannot easily eliminate or even estimate the effects that the field-worker might have. Differing effects of field-workers are a special case of the observer's paradox, since characteristics of the interviewer (e.g., race, gender, age) may actually exacerbate the effect of the presence of the field-worker.

More recent studies have shown that characteristics of the interviewer (such as gender, age, experience, social background, and race), and characteristics of the interview itself (such as the relationship between interviewer and interviewee, the strategies used by the field-worker to gather data, the role of the field-worker in the interview situation, and the presence of other interlocutors), may also affect the data from sociolinguistic fieldwork. For example, Bailey and Tillery (1999) illustrate that individual field-workers and the strategies they use to gather data can have a significant impact on the kinds of inferences made about the distribution of linguistic features. Cukor-Avila and Bailey (1995) discuss a way to create interview contexts that diminish the role of the field-worker and allow informants to interact with each other by shifting the focus of the interview from individual informants to strategic sites of linguistic interaction. Rickford and McNair-Knox (1994) explore the effects of interviewer race on topic selection and vernacular usage in interviews by African American and white field-workers with an African American Vernacular English-speaking teenager, suggesting that more reliable data come from interviews with same-race field-workers. In a study that replicates the experiment by Rickford and McNair-Knox (1994), Cukor-Avila and Bailey (2000) suggest that any effects of race of the interviewer can be ameliorated by other factors such as familiarity, the amount of time a field-worker has spent in the community, and the use of peer groups.

Clearly, then, there is a need to account for interviewer characteristics and for interview contexts in reaching conclusions based on sociolinguistic fieldwork. The problem sociolinguists face is that while neither the exact sets of characteristics nor the ways in which they interact with one another have ever been fully specified, their impact can be enormous. In order to systematically address these issues, there needs to be a greater concern with the methods used to conduct the interviews and collect the data. Early on, sociolinguistic researchers were able to replicate the methodological framework outlined in detail in Labov's studies of New York City and Philadelphia (cf. Wolfram 1969; Fasold 1972; Trudgill 1974), allowing for cross-comparisons of the data. Recently, however, cross-comparisons of linguistic features have become more difficult as methodological explanations and consistency have become secondary to quantitative concerns. Yet the validity of the linguistic analysis is only as reliable as the methods used to gather the data. And without a careful assessment of the effects of field-workers on sociolinguistic data, we can never know to what extent these data represent the typical linguistic behavior of informants.

Patricia Cukor-Avila...