The usual method adopted to satisfy the need for natural, spontaneous speech and many tokens is the sociolinguistic interview. The classic interview involves several hours of speech from every speaker. The interviewer tries to encourage casual speech by discussing topics such as childhood reminiscences and games, and personal anecdotes of dangerous, spiritual, or supernatural experiences the speaker may have had. With luck, during an extended interview the interviewee will also speak to someone other than an interviewer for a while; this is assumed to capture even more naturalistic speech. The interview may stand alone or it may be complemented with three additional verbal tasks. These very often include reading a prose passage aloud (to provide a more formal register), and the reading of a word list or minimal pairs (minimal pairs differ only in one segment, e.g. bet – bed).
A collection of interviews comprises a sociolinguistic survey or a social dialect survey. The resulting corpus of speech is similar to a regional dialect survey (in which speakers from a range of locales are asked, e.g. to name objects or ‘translate’ a set of sentences into their local, vernacular). But the data gathered in sociolinguistic survey is less subject to interference from standardized norms or perceptions about what is or is not correct. This is because the primary data consists of a stream of spontaneous speech. The trade-off associated with this spontaneity is that it is impossible to control the data that is gathered. That is, the researcher cannot be sure that all speakers will produce tokens of a variable at the same rate and in exactly the same linguistic contexts. This is one reason why sociolinguistic interview are generally very lengthy. This maximizes the chance of obtaining comparable data from all speakers.
Other methods may be used to increase the likelihood of obtaining numerous tokens of a variable. For example, if a researcher is interested in how speakers pronounce the diphthong in couch, it is possible to gain a lot of very clear tokens of the sound from the use of the so-called semantic differential tests. This involves asking interviewees a question like ‘What is the difference between a couch and a sofa?’ Or the researcher might use a rapid and anonymous survey to quickly obtain a number of tokens of a variable. A famous example of this methodology is Labov’s study of the presence or absence of /r/ in the speech of employees in several department stores in New York city. He gathered hundred of tokens by asking for directions to a product that he knew was sold on the fourth floor. Whenever someone directed him to ‘The fourth floor’, he noted carefully whether or not they pronounced the /r/ in either fourth or floor.
Recently, sociolinguists have begun to explore even more creative methods for exploring the complex relationships between social and linguistic factors and variation and change. The methods of social dialectology focus exclusively on the production of language: what variants do speakers use in different (social or linguistic) contexts? Some works has begun to also ask whether social or linguistic factors have an effect on the way speakers perceive language, e.g., do people hear something different if they think they are listening to speakers with different social attributes? Work in this area suggests that this may indeed be the case.