Constraints on production
Texts differ along the parameters set out elsewhere. To have them produced orally (and tape-record them for further processing) usually serves the purpose of getting data that represent a certain genre or the linguistic solution to certain cognitive and communicative problems. Depending on the purpose, constraints on the production (thus, on the speaker) may differ in strictness; it may be spontaneous or controlled (Himmelmann 1998). At one end of the continuum, there is the microphone (and videocamera) hidden in a room which over some time-span records the spontaneous linguistic behavior of the persons present in the room.1 At the other end, there is the linguist sitting in front of the informant, reading out each item on a list in turn, while it is the informant's task to produce the equivalent in his native language. Between these poles, there are stimuli like “Would you tell me the story of your life?”, “Explain how a traditioanl house is built!” or “Given such and such a situation, how would you say: ‘Leave that alone!’?”
The linguist may choose the degree of control he wants to exert, and may actually exert such control, which is the easier if he is guiding the production of just one informant. If the linguist does not speak the informant's language, lack of interaction tends to lead to the production of monologues. These are certainly a legitimate text genre in human societies – e.g. in story telling – and have the methodological advantage of being easy to process and analyze; but they have special properties that are not representative of everyday communication. If the aim is to obtain data of dialogical or even polylogical communication, the linguist must ask more informants to interact with each other. That entails for him to give away control of the situation to some extent. As a result, the data obtained are more variable, often not directly relevant to the linguist's epistemic interest, and usually hard to process.
Reflection on production
Just like other human actions, linguistic behavior and activity may be more or less conscious. Consciousness involves reflection on what one is doing. This reflection may be recursive, and it may be simultaneous, anterior or posterior to the action in question. And just as in non-linguistic life, a certain act may come out different if it is spontaneous than if it is subject to reflection. To give an example of posterior reflection: Telling a story may be relatively spontaneous (it may also be entirely planned or conventional). The text thus recorded will contain many sentences that do not please the speaker when he hears them played back. He will tend to correct false starts, slips of the tongue, grammatical mistakes and even his choice of words. In such a case, the linguist may assume that the product of such an emendation corresponds more closely to the speaker's own intentions and will generally accept it as a genuine piece of (more reflected) linguistic behavior which may even prove more useful for his epistemic interest than the original recording. Anyway, the linguist will not therefore discard the original recording, since it bears a – possibly interesting – paradigmatic relation to the amended version.
In a – more or less structured – informant work session, the speaker has much opportunity to reflect on his responses and to shape them according to what he thinks is expected from him and what is compatible with his self-image. Similarly, in an experiment situation, subjects will produce linguistic behavior adapted to experiment situations in general, or will produce erratic behavior because they never found themselves in such a situation before. Many different considerations will influence their linguistic behavior to the extent that it is not representative of what they would do spontaneously in an otherwise analogous situation. Just as in the case of text emendation, this interference from reflection does not render the data produced useless. However, the analyst must be aware that such data were produced under special conditions and must compare them with spontaneous data in order to make sure to what extent they are representative of general linguistic usage.
1 Needless to say, the usual research ethics apply.