It is normal for a spontaneously produced text to contain violations of rules of the language system, be they considered mistakes or errors or lapses. A linguist whose goal is the description of the language system wants to cut these out. The following considerations guide this endeavor.

Orality vs. literacy

A principal prior decision concerns the alternative between a free oral discourse and a written text:

• If the research interest is in a faithful representation of oral communication in the speech community, few or no emendations of recordings are appropriate. The transcription may then follow the conventions of communication analysis (e.g. Barras et al. 2001). Such products may, of course, be highly deficient if compared to the linguistic norm.
• If the research interest is in the system underlying speech and texts considered exemplary by members of the speech community, then written texts enjoy priority. If written communication is not established in the speech community, a method often employed is the following:
  1. A discourse of a certain genre – typically, a narrative – is produced by a competent speaker and recorded.
  2. The recording is transcribed.
  3. The transcription is heavily edited – typically in cooperation with the speaker himself – to conform to the speaker's norm.
  Such products may, of course, bear little resemblance to daily communication in the community.

When publishing a text as part of the description of a language, it is therefore indispensable to describe its production.

Editing

The author of the text himself, if confronted with his product, may find fault with it and state that it was not what he wanted to say. Then a distinction applies between a content error and a linguistic mistake. The linguist's interest is, of course, limited to amending the latter. Among such mistakes, a further distinction applies between lapsus linguae (or stilus) and violations of rules of grammar. Ignoring – i.e., correcting – the former normally goes by itself. The amendment does not even need to be noted in the text edition unless the analytic interest focuses just on lapsus linguae. Things are different with violations of rules of the linguistic system. The speaker may accidentally have formed a regular conjugation form of a verb whose conjugation is irregular, may have postposed to the end of the sentence a dependent which must be adjacent to its host, and so on. What acts in such cases is some kind of linguistic superego of the speaker. The correction of this kind of mistake follows the principles of text edition. In other words, the amendment is licit provided it is accompanied by a note quoting the original version. Later on, the linguist may choose which version to publish.

This last procedure is the more indispensable if it is another speaker of the language who finds fault with the product. If the two persons get a chance to discuss the problem, chances are that they agree on the correct version. If not, the linguist notes the version preferred by the second speaker as a variant.