The Bias Toward Single-Unit Turns in Conversation

ABSTRACT

Sacks, Schegloff, and Jefferson argued that the rules for turn taking for conversation involve a confluence of pressures that bias turn size toward single turn constructional units (TCUs), which leads to an empirical prediction that turns are more likely to be composed of single (vs. multiple) TCUs. We directly test and confirm this “single-TCU bias” by using conversation analysis, corpus linguistics, and Bayesian statistics to assess the conversational subcorpus of the British National Corpus (BNC-C), which contains 475,509 turns of talk. Our results confirm this bias, showing that 67% of turns are composed of single TCUs; we discuss why this estimate is conservative. The mean word length for single-TCU turns was 4.5 (SD = 3.4), compared to 19.9 (SD = 22.6) for multi-TCU turns. Our findings reinforce the ideas that the natural habitat for an accountable social action is the single TCU (vs. the turn), and that interaction is fundamentally organized (i.e., both produced and understood) on an action-by-action basis, which is a TCU-by-TCU basis. Data are in British English.

View correction statement:

Correction

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

¹ Schegloff further discussed this bias in several publications. For example, Schegloff (1982) discussed how it takes “work” to get more than one TCU into a turn. Subsequently, Schegloff (1987b) said: “The basic operation of the turn-taking system, in giving somebody a turn, allocates the right to produce a single turn-constructional unit—that is to say, a single lexical, phrasal, clausal or sentential construction” (p. 77). Later, Schegloff (1996) said: “There are interactional contingencies (in conversation) biasing turn size to one, and then few, TCUs” (p. 61).

² As Schegloff (1996) argued: “At every possible completion of a TCU, the turn-so-far will have amounted to—will be analyzable as—some possible action” (p. 58).

³ Levinson (2006a) observed that interaction’s “cyclic repair mechanism” (p. 88) “only works if turns are short” (p. 88), likening it (at least partially) to cognitive science’s notion of Test-Operate-Test-Exit (TOTE; Levinson, 2006b, p. 51).

⁴ Both Levinson and Torreira (2015) and Roberts et al. (2015) used the Switchboard corpus (Calhoun et al., 2010); Rühlemann (2020) used the BNC-C.

⁵ For example, see the tag question in Extract A at line 2666, which is the BNC-C transcript (The BNC symbol ‘<-|->’ represents overlap):Extract A: [KCV:2665–2667]However, in Extract A’, which is a CA re-transcription of Extract A: (a) Steven’s tag question, “Doesn’t it?” (line 5), is separated from his turn at line 3 by a micropause; (b) Katherine’s initial agreement, “Yeah,” (line 4)—which is not transcribed in the BNC-C transcript—treats Steven’s assessment turn at line 3 as being a possibly complete, discrete TCU; and (c) Katherine’s second agreement, “eYe:ah,” (line 6), treats Steven’s tag question (line 5) as a possibly complete, discrete TCU pursuing her agreement.Extract A’: It Looks Nice [KCV:2665–2667]

⁶ In response to a reviewer’s concern about Algorithm 5’s ability to adequately discriminate between single- and multi-TCU turns, we examined the transcript and, if possible, audiotape of a random sample of 500 turns (out of 30,341) that this algorithm identified as containing multiple TCUs. A total of 425 of these turns (85%) were multi-unit turns, and 75 (15%) were single-TCU turns. Although the proportions are different, this level of error was detected in our algorithm-validation exercise (discussed in the main text) and considered in our Bayesian analysis. As with Algorithm 3, Algorithm 5 makes our hypothesis test conservative because it artificially, slightly inflates the multi-TCU-turn category.

⁷ If tag questions as defined by Algorithm 3 are not considered to be TCUs—that is, if Algorithm 3 is not applied—the mean length of single-TCU turns is 4.6 words (Median = 4; Mode = 1; SD = 3.7).

⁸ More research is needed on the nature of solely nonvocal TCUs, but it is clear that they do exist. Other-initiation of repair can also be implemented solely nonvocally (Seo & Koshik, 2010). CA’s original, grammar-focused operationalization of TCUs has been progressively refined with the recognition that one of their definitional features is the accomplishment of social actions, which does not require vocality (Mondada, 2021).