Interactive repair and the foundations of language

Cover page of "Interactive repair and the foundations of language"

We have a new paper out in which we argue that the robustness and flexibility of human language is underpinned by a machinery of interactive repair. Repair is normally thought of as a kind of remedial procedure: a system for handling clarification questions, just one of those things we need to stay out of communicative trouble. Simply put (and oversimplifying only a bit), we argue we wouldn’t have complex language if it weren’t for this system of interactive repair.

We do a bunch of things in this paper: we review recent work on repair; propose it is a promising phenomenon to study in the context of comparative cognition, language development, collective computation, and interactive interfaces; and relate it to other core properties of language (reflexivity, accountability) and other elements of interactional infrastructure (turn-taking).

Anatomy and frequency of interactive repair

We start with the basics. Some facts are well-known, from the three-part structure of a repair sequence (panel A) to the high frequency of interactive repair (every 84s on average across a worldwide sample of 12 spoken and signed languages, panel D). The figure also highlights two other interesting facets of repair. First, an asymmetry in ‘interactional logic’, where interactive repair is the only kind of response in interaction that is always a possibility (an observation due originally to Harvey Sacks). Don’t try this at home. Or actually do — if you’ve ever been a kid, you know. The asymmetry is replicated once you’ve gone for repair, where you can always go ‘Huh?’ or similar, but to ask a more specific clarification question requires more work (and requires that you got at least part of the prior turn). In prior work we’ve shown that there is a general cross-cultural preference for doing this extra work, i.e. go for more specific if you can, and only go for ‘Huh?’ if you must. We point out some of the interesting features that follow from this, concluding that repair is always available, but need not always be invoked.

Four-part panel titled "Anatomy and frequency of interactive repair". Caption in the paper: "(A) With interactive repair, another participant initiates repair, inviting a repair solution by the first; the repair initiation is a pivot, pointing both back and forward. (B) While a fitted response is preferred, initiating repair is always a possible next move; likewise, within repair, while a restricted format is preferred, an open format is always an option. (C) Across diverse languages, formats for interactive repair range fall into three types, depending on how they target the trouble in prior turn and the kind of response they typically invite; these can be ranked from less to more specific in terms of the grasp of the trouble source they display. (D) Empirical cumulative distribution of independent repair sequences (black curve) as they occur over time in informal conversation in a global sample of 12 languages (grey curves). Across languages, the steepest part of the slope is around 17 s, the average 84 s, and nearly all sequences occur within a 4-min window from the last"

The second fact worth knowing about interactive repair is its sheer syntactic sophistication. In some presentations I’ve used a kind of ‘searchlight’ animation showing the ways in which repair formats zoom in on different parts of clausal structure; panel C is a still version of this, showing how repair formats provide a window onto turn structure and syntax. This is important, because we’re exposed to repair from very early on (see below), and so it is one of the key processes through which children are able to pick up the intricacies of grammatical structure.

animation of repair operations on a simple utterance


I said above that it’s oversimplifying a bit to state that we wouldn’t have language if it weren’t for repair. The oversimplification is that interactive repair stands in a dialectical relation to other core properties of language. By dialectical we mean back-and-forth causal relations by which things are co-constitutive and can become more than just the sum of their parts.

A key dialectical relation we identify is between interactive repair and reflexivity: the remarkable power of human language to refer to itself. We can use language to talk about language. This is good, because otherwise I’d be out of a job (like most if not all of us). Our argument is that the reflexivity of language is not just something that allows you to make nice recursive sentences like this one (although that, or rather this, too) — it is a property of our communication system that gains much of its adaptive value from how repair both exhibits and exploits it.

That double verb —exhibit and exploit— is an expression of the dialectical relation. Yes, repair exploits reflexivity: when I repeat part of your utterance with questioning intonation I use the language system to point to that bit you just spoke, inviting you to elaborate or clarify. But at the same time it exhibits it, puts reflexivity on display. This is especially clear in early language development, where studies show children regularly run into fairly intricate repair sequences. This exposes children from early on (perhaps before they’ve had much use of reflexivity in their own nascent language system) to the notion that you can, indeed, use language to refer to itself.

The exchanges below (Figure 2 in the paper), from work by Taylor & van den Herik (2021), show how this can work. In the exchanges below, Father’s responses to Margaret show that one can use the language system to point at itself. We don’t often think of it this way, but for an infant this is a potentially revolutionary insight. (Juliette Corrin (2010) has called this “the developmental power of ‘hm?’.) Anecdotal evidence of this revolutionary nature comes from my corpus data from Siwu and from my own kids: in both cases, at around age 2, I have examples of kids suddenly grasping that a sound like ‘Huh?’ can elicit a repetition of a prior turn, and once they get this, it becomes a really nice game for them to play. This is them discovering a key tool in our interactional toolkit and tinkering with it to see how it holds up and what they can do with it. (Later come the incessant why-questions — another example of the same discovery procedure.)

Panel showing two sequences of interactive repair in child-caregiver interaction. In A, the scene has Margaret bringing a cup to her Father in the kitchen 
M [indecipherable] 
F What did you say? 
M Want more milkie! 
F Momma said that was enough. 
M No! More! 

Panel B has Margaret in front of a DVD player. 
M Kitty-cat. Kitty-cat. 
F You mean you want to see the kitty-cat movie? 
M Yah, yah. 
F Okay. But you just watched it this morning.


Another core property of language we consider in relation to interactive repair is that of social accountability. A simple example is the first exchange above, where the child, Margaret, says something that was (to the transcriber) indecipherable, and her Father then initiates repair: “What did you say?”. This question does not just invite a redoing of Margaret’s prior turn; it also conveys to her that she is accountable for, among other things, articulating clearly.

The second exchange adds another layer of complexity: in saying “You mean you want to see the Kitty-cat movie?”, Father not only conveys there was something in need of fixing about Margaret’s initial utterance; he also demonstrates how one can turn a simple utterance into a more expansive version of itself that more explicitly features a social action, a process of enrichment that makes visible the inferential processes people may apply to each other’s talk. Importantly, in both cases, this work of holding the other accountable and making private commitments public is done through a sequence of interactive repair. Equally importantly, these are entirely mundane occurrences, not special didactic sequences. Just by doing its everyday work, the repair machinery makes reflexivity and accountability visible, for the early language learner just as well as for the analyst.

Accountability is a very rich topic across the fields of social interaction and social cognition, crucial to work on joint commitments, virtual bargaining, and the argumentative theory of reasoning, to mention just three disparate fields. In our paper we point out that social accountability it is deeply and fundamentally intertwined with the machinery for interactive repair. Far more than just a reactive quality-control system, interactive repair, as we write, “offers people opportunities at every turn to review, revise, and recalibrate around meaning in real time, editing not just the flow of information but the interpersonal commitments that the exchange of information creates” (p 1).

More in the paper

In this post I wanted to highlight only some of the points on repair’s relation to reflexivity and accountability. The paper also offers:

  1. A box Advances in conversation analytic work on repair, reviewing exciting CA work roughly from the past 5 years
  2. A box Turn-taking: broad and narrow senses, in which we propose a little more precision in cross-species discussions of “turn-taking”
  3. A discussion of possible evolutionary precursors of repair, and how notions of joint commitment and accountability may need to figure in this
  4. A proposal to view repair as a form of collective computation (see Figure below, which I am not going to explain here — it is based on this paper of ours)
  5. A review of the relevance of repair to interactive interfaces, whether text-based (as in LLMs) or voice-driven (as in Alexa and the like).
  6. A list of outstanding questions (a term the venue insists upon; I would have said “open”)
Three-panel figure depicting agent-based simulations and complexity analysis comparing repairers (interactive repair-capable agents, without reasoning) and reasoners (pragmatic reasoning-capable agents without repair) across three lexicon sizes (shown here in three shades) with moderate ambiguity. 

Panels show:
(A) Reasoners suffer decreasing communicative success with growing lexicon size. (B) Computational complexity (counted in terms of the number of computation steps, which includes the cost of additional turns for repairers) rises quadratically with lexicon size for reasoners and only linearly for repairers. (C) Additional turns for repairers lead to rapid reductions in uncertainty.

The paper

Here’s the reference, along with a link to the PDF:

Dingemanse, Mark, and N. J. Enfield. 2024. ‘Interactive Repair and the Foundations of Language’. Trends in Cognitive Sciences. (PDF)

1 thought on “Interactive repair and the foundations of language”

Leave a Reply

Your email address will not be published. Required fields are marked *