A theory of repetition and retrieval in language production.

Abstract

Repetition appears to be part of error correction and action preparation in all domains that involve producing an action sequence. The present work contends that the ubiquity of repetition is due to its role in resolving a problem inherent to planning and retrieval of action sequences: the Problem of Retrieval. Repetitions occur when the production to perform next is not activated enough to be executed. Repetitions are helpful in this situation because the repeated action sequence activates the likely continuation. We model a corpus of natural speech using a recurrent network, with words as units of production. We show that repeated material makes upcoming words more predictable, especially when more than one word is repeated. Speakers are argued to produce multiword repetitions by using backward associations to reactivate recently produced words. The existence of multiword repetitions means that speakers must decide where to reinitiate execution from. We show that production restarts from words that have seldom occurred in a predictive precedingword context and have often occurred utterance-initially. These results are explained by competition between preceding-context and top-down cues over the course of language learning. The proposed theory improves on structural accounts of repetition disfluencies, and integrates repetition disfluencies in language production with repetitions observed in other domains of skilled action.

The competition–compensation account of developmental language disorder

“…We characterized the over‐extension of bare form to ‐ed contexts as a type of compensation, and demonstrated that the production of bare form was favored over other potential candidates such as the 3rd person present‐tense ‐s in generalization tests. Using an experimental paradigm that involved manipulating the accessibility of a form while keeping its frequency constant, Harmon and Kapatsinski (2017) demonstrated that when high frequency results in increase in accessibility of a form, speakers extend that form, as opposed to its competitor, to novel semantically related contexts (see also Harmon, 2019; Kapatsinski, 2018; Koranda et al., 2021; Kapatsinski, 2022). The high accessibility of the bare form, coupled with its semantic and form‐based similarity to its inflected past‐tense form due to stem overlap, leads to its repeated extension to past‐tense contexts, resulting in inconsistent inflectional marking (see also, Hoeffner & McClelland, 1993).…”

“…Reduced exposure to ‐ed hapaxes weakens the productivity of ‐ed . For a form to be productive, it must be parsed out of the contexts in which it has been experienced (Hay & Baayen, 2002) and must be accessible independently of those contexts (Harmon & Kapatsinski, 2021). While applying a suffix to hapax legomena requires parsing the suffix independently of the stem, its repeated application to a small number of stems results in chunking the suffix with those stems (e.g., Bybee & Brewer, 1980; Kapatsinski, 2010; Stemberger & MacWhinney, 1986).…”

Is the Mind Inherently Predicting? Exploring Forward and Backward Looking in Language Processing

“…Language users do appear to make use of backward looking in many experimental settings, for example, to fill in words they have missed in comprehension (Gwilliams, Linzen, Poeppel, & Marantz, 2018 ; Lieberman, 1963 ), to repeat words that help to access upcoming words (Harmon & Kapatsinski, 2021 ), or to fill in acoustically ambiguous words. For example, in the classic experiment of phoneme restoration by Warren ( 1970 ), subjects heard spoken sentences in which a word mid‐sentence had been partly covered by a cough (signaled here by a *), and thus was made acoustically ambiguous.…”

Morphology in a Parallel, Distributed, Interactive Architecture of Language Production

“…This assumption is likely a simplification but it allows us to focus on processing mechanisms in this paper without addressing learning mechanisms. Once planning is complete, we assume that learning updates associations from forms to meanings, increasing the weights of associations from the semantic features of the message to the chunks forming the constructed plan, and decreasing association weights to chunks that were not selected (see Kapatsinski, 2018a ; Baayen et al, 2019 ; Harmon and Kapatsinski, 2021 ; for possible mechanisms). Through learning, the system becomes and remains sensitive to the frequencies with which various semantic features co-occur with various forms in production.…”

“…Usage-based linguistics has argued that sublexical units emerge from generalization over experienced utterances and other units that can stand on their own, like words (Bybee, 1985 , 2001 ). From this perspective, sublexical units like -ism should be able to gain autonomy when they occur in a wide variety of contexts (Bybee, 1985 , 2001 ; Harmon and Kapatsinski, 2021 ). Consequently, examples in which units appear to gain autonomy despite occurring in a fixed context present a challenge to the theory.…”