Misspoken words affect the perception and retrieval of intended words

ABSTRACT

Everyday speech contains disfluencies, including unintentionally spoken words. What is the fate of the misspoken word in the comprehender’s memory? In this study, we asked whether: (1) the gender of misspoken words lingers and affects how the intended word is perceived, and (2) whether and how lingering representations can cause interference during the retrieval of the intended word. In two experiments, participants provided spoken responses to given prompts. In Experiment 1, participants used masculine or feminine pronouns to refer to gender-neutral words (passenger) depending on the gender of a preceding misspoken word (pilot or stewardess), suggesting that the gender of reparanda lingers. Experiment 2 showed that the presence of a misspoken word resulted in a reduction of pronominal reference to the intended word, suggesting that the misspoken word causes interference when the intended word is being retrieved by functioning as an additional discourse entity.

KEYWORDS:

• Disfluency
• Lingering memory representations
• Referential processing
• Good Enough language processing
• Perception
• Encoding
• Retrieval

Disclosure statement

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

Notes

1 There is a distinction between “accessibility” and “availability” of information in the literature, with the former referring to retrieval speed and the latter referring to probability of successful retrieval of the target memory item (e.g. McElree, 2000; McElree et al., 2003; Foraker & McElree, 2007; Van Dyke & McElree, 2011). However, the current paper is agnostic to this distinction. We therefore use the term “accessibility” throughout the paper and we use this term interchangeably with “ease of retrieval”.

2 Our use of forms that assume these gender stereotypes should in no way be taken to imply acceptance or endorsement of these gender stereotypes. In fact, we look forward to the day when this manipulation is ineffective.

3 It is important to note that we also calculated the degree of semantic similarity between male reparanda and the repairs as well as female reparanda and the repairs using Latent Semantic Analysis (http://lsa.colorado.edu/). We then ran a t-test on the similarity values and found no significant difference between the two conditions. Thus, any potential differences between the conditions could not be caused by the degree of semantic similarity between the conditions.

4 Most of our data for this analysis came from object, possessive, and reflexive (i.e. non-subject) pronouns.

5 Note that the low number of analyzable datapoints is caused by the nature of the task plus some technical issues, and does not invalidate the results. In an open-ended task like ours, participants are free to say whatever they wish and in whatever form they choose. Consequently, they may or may not use a “gender-revealing” pronoun. Note also that the total number of gendered pronouns was almost equally distributed across the three conditions (see below), indicating that the excluded trials are not biased based on condition.

6 The random effects for this model included random intercepts for both subjects and items, as well as by-subjects and by-items random slopes for the effect of Condition.

7 The random effects for this model also included random intercepts for both subjects and items, as well as by-subjects random slopes for the effect of Condition.

8 Note that Julie is not technically a repair in the Fluent condition, but we will call it a repair for the sake of simplicity.

9 Note that the total number of references to the first- and the second-mentioned words add up to 2420, whereas the total number of analyzable responses was 3334 (see above). The difference is due to 914 of analyzable responses which were analyzed as collective references (see below).

10 To examine the effect of syntactic role/serial position, we also ran a model predicting form of reference as a function of “Word Position”. The results revealed significantly greater probability of pronominal reference to the syntactic subject/the first-mentioned referential candidate than to the syntactic object/the second-mentioned referential candidate (β = −3.07, SE = 0.35, z = −8.59, p < .001).