SPACER: A Parallel Dataset of Speech Production And Comprehension of Error Repairs

TL;DR

SPACER provides a parallel dataset that links naturalistic speech errors and repairs in production with comprehension edits, enabling integrated study of error monitoring across both modalities. It combines Switchboard-derived single-word substitutions with a web-based correction task, yielding 1056 initial utterances (576 SC, 480 SU) and 5808 comprehender responses across 66 participants. Analyses reveal asymmetries: speakers tend to self-repair when semantic and phonemic deviations are large, whereas comprehenders tend to correct errors that are phonemically similar or contextually unsupported, suggesting complementary strategies. The dataset supports principled, computational investigations into rational-inference models of error correction and offers a resource to bridge production and comprehension research with broad applicability to language science.

Abstract

Speech errors are a natural part of communication, yet they rarely lead to complete communicative failure because both speakers and comprehenders can detect and correct errors. Although prior research has examined error monitoring and correction in production and comprehension separately, integrated investigation of both systems has been impeded by the scarcity of parallel data. In this study, we present SPACER, a parallel dataset that captures how naturalistic speech errors are corrected by both speakers and comprehenders. We focus on single-word substitution errors extracted from the Switchboard corpus, accompanied by speaker's self-repairs and comprehenders' responses from an offline text-editing experiment. Our exploratory analysis suggests asymmetries in error correction strategies: speakers are more likely to repair errors that introduce greater semantic and phonemic deviations, whereas comprehenders tend to correct errors that are phonemically similar to more plausible alternatives or do not fit into prior contexts. Our dataset enables future research on integrated approaches toward studying language production and comprehension.

Figures (11)

• Figure 1: An illustration of the dataset design. Suppose a speaker produces an initial utterance. Both the speaker and the comprehender may engage in error monitoring and correction processes, resulting in speaker final production and comprehender final response being either the same as the initial utterance (represented in gray bubbles) or different from it (represented in green bubbles). Each utterance is annotated by four to six comprehenders.
• Figure 2: Examples of speaker uncorrected (SU), speaker corrected (SC), comprehender uncorrected (CU), and comprehender corrected (CC) utterances in the SPACER dataset. Words highlighted in red were initially produced a speaker and later corrected. Words highlighted in green are corrections made by either the speaker in the original context or by a participant in the comprehension experiment. A grey highlight indicates that the word was not corrected by the participant. Note that for each SU and SC utterance, there may be up to four responses, which we classify as either CC or CU responses depending on whether or not the participant made a correction in their response.
• Figure 3: An illustration of comprehension experiment. A comprehender is presented with the key sentence together with preceding context. The comprehender is instructed to make necessary edits in the textbox and slide bar to indicate their confidence level.
• Figure 4: The POS categories of speaker-produced errors and corresponding repairs in speaker corrected utterances.
• Figure 5: The number of corrected and uncorrected responses across different part-of-speech (POS) categories of presented errors in the speaker corrected utterances.