Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2022 May 1.
Published in final edited form as: Cognition. 2021 Feb 17;210:104577. doi: 10.1016/j.cognition.2020.104577

Cascading activation in phonological planning and articulation: Evidence from naturalistic speech errors

John Alderete 1,*, Melissa Baese-Berk 2, Keith Leung 1, Matt Goldrick 3
PMCID: PMC8009837  NIHMSID: NIHMS1662585  PMID: 33609911

Abstract

Speaking involves both retrieving the sounds of a word (phonological planning) and realizing these selected sounds in fluid speech (articulation). Recent phonetic research on speech errors has argued that multiple candidate sounds in phonological planning can influence articulation because the pronunciation of mis-selected error sounds is slightly skewed toward unselected target sounds. Yet research to date has only examined these phonetic distortions in experimentally-elicited errors, leaving doubt as to whether they reflect tendencies in spontaneous speech. Here, we analyzed the pronunciation of speech errors of English-speaking adults in natural conversations relative to matched correct words by the same speakers, and found the conjectured phonetic distortions. Comparison of these data with a larger set of experimentally-elicited errors failed to reveal significant differences between the two types of errors. These findings provide ecologically-valid data supporting models that allow for information about multiple planning representations to simultaneously influence speech articulation.

Keywords: speech production, speech errors, cascading activation, phonological encoding, articulation, phonetics

1. Introduction

Modern psycholinguistics understands speaking as a system of distinct processes that combine to produce fluid speech from an underlying message. In particular, single word productions involve mapping intended concepts to words (lexical selection), specifying the sounds of these words (phonological planning), and executing these sounds as motor actions (articulation). Though distinct, these production processes may interact with each other in interesting ways. In speech errors, for example, so-called mixed errors like /cat/ → rat involve the mis-selection of sounds in phonological planning (r for k), but they are assumed to arise from the activations of both target (cat) and competitor (rat) words in lexical selection (Harley, 1984; Martin, Gagnon, Schwartz, Dell, & Saffran, 1996). Likewise, picture-naming studies have shown that competitors that are activated in lexical selection affect phonological planning, reducing response latencies in words that overlap phonologically with competitors relative to words that do not (Peterson & Savoy, 1998). These examples demonstrate cascading activation, a kind of interaction in which multiple candidate representations activated in one process can influence outcomes in a process downstream (Dell, 1986; Melinger, Branigan, & Pickering, 2014).

While the evidence for cascading activation has tended to focus on interactions internal to speech planning, the phonetic properties of speech errors provide an intriguing form of evidence for cascades. If activation from multiple candidate representations in phonological planning does cascade and influence articulatory processes, articulations should show a blend of properties from multiple sounds. For example, as illustrated in Figure 1, in a speech error like /patch/ → bbatch, cascading activation from the partially activated target sound /p/ pulls the intruding b sound towards the intended p sound, distorting its pronunciation relative to a correct /batch/ → batch production. This predicted ‘trace’ of intended sounds1 has been borne out in a number of studies using both acoustic and articulatory methods (e.g., Frisch & Wright, 2002; Goldrick & Blumstein, 2006; Goldstein, Pouplier, Chena, Saltzman, & Byrd, 2007; McMillan & Corley, 2010; Pouplier, 2007; see Goldrick, Keshet, Gustafson, Heller, & Needle, 2016 for review). Related research has shown these phonetic traces can also arise from competition among morphologically related forms (Ernestus & Baayen, 2006; cf. Seyfarth, Vander Klok, & Garellek, 2019) and explicitly primed competitor forms (Yuen, Davis, Brysbaert, & Rastle, 2010).

Figure 1.

Figure 1

Open in a new tab

a. Waveform showing voice-onset time (VOT) for intruder sound b in speech error Baby eye [b]atch (Simon Fraser University Speech Error Database – English, record ID13005).

b. Visualization of cascading activation from the mis-selected error sound b in double circle and intended sound p of intended word patch. The selected phoneme b sends a large amount of activation to the glottal gesture consistent with voiced stops, yet the weakly-activated competitor p also sends a smaller activation signal to its corresponding gesture – resulting in the intermediate articulation between the two categories (see Goldrick & Chu, 2014 for simulation of this effect using gestural articulations from Browman & Goldstein, 1992).

This new account departs radically from past discrete models of language production that systematically exclude cascading activation by encapsulating production processes (Garrett, 1980; Levelt, Roelofs, & Meyer, 1999). Do these data provide sufficient evidence for rejecting such a model? While traces have only been documented in controlled experiments in the laboratory, it has been argued that these experiments create conditions similar to the ones leading to naturalistic speech errors (Baars, 1992), and naturalistic and induced errors pattern alike in sound, morphological, and lexical-syntactic structure (Stemberger, 1992). On the other hand, some researchers have questioned the validity of errors drawn from error-inducing experimental paradigms and their relevance to theories of production (Katsika, Shattuck-Hufnagel, Mooshammer, Tiede, & Goldstein, 2014; Keyser & Stevens, 2006). Given the theoretical importance of trace effects, we address this issue by investigating trace effects in naturally occurring speech errors. We compare these errors to a large corpus of experimentally-induced errors to determine if there are systematic differences in errors occurring in the wild versus the lab.2

2. Methods

2.1. Materials – the spontaneous speech error corpus

The speech error data were drawn from SFUSED English, a large database of speech errors based on audio recordings of natural conversations (Alderete, 2019). The speech errors were collected by the first author and research assistants in his lab that had undergone a month of training in error analysis and detection. Once collected, each error was verified with audio backup and classified by the first author using a standard classification system (Stemberger, 1982/1985, 1993). Analysis of SFUSED English has demonstrated that its methodology is robust to many of the problems that have plagued prior research, including skewed distributions due to perceptual biases, poor sample coverage, and mistakes in transcription and classification (Alderete & Davies, 2019).

We analyzed errors of phonological planning in which a sound similar to the target sound was mis-selected, as in /push/ → bush. Specifically, building on prior work (Goldrick, Baker, Murphy, & Baese-Berk, 2011; Goldrick & Blumstein, 2006; Goldrick et al., 2016; McMillan & Corley, 2010), we investigated speech errors within pairs of stop consonants that contrast in phonetic voicing (b p, d t, and g k). Our initial search of the database produced 141 phonological errors with these pairings, but they occurred in a variety of prosodic environments known to affect stop voicing. In an effort to control for this variation, we restricted the errors to those occurring in the following prosodic contexts: phrase-medial, word-initial, and simple onsets (i.e., pre-vocalic). These restrictions, and sampling for matched correct words (see below), reduced the original list to 86 errors from 36 talkers.

2.2. Materials – sampling matched words

The productions of stops in speech errors were compared against a set of matched correct words. When selecting these control words, we focused on controlling for the phonetic measure of interest, Voice Onset Time (VOT), the duration between the stop burst and the onset of periodicity in the following vowel (see Goldrick & Blumstein, 2006 for application to speech error analysis). VOT is the primary cue to voicing in English, and the primary measure for detecting trace effects in prior studies. By using spontaneous speech, we address concerns about the ecological validity of the data. In doing so, however, we also open ourselves up to the influence of possible confounding factors of phonetic context (e.g., position within the phrase) on VOT. To select properties to control for when matching errors to correct words, we conducted a pilot probe of production data from two speakers in our corpus, examining the influence of factors on VOT in previous phonetics studies (Davidson, 2011; Lisker & Abramson, 1964, 1967). Based on the results of this probe (included as supplementary material), correct words were selected so as to match the error in: prosodic frame (phrase-medial, word initial, simple onset); stress level of following vowel; and sonorancy of preceding segment (obstruent vs. sonorant). All correct tokens were screened for special intonation and disfluencies.

The matched correct words were drawn in semi-random fashion from the same corpus of audio recordings that the speech errors were extracted from. Transcripts of these recordings were generated using the IBM Watson Speech-to-Text service, and an index was created with the forced aligner Gentle v0.10.1 (Ochshom & Hawkins, 2019). Using the index, the recording was searched from the beginning (after initial scripted material) for words matching a given error’s properties, and after finding a match, skipping to the next minute marker to resume the search. This process was iterated with the goal of identifying 20 matched words. Restarting at minute markers enabled us to randomize the selection of words in a timely fashion, while still collect a sufficient number of correct productions. Because a given talker’s speech might be infrequent in the recording, the target sound structure could be rare (e.g., unstressed initial syllables), or the preceding segment class was challenging to match, we were only able to identify a mean of 10.4 (s.d. 3.58) matched correct productions for each error production. This sampling procedure produced 739 correct words matched to 86 errors.

2.3. Materials – the experimentally-induced speech error corpus

To compare these spontaneous errors with previous experimental work, we reanalyzed VOT data from two larger-scale studies of tongue twisters (Goldrick et al., 2011: n = 10 participants; Goldrick et al., 2016: n = 34; we only analyzed data that met inclusion criteria for each study). Both experiments involved tongue twisters designed to induce errors on word-initial consonants contrasting in voicing (e.g., when producing pin bin bin pin rapidly, participants make many errors swapping b and p). VOTs were automatically categorized3 as voiced vs. voiceless by fitting a two-component mixture model at each place of articulation for all of the productions by each participant. Errors were productions that were more likely to come from the incorrect versus correct mixture (see Goldrick et al., 2011). Except the very first syllable produced in a trial (which was excluded), all of the productions occurred adjacent to identical sounds, allowing errors and correct productions to be matched for phonetic context. (Inconsistent prosody across participants/trials made it infeasible to match along this dimension.) We also excluded cases where the number of errors produced by a speaker on a given sound exceeded the number of correct productions. This yielded 69,054 correct productions matched to 9,963 errors.

2.4. Data analysis

All of the data and statistical analysis scripts can be accessed at https://osf.io/7mf2p/. To determine if traces were present, we examined the absolute and relative differences between the mean VOTs of errors and correct productions of the stop consonant. As speech errors and correct productions sometimes occurred within the same context for the same speaker, they were grouped into sets defined by speaker, produced sound, and phonetic context (as defined above), yielding 71 distinct sets for spontaneous errors and 259 sets for experimental errors. The trace within each set was calculated by the mean trace size (mean VOT of errors within a set – mean VOT of correct productions in the same set) and mean trace ratio (error mean VOT / correct mean VOT). A bootstrap procedure (Goldrick et al., 2016) was then used to estimate 95% confidence intervals for these statistics. Error-correct word sets were randomly re-sampled with replacement, with an equally-sized sample drawn across error and correct productions. For experimental errors, each re-sampling used a random subset of the available sets, with the size of the subset equal to the number of spontaneous errors sets. Means, mean trace size, and mean trace ratio were calculated within each random re-sampling. This procedure was repeated 10,000 times; confidence intervals were estimated by the lower 2.5th and upper 97.5th percentiles.

3. Results

For both spontaneous and experimental productions, the bootstrapped distributions of error and matched correct words exhibited the hypothesized trace effect. Figure 2a shows the results for voiced stops (b d g) and Figure 2b the results for voiceless stops (p t k). As hypothesized, based on results reviewed in the introduction, the VOT of voiced stops in errors was slightly larger than matched stops in correct words. This held for spontaneous productions (error mean 14.0 msec vs. correct mean 12.0 ms; mean trace size 2.0 ms, mean trace ratio 1.17, one-tailed p < .03) and experimental productions (error mean 25.3 msec vs. correct mean 21.1 ms; mean trace size 4.1 ms, mean trace ratio 1.20, one-tailed p < .0001). Conversely, VOT in voiceless stops was slightly lower in errors. This held for spontaneous productions (error mean 51.2 msec vs. correct mean 58.2 ms; mean trace size −7.0 ms, mean trace ratio 0.88, one-tailed p < .02) and experimental productions (error mean 54.3 msec vs. correct mean 65.4 ms; mean trace size −11.1 ms, mean trace ratio 0.83, one-tailed p < .0001).

Figure 2.

Figure 2

Figure 2

Open in a new tab

a. Phonetic properties of spontaneous (left panel) and experimentally-induced (right panel) productions of voiceless → voiced errors and matched correct voiced productions. Grey points show values for individual sets of matched productions (lines connect correct and error means for the same set). Red points and lines show mean across sets. Center panel: Mean trace size and trace ratio for spontaneous and experimentally-induced errors. Error bars show bootstrapped 95% confidence intervals.

b. Phonetic properties of spontaneous (left panel) and experimentally-induced (right panel) productions of voiced → voiceless errors and matched correct voiceless productions. Grey points show values for individual sets of matched productions (lines connect correct and error means for the same set). Red points and lines show mean across sets. Center panel: Mean trace size and trace ratio for spontaneous and experimentally-induced errors. Error bars show bootstrapped 95% confidence intervals.

As shown by the means in the center panels of Figures 2a and 2b, the magnitude of the traces in spontaneous errors tended to be numerically weaker than experimental errors in absolute and relative terms for both voiced and voiceless stops. However, there was considerable overlap in the confidence intervals (one-tailed test, voiced spontaneous trace ratio < experimental: p < .39; voiceless spontaneous ratio > experimental: p < .62). We therefore fail to reject the null hypothesis of no difference between the groups.

4. Discussion

Models of language production have formalized phonological planning and articulatory processing as distinct production processes, but the nature of their interaction is controversial. By documenting phonetic traces in naturalistic speech errors, we have strengthened the case for allowing partial activation of target phonological representations to cascade to articulation (Goldrick & Blumstein, 2006), a phenomenon which to date has only been explored in the laboratory. Our findings also address a common concern that speech errors induced in experiments do not reflect the reality of natural speech (Keyser & Stevens, 2006). Our naturalistic speech errors pattern with induced errors, supporting a broader comparison of the two methods (Stemberger, 1992). Finally, we have described a new method of comparing naturalistic speech errors with appropriate baseline data, including techniques for sampling words without errors, identifying which factors matter for sampling, and analyzing differences between error and correct words. The success of our methods demonstrates the possibility of probing rather subtle effects with naturalistic data, consistent with renewed efforts to document language processing in ecologically-valid contexts (Speed, Wnuk, & Majid, 2018).

Our findings suggest that models of speech production require a mechanism for speech errors that involves the co-production of both target and error sounds. Interactive models of speech production can incorporate such a mechanism: cascading activation from partially activated phonological categories to articulatory processing, combined with non-discrete selection mechanisms (Figure 1). Model simulations of cascading activation, where target and error sounds are co-activated and send an activation signal to phonetic features, have documented quantitatively how these core assumptions produce differences between correct and error sounds in both voicing and place of articulation errors (Goldrick & Chu, 2014; Smolensky, Goldrick, & Mathis, 2014; see also Tilsen, 2019 for a co-activation proposal in the Articulatory Phonology framework). On the other hand, models lacking such a mechanism do not predict systematic differences between error and intended sounds. For example, WEAVER++ (Levelt et al., 1999; Roelofs, 1992) encapsulates the selection of segments from the creation of syllable program nodes (pre-compiled syllables associated with articulatory programs). As a result, the articulatory programs that arise from incorrectly selected segments are predicted to be indistinguishable from the programs retrieved for error-free words.

In addition to predicting the existence of trace effects, models with cascading activation also predict the relatively small magnitude of these effects. This is a consequence of selection mechanisms that greatly enhance the activation of representations that are selected for later processing relative to competitor forms (Rapp & Goldrick, 2000). In the case of traces, this assumption predicts that the selected error sound will dominate processing, resulting in relatively small tilt towards the unselected sound (Goldrick & Chu, 2014).

In the context of experimental tasks with repetitive speech, traces can also arise within articulatory processes. Across many motor domains, alternating between different movements (lift left hand, lift right hand, repeat) is inherently less dynamically stable than repeating synchronous actions (lift both hands, repeat). Faster rates for alternating actions can lead to spontaneous shift to synchronous movements (Haken, Peper, Beek, & Daffertshofer, 1996). Many speech error induction experiments involve producing sequences of alternating movements (e.g., movements associated with /b/ and /p/ in the tongue twister big pig pig big). Producing such sequences at fast rates could, as in other motor domains, yield synchronized productions (e.g., simultaneous production of intruding /b/ and intended /p/) – a trace effect (Goldstein et al., 2007; Pouplier, 2007). Along with cascading activation, this mechanism likely contributes to trace effects observed in tongue-twister style tasks (Goldrick et al., 2016). Does it contribute to errors in spontaneous speech as well? While the data in the current study show that blended speech errors can also occur in spontaneous speech that does not involve repetitive productions, other work has shown that repetition significantly influences spontaneous errors (Stemberger, 2009). Understanding the contribution from each mechanism – cascading activation and oscillatory planning dynamics – will therefore involve investigating phonetic distortions in spontaneous speech errors across a wider range of contexts.

While spontaneous and experimentally-induced errors show similar qualitative patterns, our results suggest there may be subtle quantitative differences in the magnitude of effects. This could reflect the high degree of phonetic variation in spontaneous speech (which makes matching error and correct productions difficult). Alternatively, it could reflect differences in speech planning processes in naturalistic, connected speech versus artificial tongue twisters, which, as discussed above, may be affected differently by cascades and oscillatory dynamics.

More generally, rigorous investigation of these questions will require larger sample sizes. Building on work with experimental tasks (Goldrick et al., 2016), such research will likely be driven by advances in automated analysis of speech. Adaptation of existing methods to the highly variable nature of speech produced in diverse contexts by diverse individuals is a key area for methodological development.

Supplementary Material

1

Acknowledgements

We would like to thank Jane Li for assistance with data collection and the audience at the 61st annual meeting of the Psychonomics Society for comments and questions. This work was supported in part by a Social Science and Humanities Research Council of Canada grant (435-2014-0452), a National Science Foundation grant (BCS0846147), and a National Institutes of Health grant (HD077140).

Footnotes

Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

1

Traces may also arise due to the dynamic instability of alternating movements (e.g., bad patch) versus repeated synchronous actions (pad patch) – a spontaneous transition that can yield co-production of two speech sounds (Goldstein et al., 2007; Pouplier, 2007). We return to this mechanism in the general discussion.

2

Our emphasis on spontaneously produced speech errors is not intended to reinforce a false dichotomy between naturalistic speech and lab speech, and de-value the latter. Rather, we seek to address a concern that experimentally-induced errors are constrained in ways (e.g., fixed prosodic frames, controlled conceptual content, artificial prompts and cut-offs, and extensive use of repetition and nonce words) that may produce phonetic effects that are not observed in speech that lacks these constraints.

3

Although differences between these techniques have not been systematically examined, Goldrick and Blumstein (2006) used perceptual criteria to identify experimentally-induced speech errors. The mean traces sizes reported there are well-within the 95% confidence interval of errors reported here, suggesting substantial similarity between results obtained using the two methods.

References

  1. Alderete J (2019). Simon Fraser University Speech Error Database - English Beta, [www.sfu.ca/people/alderete/sfused.] [DOI] [PMC free article] [PubMed]
  2. Alderete J, & Davies M (2019). Investigating perceptual biases, data reliability, and data discovery in a methodology for collecting speech errors from audio recordings. Language and Speech 62, 281–317. [DOI] [PubMed] [Google Scholar]
  3. Baars BJ (1992). A dozen competing-plans techniques for inducing predictable slips in speech and action. In Baars BJ (Ed.), Experimental slips and human error: Exploring the architecture of volition (pp. 129–150). New York: Plenum Press. [Google Scholar]
  4. Browman C, & Goldstein L (1992). Articulatory phonology: An overview. Phonetica, 49, 155–180. [DOI] [PubMed] [Google Scholar]
  5. Davidson L (2011). Characteristics of stop releases in American English spontaneous speech. Speech Communication, 53, 1042–1058. [Google Scholar]
  6. Dell GS (1986). A spreading-activation theory of retrieval in sentence production. Psychological Review, 93, 283–321. [PubMed] [Google Scholar]
  7. Emestus M, & Baayen H (2006). The functionality of incomplete neutralization in Dutch: The case of past-tense formation. Laboratory Phonology, 8, 27–49. [Google Scholar]
  8. Frisch SA, & Wright R (2002). The phonetics of phonological speech errors: An acoustic analysis of slips of the tongue. Journal of Phonetics, 30, 139–162. [Google Scholar]
  9. Garrett M (1980). Levels of processing in sentence production. In Butterworth B (Ed.), Language production: Vol. I. Speech and talk (pp. 177–220). London: Academic Press. [Google Scholar]
  10. Goldrick M, Baker HR, Murphy A, & Baese-Berk M (2011). Interaction and representational integration: Evidence from speech errors. Cognition, 121, 58–72. [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Goldrick M, & Blumstein S (2006). Cascading activation from phonological planning to articulatory processes: Evidence from tongue twisters. Language and Cognitive Processes, 21, 649–683. [Google Scholar]
  12. Goldrick M, & Chu K (2014). Gradient co-activation and speech error articulation: Comment on Pouplier and Goldstein (2010). Language, Cognition and Neuroscience, 29, 452–458. [Google Scholar]
  13. Goldrick M, Keshet J, Gustafson E, Heller J, & Needle J (2016). Automatic analysis of slips of the tongue: Insights into the cognitive architecture of speech production. Cognition, 149, 31–39. [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Goldstein L, Pouplier M, Chena L, Saltzman EL, & Byrd D (2007). Dynamic action units slip in speech production errors. Cognition, 103, 386–412. [DOI] [PMC free article] [PubMed] [Google Scholar]
  15. Haken H, Peper CE, Beek PJ, & Daffertshofer A (1996). A model of phase transitions in human hand movements during multifrequency tapping. Physica D, 90, 179–196. [Google Scholar]
  16. Harley TA (1984). A critique of top-down independent level models of speech production: Evidence from non-plan-internal speech errors. Cognitive Science, 8, 191–219. [Google Scholar]
  17. IBM. IBM Cloud Speech to Text. Retrieved from https://cloud.ibm.com/catalog/services/speech-to-text [Google Scholar]
  18. Katsika A, Shattuck-Hufnagel S, Mooshammer C, Tiede M, & Goldstein L (2014). Effects of compatible versus competing rhythmic grouping on errors and timing variability in speech. Language and Speech, 57, 544–562. [DOI] [PubMed] [Google Scholar]
  19. Keyser SJ, & Stevens KN (2006). Enhancement and overlap in the speech chain. Language, 82, 33–63. [Google Scholar]
  20. Levelt WJM, Roelofs A, & Meyer AS (1999). A theory of lexical access in speech production. Behavioral and Brain Sciences, 22, 1–75. [DOI] [PubMed] [Google Scholar]
  21. Lisker L, & Abramson A (1964). A cross-language study of voicing in initial stops: Acoustical measurements. Word, 20, 384–422. [Google Scholar]
  22. Lisker L, & Abramson A (1967). Some effects of context on voice onset time in English stops. Language and Speech, 10, 1–28. [DOI] [PubMed] [Google Scholar]
  23. Martin N, Gagnon DA, Schwartz MF, Dell GS, & Saffran EM (1996). Phonological facilitation of semantic errors in normal and aphasic speakers. Language and Cognitive Processes, 11, 257– 282. [Google Scholar]
  24. McMillan CT, & Corley M (2010). Cascading influences on the production of speech: Evidence from articulation. Cognition, 117, 243–260. [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Melinger A, Branigan HP, & Pickering MJ (2014). Parallel processing in language production. Language, Cognition and Neuroscience, 29(6), 663–683. [Google Scholar]
  26. Ochshorn RM, & Hawkins M (2019). Gentle forced aligner v0.10.1. Retrieved from http://lowerquality.com/gentle/ [Google Scholar]
  27. Peterson RR, & Savoy P (1998). Lexical selection and phonological encoding during language production: Evidence for cascading processing. Journal of Experimental Psychology: Learning, Memory and Cognition, 24, 539–557. [Google Scholar]
  28. Pouplier M (2007). Tone kinematics during utterances elicited with the SLIP technique. Language and Speech, 50, 311–341. [DOI] [PubMed] [Google Scholar]
  29. Rapp B, & Goldrick M (2000). Discreteness and interactivity in spoken word production. Psychological Review, 107, 460–499. [DOI] [PubMed] [Google Scholar]
  30. Roelofs A (1992). A spreading-activation theory of lemma retrieval in speaking. Cognition, 42, 107–142. [DOI] [PubMed] [Google Scholar]
  31. Seyfarth S, Vander Klok J, & Garellek M (2019). Evidence against interactive effects on articulation in Javanese verb paradigms. Psychonomic Bulletin & Review, 26, 1690–1696. [DOI] [PubMed] [Google Scholar]
  32. Smolensky P, Goldrick M, & Mathis D (2014). Optimization and quantization in gradient symbol systems: A framework for integrating the continuous and the discrete in cognition. Cognitive Science, 38, 1107–1138. [DOI] [PubMed] [Google Scholar]
  33. Speed LJ, Wnuk E, & Majid A (2018). Studying psycholinguistics out of the lab. In de Groot AM & Hagoort P (Eds.), Research methods in psycholinguistics and the neurobiology of language (pp. 190–207). Oxford: Wiley Blackwell. [Google Scholar]
  34. Stemberger JP (1982/1985). The lexicon in a model of language production. New York: Garland. [Google Scholar]
  35. Stemberger JP (1992). The reliability and replicability of naturalistic speech error data. In Baars BJ (Ed.), Experimental slips and human error: Exploring the architecture of volition (pp. 195–215). New York: Plenum Press. [Google Scholar]
  36. Stemberger JP (1993). Spontaneous and evoked slips of the tongue. In Blanken G, Dittmann J, Grimm H, Marshall JC, & Wallesch C-W (Eds.), Linguistic disorders and pathologies. An international handbook (pp. 53–65). Berlin: Walter de Gruyter. [Google Scholar]
  37. Stemberger JP (2009). Preventing perseveration in language production. Language and Cognitive Processes, 24, 1431–1470. [Google Scholar]
  38. Tilsen S (2019). Motoric mechanisms for the emergence of non-local phonological patterns. Frontiers in Psychology, 10, 1–25. [DOI] [PMC free article] [PubMed] [Google Scholar]
  39. Yuen I, Davis MH, Brysbaert M, & Rastle K (2010). Activation of articulatory information in speech perception. Proceedings of the National Academy of Sciences of the United States of America, 107, 592–597. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

1
NIHMS1662585-supplement-1.pdf (150.1KB, pdf)

RESOURCES