---
title: "Effects of Duration, Locality, and Surprisal in Speech Disfluency"
description: "The paper asks whether two theories of comprehension difficulty, Surprisal Theory… 5.04% accuracy. 41 references, cited by 9 papers here."
authors: ["Samvit Dammalapati", "Rajakrishnan Rajkumar", "Sidharth Ranjan", "Sumeet Agarwal"]
year: 2021
venue: "Proceedings of the Society for Computation in Linguistics 2021"
source: "https://aclanthology.org/2021.scil-1.9/"
licence: "CC BY 4.0 (ACL Anthology)"
licence_url: "https://creativecommons.org/licenses/by/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/dammalapati-2021-effects-duration-locality-surprisal-speech/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/dammalapati-2021-effects-duration-locality-surprisal-speech/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2021](https://md.research.elamur.ai/free/en/papers/2021/page/6/) / Effects of Duration, Locality, and Surprisal in Speech Disfluency Prediction in English Spontaneous Speech

# Effects of Duration, Locality, and Surprisal in Speech Disfluency Prediction in English Spontaneous Speech

**Authors:** Samvit Dammalapati, [Rajakrishnan Rajkumar](https://md.research.elamur.ai/free/en/authors/rajakrishnan-rajkumar/), Sidharth Ranjan and [Sumeet Agarwal](https://md.research.elamur.ai/free/en/authors/sumeet-agarwal/) · **Year:** 2021 · **Venue:** [Proceedings of the Society for Computation in Linguistics 2021](https://md.research.elamur.ai/free/en/venues/scil/) ([SCiL 2021](https://md.research.elamur.ai/free/en/venues/scil/2021/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/2021.scil-1.9/) · [PDF](https://aclanthology.org/2021.scil-1.9.pdf) · [OpenAlex](https://openalex.org/W3133525917)

**Citations:** 5 citations (OpenAlex, 2026-10-14) · [9](https://md.research.elamur.ai/free/en/papers/dammalapati-2021-effects-duration-locality-surprisal-speech/cited-by/) in our library, of the 141,840 papers whose bibliographies we read

## Summary
The paper asks whether two theories of comprehension difficulty, Surprisal Theory and Dependency Locality Theory (DLT), can also predict where speakers become disfluent. It looks at two kinds of disfluency: fillers such as *uh* and *um*, and reparandums, the stretches a speaker abandons and then corrects. The data come from the Switchboard corpus of American English conversation. After removing disfluencies, the authors trained binary Generalized Linear Mixed Models that predict, for each word position, whether a disfluency occurs there. The predictors were lexical surprisal, syntactic surprisal, DLT integration cost, DLT storage cost and word duration, with a random intercept per word.

Disfluencies tend to sit between two kinds of words:
- **Before:** easy, predictable, mostly function words that the speaker lengthens.
- **After:** hard words with high surprisal and high locality cost.

The strongest predictors were the duration of the preceding word, the lexical surprisal of the following word, and storage cost. The authors read this as speakers stretching easy words to gain planning time, a buffer that sometimes fails. The work links psycholinguistic cost metrics to production and to disfluency modelling.

## Key points
- **Data:** Switchboard transcripts, with a separate binary model for fillers and for reparandums. Each model is fit in R with `lme4` as `disfluency ~ lexsurp + synsurp + IC + SC + duration + (1|word)`.
- **Lexical surprisal:** a trigram model trained with SRILM on the Open American National Corpus.
- **Syntactic surprisal:** a PCFG parser trained on Penn Treebank sections 2–21 through ModelBlocks, an incremental Berkeley-parser implementation.
- **DLT costs:** computed on Stanford dependency conversions of the disfluency-stripped constituency trees.
  - Integration cost sums the lengths of the backward links a word closes.
  - Storage cost counts the dependencies still waiting for a head.
- **Effect sizes:** lexical surprisal adds 5.04% accuracy for fillers. Storage cost adds 2.12% for reparandums, more than surprisal does there. Integration cost gives no notable gain.
- **Correlations:** lexical surprisal correlates 0.49 with duration, the highest of any feature. Surprisal and DLT measures correlate weakly, at most 0.151, so they are largely complementary.

## Interesting ideas
- Storage cost is a cheap, forward-looking difficulty score: the number of open dependencies at each word, read directly off a dependency tree. It predicted upcoming trouble better than backward-looking integration cost.
- Pairing a probabilistic signal (surprisal) with a structural one (locality) works well because the two barely correlate, so each catches difficulty the other misses.
- A random intercept per word keeps lexical identity from masquerading as an effect of the structural predictors.
- To get clean dependencies for conversational speech, the authors removed disfluencies from the constituency trees before converting them, not after.
- Disfluencies follow a recognisable signature: easy function words, often lengthened, followed by a hard, surprising word.

## For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

- **I1** (tree): The parser could compute open-dependency counts per token and use peaks as an explainable flag for spans where attachments are likely to go wrong and need a second look.
- **I2** (learn): When choosing silver sentences or deciding where witness parsers should be trusted, combine an LM-surprisal score with a tree-based locality score rather than relying on either alone.
- **I3** (learn): Evaluations of a feature's contribution could include per-lemma baselines as a control, so that claimed gains are not just memorised word identity.
- **I4** (tree): For any spoken-English treebank, such as Switchboard-style data, strip filler and reparandum nodes before UD conversion so the dialect converter never builds arcs to aborted fragments.
- **I5** (tokens): A transcript tokenizer could mark a "repair zone" when repeated function words run up to a high-surprisal token, letting later stages drop the reparandum with an explicit, traceable reason.

*Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.*

## Institutions

- [Indian Institute of Technology Delhi](https://md.research.elamur.ai/free/en/universities/indian-institute-of-technology-delhi/) — University, New Delhi, India · 10 works · 915 citations
- [Indian Institute of Science Education and Research, Bhopal](https://md.research.elamur.ai/free/en/universities/indian-institute-of-science-education-and-research-bhopal/) — University, Bhopal, India · 3 works · 20 citations

📚 Shelves (7): [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 9,296 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 9,206 works · [💭 Psychology](https://md.research.elamur.ai/free/en/shelves/psychology/) · 798 works · [💭 Experimental and Cognitive Psychology](https://md.research.elamur.ai/free/en/shelves/psychology/experimental-and-cognitive-psychology/) · 463 works · [💭 Developmental and Educational Psychology](https://md.research.elamur.ai/free/en/shelves/psychology/developmental-and-educational-psychology/) · 260 works · [🧠 Neuroscience](https://md.research.elamur.ai/free/en/shelves/neuroscience/) · 95 works · [🧠 Cognitive Neuroscience](https://md.research.elamur.ai/free/en/shelves/neuroscience/cognitive-neuroscience/) · 95 works

🧭 Topics (2): [English](https://md.research.elamur.ai/free/en/topics/english/) · 1,373 works · [🧒 Psycholinguistics & language acquisition](https://md.research.elamur.ai/free/en/topics/psycholinguistics/) · 66 works

## Cite

Cite as: Samvit Dammalapati, Rajakrishnan Rajkumar, Sidharth Ranjan and Sumeet Agarwal. 2021. Effects of Duration, Locality, and Surprisal in Speech Disfluency Prediction in English Spontaneous Speech. Proceedings of the Society for Computation in Linguistics 2021.

The paper itself, not this page. BibTeX from the library's record of the work.

```bibtex
@inproceedings{dammalapati2021effects,
  title = {{Effects of Duration, Locality, and Surprisal in Speech Disfluency Prediction in English Spontaneous Speech}},
  author = {Samvit Dammalapati, Rajakrishnan Rajkumar, Sidharth Ranjan and Sumeet Agarwal},
  booktitle = {Proceedings of the Society for Computation in Linguistics 2021},
  year = {2021},
  doi = {10.7275/s5yn-7d90},
  url = {https://doi.org/10.7275/s5yn-7d90}
}
```

- [← previous Current Trends and Approaches in Synonyms Extraction: Potential Adaptation to Arabic](https://md.research.elamur.ai/free/en/papers/naser-karajah-2022-current-trends-approaches-synonyms-extraction/)
- [next → Evaluating Text Generation from Discourse Representation Structures](https://md.research.elamur.ai/free/en/papers/wang-2021-evaluating-text-generation-discourse-representation/)
- [↑ back to papers of 2021](https://md.research.elamur.ai/free/en/papers/2021/page/6/)

## References

41 references, 3 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Jennifer E Arnold, Maria Fagnano, and Michael K Tanenhaus. 2003. Disfluencies signal theee, um, new information. Journal of psycholinguistic research, 32(1):25–36. [↗](https://doi.org/10.1023/a:1021980931292)
2. Vera Demberg and Frank Keller. 2008. Data from eye-tracking corpora as evidence for theories of syntactic processing complexity. Cognition, 109(2):193–210. [↗](https://doi.org/10.1016/j.cognition.2008.07.008)
3. Jennifer E Arnold, Anthony Losongco, Thomas Wasow, and Ryan Ginstrom. 2000. Heaviness vs. newness: The effects of structural complexity and discourse status on constituent ordering. Language, 76(1):28–55. [↗](https://doi.org/10.1353/lan.2000.0045)
4. Vera Demberg, Asad B. Sayeed, Philip J. Gorinski, and Nikolaos Engonopoulos. 2012. Syntactic surprisal affects spoken word duration in conversational contexts. In Proceedings of the 2012 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning, EMNLP-CoNLL ’12, pages 356–367, Stroudsburg, PA, USA. Association for Computational Linguistics. [find](https://scholar.google.com/scholar?q=Vera+Demberg%2C+Asad+B%2E+Sayeed%2C+Philip+J%2E+Gorinski%2C+and+Nikolaos+Engonopoulos%2E+2012%2E+Syntactic+surprisal+affects+spoken+word+duration+in+conversational+contexts%2E+In+Proceedings+of+the+2012+Joint+Confere)
5. Dale J Barr. 2001. Trouble in mind: Paralinguistic indices of effort and uncertainty in communication. Oralité et gestualité: Interactions et comportements multimodaux dans la communication, pages 597– 600. [find](https://scholar.google.com/scholar?q=Dale+J+Barr%2E+2001%2E+Trouble+in+mind%3A+Paralinguistic+indices+of+effort+and+uncertainty+in+communication%2E+Oralite%CC%81+et+gestualite%CC%81%3A+Interactions+et+comportements+multimodaux+dans+la+communication%2C+pages+5)
6. Neeraj Deshmukh, Aravind Ganapathiraju, Andi Gleeson, Jonathan Hamaker, and Joseph Picone. 1998. Resegmentation of switchboard. In Fifth international conference on spoken language processing. [find](https://scholar.google.com/scholar?q=Neeraj+Deshmukh%2C+Aravind+Ganapathiraju%2C+Andi+Gleeson%2C+Jonathan+Hamaker%2C+and+Joseph+Picone%2E+1998%2E+Resegmentation+of+switchboard%2E+In+Fifth+international+conference+on+spoken+language+processing%2E)
7. Alan Bell, Jason M Brenier, Michelle Gregory, Cynthia Girand, and Dan Jurafsky. 2009. Predictability effects on durations of content and function words in conversational english. Journal of Memory and Language, 60(1):92–111. [↗](https://doi.org/10.1016/j.jml.2008.06.003)
8. Merrill Garrett. 1980. Levels of processing in sentence production. In Language production Vol. 1: Speech and talk, pages 177–220. Academic Press. [find](https://scholar.google.com/scholar?q=Merrill+Garrett%2E+1980%2E+Levels+of+processing+in+sentence+production%2E+In+Language+production+Vol%2E+1%3A+Speech+and+talk%2C+pages+177%E2%80%93220%2E+Academic+Press%2E)
9. Alan Bell, Daniel Jurafsky, Eric Fosler-Lussier, Cynthia Girand, Michelle Gregory, and Daniel Gildea. 2003. Effects of disfluencies, predictability, and utterance position on word form variation in english conversation. The Journal of the Acoustical Society of America, 113(2):1001–1024. [↗](https://doi.org/10.1121/1.1534836)
10. Merrill F Garrett. 1975. The analysis of sentence production. In Psychology of learning and motivation, volume 9, pages 133–177. Elsevier. [↗](https://doi.org/10.1016/s0079-7421(08)60270-4)
11. J. Kathryn Bock and Richard K Warren. 1985. Conceptual accessibility and syntactic structure in sentence formulation. Cognition, 21:47–67. [↗](https://doi.org/10.1016/0010-0277(85)90023-x)
12. Edward Gibson. 2000. The dependency locality theory: A distance-based theory of linguistic complexity. Image, language, brain, pages 95–126. [↗](https://doi.org/10.7551/mitpress/3654.003.0008)
13. Kathryn Bock and Willem JM Levelt. 1994. guage production: Grammatical encoding. demic Press. [find](https://scholar.google.com/scholar?q=Kathryn+Bock+and+Willem+JM+Levelt%2E+1994%2E+guage+production%3A+Grammatical+encoding%2E+demic+Press%2E)
14. J. J. Godfrey, E. C. Holliman, and J. McDaniel. 1992. Switchboard: telephone speech corpus for research and development. In [Proceedings] ICASSP-92: 1992 IEEE International Conference on Acoustics, Speech, and Signal Processing, volume 1, pages 517–520 vol.1. [find](https://scholar.google.com/scholar?q=J%2E+J%2E+Godfrey%2C+E%2E+C%2E+Holliman%2C+and+J%2E+McDaniel%2E+1992%2E+Switchboard%3A+telephone+speech+corpus+for+research+and+development%2E+In+%5BProceedings%5D+ICASSP%2D92%3A+1992+IEEE+International+Conference+on+Acoustics%2C+Sp)
15. Lan-Aca-Sasha Calhoun, Jean Carletta, Jason M Brenier, Neil Mayo, Dan Jurafsky, Mark Steedman, and David Beaver. 2010. The nxt-format switchboard corpus: a rich resource for investigating the syntax, semantics, pragmatics and prosody of dialogue. Language resources and evaluation, 44(4):387–419. [↗](https://doi.org/10.1007/s10579-010-9120-1)
16. John Hale. 2001. [A probabilistic earley parser as a psycholinguistic model](https://md.research.elamur.ai/free/en/papers/hale-2001-probabilistic-earley-parser-as-psycholinguistic/). In Proceedings of the second meeting of the North American Chapter of the Association for Computational Linguistics on Language technologies, pages 1–8. Association for Computational Linguistics.
17. Herbert H Clark and Thomas Wasow. 1998. Repeating words in spontaneous speech. Cognitive psychology, 37(3):201–242. [↗](https://doi.org/10.1006/cogp.1998.0693)
18. Daphna Heller, Jennifer E Arnold, Natalie Klein, and Michael K Tanenhaus. 2015. Inferring difficulty: Flexibility in the real-time processing of disfluency. Language and speech, 58(2):190–203. [↗](https://doi.org/10.1177/0023830914528107)
19. Samvit Dammalapati, Rajakrishnan Rajkumar, and Sumeet Agarwal. 2019. Expectation and locality effects in the prediction of disfluent fillers and repairs Matthew Honnibal and Mark Johnson. 2014. [Joint incremental disfluency detection and dependency parsing](https://md.research.elamur.ai/free/en/papers/honnibal-2014-joint-incremental-disfluency-detection-dependency/). Transactions of the Association for Computational Linguistics, 2:131–142.
20. Nancy Ide and Keith Suderman. 2004. The american national corpus first release. In LREC. [find](https://scholar.google.com/scholar?q=Nancy+Ide+and+Keith+Suderman%2E+2004%2E+The+american+national+corpus+first+release%2E+In+LREC%2E)
21. Jeremy G Kahn, Matthew Lease, Eugene Charniak, Mark Johnson, and Mari Ostendorf. 2005. Effective use of prosody in parsing conversational speech. In Proceedings of the conference on human language technology and empirical methods in natural language processing, pages 233–240. Association for Computational Linguistics. [find](https://scholar.google.com/scholar?q=Jeremy+G+Kahn%2C+Matthew+Lease%2C+Eugene+Charniak%2C+Mark+Johnson%2C+and+Mari+Ostendorf%2E+2005%2E+Effective+use+of+prosody+in+parsing+conversational+speech%2E+In+Proceedings+of+the+conference+on+human+language+tec)
22. Steven G Lapointe and Gary S Dell. 1989. A synthesis of some recent work in sentence production. In Linguistic structure in language processing, pages 107–156. Springer. [↗](https://doi.org/10.1007/978-94-009-2729-2_4)
23. Willem JM Levelt. 1992. Accessing words in speech production: Stages, processes and representations. Cognition. [↗](https://doi.org/10.1016/0010-0277(92)90038-j)
24. Willem JM Levelt, Ardi Roelofs, and Antje S Meyer. 1999. A theory of lexical access in speech production. Behavioral and brain sciences, 22:1–38. [↗](https://doi.org/10.1017/s0140525x99001776)
25. Roger Levy. 2008. Expectation-based syntactic comprehension. Cognition, 106(3):1126–1177. [↗](https://doi.org/10.1016/j.cognition.2007.05.006)
26. Elizabeth Shriberg. 1995. Acoustic properties of disfluent repetitions. In Proceedings of the international congress of phonetic sciences, volume 4, pages 384– 387. [find](https://scholar.google.com/scholar?q=Elizabeth+Shriberg%2E+1995%2E+Acoustic+properties+of+disfluent+repetitions%2E+In+Proceedings+of+the+international+congress+of+phonetic+sciences%2C+volume+4%2C+pages+384%E2%80%93+387%2E)
27. Elizabeth Shriberg, Rebecca Bates, and Andreas Stolcke. 1997. A prosody only decision-tree model for disfluency detection. In Fifth European Conference on Speech Communication and Technology. [find](https://scholar.google.com/scholar?q=Elizabeth+Shriberg%2C+Rebecca+Bates%2C+and+Andreas+Stolcke%2E+1997%2E+A+prosody+only+decision%2Dtree+model+for+disfluency+detection%2E+In+Fifth+European+Conference+on+Speech+Communication+and+Technology%2E)
28. Andreas Stolcke. 2002. Srilm-an extensible language modeling toolkit. In Seventh international conference on spoken language processing. [find](https://scholar.google.com/scholar?q=Andreas+Stolcke%2E+2002%2E+Srilm%2Dan+extensible+language+modeling+toolkit%2E+In+Seventh+international+conference+on+spoken+language+processing%2E)
29. Trang Tran, Shubham Toshniwal, Mohit Bansal, Kevin Gimpel, Karen Livescu, and Mari Ostendorf. 2017. Parsing speech: a neural approach to integrating lexical and acoustic-prosodic information. arXiv preprint arXiv:1704.07287. [↗](https://arxiv.org/abs/1704.07287)
30. Jean E Fox Tree and Herbert H Clark. 1997. Pronouncing “the” as “thee” to signal problems in speaking. Cognition, 62(2):151–167. [↗](https://doi.org/10.1016/s0010-0277(96)00781-0)
31. Mitchell Marcus, Grace Kim, Mary Ann Marcinkiewicz, Robert MacIntyre, Ann Bies, Mark Ferguson, Karen Katz, and Britta Schasberger. 1994. [The penn treebank: Annotating predicate argument structure](https://md.research.elamur.ai/free/en/papers/marcus-1994-penn-treebank-annotating-predicate-argument/). In Proceedings of the Workshop on Human Language Technology, HLT ’94, pages 114–119, Stroudsburg, PA, USA. Association for Computational Linguistics.
32. Shaolei Wang, Wanxiang Che, Yue Zhang, Meishan Zhang, and Ting Liu. 2017. Transition-based disfluency detection using lstms. In Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, pages 2785–2794. [find](https://scholar.google.com/scholar?q=Shaolei+Wang%2C+Wanxiang+Che%2C+Yue+Zhang%2C+Meishan+Zhang%2C+and+Ting+Liu%2E+2017%2E+Transition%2Dbased+disfluency+detection+using+lstms%2E+In+Proceedings+of+the+2017+Conference+on+Empirical+Methods+in+Natural+Langu)
33. Quinn McNemar. 1947. Note on the sampling error of the difference between correlated proportions or percentages. Psychometrika, 12(2):153–157. [↗](https://doi.org/10.1007/bf02295996)
34. Vicky Zayats, Mari Ostendorf, and Hannaneh Hajishirzi. 2016. Disfluency detection using a bidirectional lstm. arXiv preprint arXiv:1604.03209. [↗](https://arxiv.org/abs/1604.03209)
35. Benjamin Munson. 2007. Lexical access, lexical representation, and vowel production. Laboratory phonology, 9:201–228. [find](https://scholar.google.com/scholar?q=Benjamin+Munson%2E+2007%2E+Lexical+access%2C+lexical+representation%2C+and+vowel+production%2E+Laboratory+phonology%2C+9%3A201%E2%80%93228%2E)
36. Janet Pierrehumbert et al. 2002. Word-specific phonetics. Laboratory phonology, 7. [↗](https://doi.org/10.1515/9783110197105.1.101)
37. José C Pinheiro and Douglas M Bates. 2000. Linear mixed-effects models: basic concepts and examples. Mixed-effects models in S and S-Plus, pages 3–56. [find](https://scholar.google.com/scholar?q=Jose%CC%81+C+Pinheiro+and+Douglas+M+Bates%2E+2000%2E+Linear+mixed%2Deffects+models%3A+basic+concepts+and+examples%2E+Mixed%2Deffects+models+in+S+and+S%2DPlus%2C+pages+3%E2%80%9356%2E)
38. Gregory Scontras, William Badecker, Lisa Shank, Eunice Lim, and Evelina Fedorenko. 2015. Syntactic complexity effects in sentence production. Cognitive Science, 39(3):559–583. [↗](https://doi.org/10.1111/cogs.12168)
39. Claude Elwood Shannon. 1948. A mathematical theory of communication. Bell system technical journal, 27(3):379–423. [↗](https://doi.org/10.1002/j.1538-7305.1948.tb01338.x)
40. Victoria Zayats, Mari Ostendorf, and Hannaneh Hajishirzi. 2014. Multi-domain disfluency and repair detection. In Fifteenth Annual Conference of the International Speech Communication Association. [find](https://scholar.google.com/scholar?q=Victoria+Zayats%2C+Mari+Ostendorf%2C+and+Hannaneh+Hajishirzi%2E+2014%2E+Multi%2Ddomain+disfluency+and+repair+detection%2E+In+Fifteenth+Annual+Conference+of+the+International+Speech+Communication+Association%2E)
41. Simon Zwarts and Mark Johnson. 2011. The impact of language models and loss functions on repair disfluency detection. In Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies-Volume 1, pages 703–711. Association for Computational Linguistics. *Figure 1: Scatter plot of random intercepts for different POS tags* *Figure 2: Percentage count of different POS tags split by Storage Costs* [find](https://scholar.google.com/scholar?q=Simon+Zwarts+and+Mark+Johnson%2E+2011%2E+The+impact+of+language+models+and+loss+functions+on+repair+disfluency+detection%2E+In+Proceedings+of+the+49th+Annual+Meeting+of+the+Association+for+Computational+Lin)

*Updated 24 October 2026*