Elamur Research⌕
☰
home / papers / 1990 / A Computational Grammar of Discourse-Neutral Prosodic Phrasing in English

A Computational Grammar of Discourse-Neutral Prosodic Phrasing in English

Paper · 1990 · J. Bachenko and E. Fitzpatrick

Summary

Bachenko and Fitzpatrick address how a text-to-speech system can decide where to put prosodic breaks, and how strong they should be, using only the sentence and no knowledge of the surrounding discourse. They call this level "discourse-neutral" phrasing. Their earlier 1986 system tied phrasing to clause structure and predicate-argument relations. Here they drop both assumptions. In the new grammar, a parser supplies word categories, NP/PP/AdjP chunks and their heads, and prosody rules then add three further factors: adjacency to a verb, linear word order, and constituent length. Because length is treated as a separate phonological pressure, prosodic phrases can cut across major syntactic boundaries, as human speech often does. The rules were checked against roughly 500 sentences whose breaks were transcribed independently by both authors, and against sentences from G&G. Few misses were serious, and false primary breaks were almost absent. The work shows that speech-oriented parsing needs far less structure than parsing for translation or retrieval, and that phrasing behaves as a syntax-informed but length-filtered system.

Key points

  • Reference data. The gold data came from taped speech: The Speckled Band, a Sherlock Holmes tale performed by professional actors, and an Everest documentary mixing prepared narration with spontaneous interviews.
    • Both authors marked primary breaks (cued by pause), secondary breaks (cued by pitch change), or no break.
    • Disagreements were thrown out, leaving about 500 sentences.
  • Inputs. The system was built for the Olive-Liberman synthesizer. It needs no VP or S nodes and no argument/adjunct distinction, since observed phrasing treated complements and adjuncts alike.
  • Appendix A results (21 sentences). Only 2 of the 57 human boundaries were missed in a way judged truly problematic. No spurious primary break was ever produced. Secondary breaks were overgenerated at 7 spots, of which 3 were harmful.
  • G&G results. The system matched 12 of 14 primary boundaries exactly. Both misses, and the one serious overgeneration, involved sentential subjects, because the system discards the S node.
  • Main error source. Most failures came from the parser rather than the prosody rules: wrong part-of-speech tags, mis-analysed pre-head modifiers, and unrecognized idioms.

Interesting ideas

  • I1. Keeping only items on which two independent annotators agree yields a smaller but trustworthy gold set, with disagreement itself flagged as unreliable signal.
  • I2. Evaluation splits errors into misses and overgenerations, then grades each by whether it actually misleads, following a "first do no harm" principle rather than raw accuracy.
  • I3. Phrasing is modelled as a neutral default that discourse factors (emphasis, contrast, coreference) may later override, keeping the two layers cleanly separated.
  • I4. Structure proposes break points, but an independent length-balancing constraint decides which ones surface, so output units need not mirror the tree.
  • I5. The authors show that a downstream task may need only a thin slice of syntax (categories, small phrases, heads), which makes it robust to deeper parse errors.

For Mova

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

  • I1 (learn): Silver data could keep only edges where the witness parsers agree, and disagreeing sentences could be logged separately as a pool for the LLM teacher tiers.
  • I2 (learn): Evaluation reports could weight confident wrong outputs (a false fact, a wrong answer) much more heavily than abstentions, and track the two error types separately.
  • I3 (frames): Prag could first compute a context-free literal reading of each clause, then apply implicature, irony or contrast as explicit, explainable overrides on that default.
  • I4 (back): Generation and the text → graph → text compressor could pick clause splits and comma placement from UD structure, filtered by a chunk-length balance, to avoid run-on or choppy sentences.
  • I5 (tree): Mova could expose a shallow "chunks + heads" view of the UD tree as a fallback for consumers that do not need full clause attachment when the parser is unsure.

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

🏢AT&TBedminster · United States40 works · 75,466 citations

 Fields, topics and concepts

📚 Shelves 6The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 worksPsychology · 1,649 worksExperimental and Cognitive Psychology · 930 works

🧭 Topics 3Research subjects and languages the work is about.English · 2,379 worksDiscourse analysis · 471 worksSpeech & spoken language · 519 works

Cite

Cite as J. Bachenko and E. Fitzpatrick. 1990. A Computational Grammar of Discourse-Neutral Prosodic Phrasing in English. Computational Linguistics.

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

@article{bachenko1990computational,
  title = {{A Computational Grammar of Discourse-Neutral Prosodic Phrasing in English}},
  author = {J. Bachenko and E. Fitzpatrick},
  journal = {Computational Linguistics},
  year = {1990},
  url = {https://aclanthology.org/J90-3003/}
}

References

40 references, 2 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Allen, G. 1975 Speech Rhythm: Its Relation to Performance Universals and Articulatory Timing. Journal of Phonetics 3: 75-86. ↗
  2. Allen, J. 1976 Synthesis of Speech from Unrestricted Text. Proceedings of the IEEE 4: 433442. ↗
  3. Bachenko, J.; Hindle, D.; and Fitzpatrick, E. 1983 Constraining a Deterministic Parser. Proceedings of the National Conference on Artificial Intelligence (AAAI-83). find
  4. Bachenko, J.; Fitzpatrick, E.; and Wright, C. E. 1986 The Contribution of Parsing to Prosodic Phrasing in an Experimental Text-to-Speech System. Proceedings of the 24th Annual Meeting of the Association for Computational Linguistics, 145-153. find
  5. Bierwisch, M. 1966 Regeln fur die Intonation deutscher Satze. In: Bierwisch, M. (ed.), Studia Grammatica VII." Untersuchungen uber Akzent und Intonation im Deutschen. Akademie-Verlag, Berlin: 99-201. find
  6. Bing, J. 1985 Aspects of Prosody. Garland Press, New York, New York. find
  7. Cahn, J. 1988 From Sad to Glad: Emotional Computer Voices. Proceedings of Speech Tech "88: 35-36. find
  8. Chomsky, N. 1965 Aspects of the Theory of Syntax. MIT Press, Cambridge, MA. ↗
  9. Chomsky, N. and Halle, M. 1968 The Sound Pattern of English. Harper & Row, New York, New York. find
  10. Church, K. 1988 A Stochastic Parts Program and Noun Phrase Parser for Unrestricted Text. Proceedings of the Second Conference on Applied Natural Language Processing (ACL): 136-143. ↗
  11. Cooper, W. E. 1976 Syntactic Control of Timing in Speech Production: A Study of Complement Clauses. Journal of Phonetics 4:151-171. ↗
  12. Cooper, W. and Paccia-Cooper, J. 1980 Syntax and Speech. Harvard University Press, Cambridge, MA. ↗
  13. Crystal, D. 1969 Prosodic Systems and Intonation in English. Cambridge University Press, Cambridge, U.K. find
  14. Dommergues, J.-Y. and Grosjean, F. 1981 Performance Structures in the Recall of Sentences. Memory and Cognition 9: 478-486. ↗
  15. Downing, B. 1970 Syntactic Structure and Phonological Phrasing in English. Ph.D. Dissertation, University of Texas, Austin, TX. find
  16. Elovitz, H.; Johnson, R.; McHugh, A.; and Shore, J. E. 1976 Letter-to-Sound Rules for Automatic Translation of English Text to Phonetics. IEEE Transactions on Acoustics, Speech, and Signal Processing 6: 446-459. ↗
  17. Emorine, O. M. and Martin, P. M. 1988 The Multivoc Text-to-Speech System. Proceedings of the Second Conference on Applied Natural Language Processing (,4 CL): 115-120. find
  18. Fitzpatrick, E. 1989 The Preferred Prosodic Phrasing of Prepositional Phrases. Unpublished data. find
  19. Fitzpatrick, E. and Bachenko, J. 1989 Parsing for Prosody: What a Text-to-Speech System Needs from Syntax. Proceedings of IEEE Artificial Intelligence Systems in Government (AISIG) "89. find
  20. Gee, J. P. and Grosjean, F. 1983 Performance Structures: A Psycholinguistic and Linguistic Appraisal. Cognitive Psychology 15:411-458. ↗
  21. Grosjean, F. and Gee, J. P. 1987 Prosodic Structure and Spoken Word Recognition. Cognition 25:135-155. ↗
  22. Grosjean, F.; Grosjean, L.; and Lane, H. 1979 The Patterns of Silence: Performance Structures in Sentence Production. Cognitive Psychology 11: 58-81. ↗
  23. Hillinger, M.; James, C. T.; Zell, D. L.; and Prato, L. M. 1976 The Influence of Prescriptive and Subjective Phrase Markers on Retrieval Latencies. Bulletin of the Psychonomic Society 8: 353-355. ↗
  24. Hindle, D. 1983 User Manual for Fidditch, a Deterministic Parser. NRL Technical Memorandum #7590-142. find
  25. Hirschberg, J. and Litman, D. 1987 Now Let's Talk About Now: Identifying Cue Phrases Intonationally. Proceedings of the 25th Annual Meeting of the Association for Computational Linguistics, 163-171 find
  26. Klatt, D. H. 1987 Review of Text-to-Speech Conversion for English. Journal of the Acoustic Society of America 82: 737-793. ↗
  27. Klatt, D. H. 1975 Vowel Lengthening Is Syntactically Determined in Connected Discourse. Journal of Phonetics 3: 129-140. find
  28. Koster, Jan. 1978 Why Subject Ss Don't Exist. In: Keyser, Samuel J. Recent Transformational Studies in European Languages. Linguistic Inquiry Monograph 3. MIT Press, Cambridge, MA. find
  29. Langendoen, D. T. 1975 Finite-State Parsing of Phrase-Structure Languages and the Status of Readjustment Rules in Grammar. Linguistic Inquiry 6: 533-554. find
  30. Litman, D. and Hirschberg, J. 1990. Disambiguating Cue Phrases in Text and Speech. Proceedings of the 13th International Conference on Computational Linguistics (COLING). find
  31. Liberman, M. Y. and Prince, A. 1977 On Stress and Linguistic Rhythm. Linguistic Inquiry 8: 249-336. find
  32. Luce, P. A.; Feustel, T. C.; and Pisoni, D. B. 1983 Capacity Demands in Short-Term Memory for Synthetic and Natural Speech. Human Factors 25: 17-32. ↗
  33. Martin, E. 1970 Toward an Analysis of Subjective Phrase Structure. Psychological Bulletin 74:153-166. ↗
  34. Nespor., M. and Vogel, I. 1986 Prosodic Phonology. Foris Publications, Dordrecht, The Netherlands. find
  35. Olive, J. P. and Liberman, M. Y. 1985 Text-to-Speech — An Overview. Journal of the Acoustic Society of America, Supplement 1: 78, $6. find
  36. O'Shaughnessy, D. D. 1989 Parsing with a Small Dictionary for Applications s~:ch as Text-to-Speech. Computational Linguistics 15: 97-108. find
  37. Pierrehurnbert, J. B. 1980 The Phonetics and Phonology of English Intonation. Ph.D. Dissertation, MIT, Cambridge, MA. find
  38. Selkirk, E. O. 1984 Phonology and Syntax: The Relation between Sound and Structure. MIT Press, Cambridge, MA. find
  39. Sproat, R. W. and Liberman, M. Y. 1987 Toward Treating English Nominals Correctly. Proceedings of the 25th Annual Meeting of the Association for Computational Linguistics, 140-146.
  40. Streeter, L. A. 1978 Acoustic Determinants of Boundary Perception. Journal of the Acoustical Society of America 64:1582-1592. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/bachenko-1990-computational-grammar-discourse-neutral-prosodic/