Using Electronic Dictionaries and NooJ to Generate Sentences Containing English Phrasal Verbs

What it is about

The author uses the NooJ platform, which already recognises phrasal verbs (PV) in corpora with a dictionary of about 1,200 entries and a grammar, to generate sentences instead. A generation grammar takes one sentence with a PV and produces variants: particle movement (let out the dogs / let the dogs out), preterit, negation, clefting of subject or object, modal insertion from nine modals, aspect (start, continue, finish), question formation and passive voice. For each sentence with a PV the grammar gives 2,694 entries, and the example "Max figures out the problem" yields over 2,500 variants. The grammar handles only singular nouns in present and past tense, and every output is grammatical by construction; no measurement of naturalness is given. In an earlier study the recogniser reached 88% accuracy, which rose to 98% on the Dickens and Melville novels when only six particles (out, up, down, away, back, off) were kept instead of twelve.

Ideas

I1. en/src/morph.rs and 28. Voice and the passive: write a small generator over a fixed transformation list (particle movement, preterit, negation, passive, cleft). Each output gets its transformation trace, so a failure says which step broke.
I2. Use the generated sets as a test for the parser and the linearizer: a generated sentence has a known UD tree, so en/src/lin.rs and tools/roundtrip can be scored on word order after particle movement, with the stated rule as the reason.
I3. Test the reduction to six particles inside Mova's phrasal-verb identification: compare precision with the full particle list and with the six-particle list on a held-out set, run twice, and report both numbers with the date.
I4. Negative control: the generator should produce sentences that the identifier rejects when the particle is a preposition (look over the fence), so a generated set with flipped particles must score lower.

For Mova

The chapters are 13. Verbs for the verb-particle list and 29. Ellipsis, word order and information structure for particle movement. The code points are en/src/morph.rs for inflection and en/src/lin.rs for the word order, with tools/roundtrip as the harness. The measure is exact match of generated word order against the expected tree, and precision of particle identification with six versus twelve particles, held out and repeated for determinism.

Cite

The paper itself, not this card. BibTeX from the library's record of the work. DOI 10.18653/v1/w17-3805

@inproceedings{machonis2017using,
  title = {{Using Electronic Dictionaries and NooJ to Generate Sentences Containing English Phrasal Verbs}},
  author = {Machonis, Peter A.},
  booktitle = {Proceedings of the Linguistic Resources for Automatic Natural Language Generation - LiRA@NLG},
  year = {2017},
  doi = {10.18653/v1/w17-3805},
  url = {https://doi.org/10.18653/v1/w17-3805}
}

References

5 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

Maurice Gross. 1994. Constructing Lexicon-Grammars. Computational Approaches to the Lexicon (eds. Atkins and Zampolli), 213-263. find
Oxford University Press, Oxford, UK Maurice Gross. 1996. Lexicon Grammar. Concise Encyclopedia of Syntactic Theories (eds. K. Brown and J. Miller), 244-258. Elsevier, New York. find
Risto Hiltunen. 1994. On Phrasal Verbs in Early Modern English: Notes on Lexis and Style. Studies in Early Modern English (ed. D. Kastovsky), 129-140. Mouton de Gruyter, Berlin, Germany. find
Lien Huynh Hodapp. 2010. Eliminating phrasal verbs in academic writing. Minnesota State University Memorial Library University Archives, Mankato, Minnesota. find
Peter A. Machonis. 2010. English Phrasal Verbs: from Lexicon-Grammar to Natural Language Processing. Southern Journal of Linguistics 34(1): 21-48. find

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/machonis-2017-electronic-dictionaries-nooj-generate-sentences/