Elamur Research⌕
☰
home / papers / 2016 / Predicting the Direction of Derivation in English Conversion

Predicting the Direction of Derivation in English Conversion

Paper · 2016 · Max Kisselew, Laura Rimell, Alexis Palmer et al.

Summary

English often switches a word between noun and verb with no visible affix (tunnel, walk, email). The authors ask whether corpus statistics can recover which use came first historically. They treat earlier attestation as a theory-neutral stand-in for the direction of conversion.

The gold standard comes from CELEX: about 2,000 monomorphemic lemmas labelled N-to-V or V-to-N. Noun and verb occurrences were drawn from the PoS-tagged BNC and ukWaC corpora. Two hypotheses were tested:

  • the derived category is rarer than its base;
  • the derived category is semantically narrower than its base, measured by entropy and KL divergence over sparse PPMI context vectors built from frequency-matched samples.

Frequency works well for nouns turned into verbs but poorly for the reverse. Specificity shows the opposite pattern. A logistic regression combining both reaches roughly 77% accuracy in each direction.

The asymmetry matters for theory. Denominal verbs behave as if built from a category-neutral root. Deverbal nouns behave like real derivations that add meaning.

Key points

  • Gold data: 1,044 N-to-V and 948 V-to-N lemmas from CELEX (WebCelex). Conversion entries are identified by MorphStatus=Z, and bases are restricted to monomorphemic forms.

  • Corpus: BNC plus ukWaC, 2.36 billion lemmatized, PoS-tagged tokens.

    • A manual check found noun/verb tags about 85% correct (κ=0.68).
    • Wikipedia was dropped because its tagging was less reliable.

  • Vectors: 10,000-dimensional bag-of-words context vectors with a ±5 word window, PPMI weighting and L1 normalization.

    • Each noun/verb pair is downsampled to min(f_N, f_V) occurrences.
    • The KL reference distribution is the centroid of about 28,000 vectors.
    • Entropy and KL correlate at ρ=0.91.

  • Accuracy by direction (N-to-V / V-to-N / all):

    ModelN-to-VV-to-NAll
    Frequency84.7%58.7%72.3%
    Entropy50.1%75.5%62.2%
    KL divergence53.8%76.7%64.6%
    Combined (Freq + H + KL)77.4%76.0%76.8%
    Majority-class baseline (all)——52.4%

  • Normalizing predictors by part-of-speech means barely changed results (76.8% → 77.0%). The effect is therefore not just a generic noun-versus-verb difference.

Interesting ideas

  • I1. Before comparing information measures between two words, rebuild both vectors from the same number of randomly chosen occurrences, so that density differences caused by frequency cannot fake a specificity gap.
  • I2. Use sparse, probabilistically interpretable count vectors (not dense embeddings) whenever the downstream score is entropy or KL against a corpus-wide centroid.
  • I3. Frequency and specificity are complementary cues, each strong for one conversion direction, and combining them gives balanced accuracy.
  • I4. Denominal verbs carry loose, noun-anchored meanings, while deverbal nouns regularly name the event or its result.
  • I5. When noun/verb tagging is ambiguous (gerunds, participle-like adjectives), overt inflection serves as a decisive tie-breaker, because inflectional suffixes strictly select their base's category.

For Mova

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

  • I1 (learn): Any comparison of witness-parser statistics or corpus-derived scores between rare and common items could use frequency-matched resampling as a built-in negative control.
  • I2 (world): A KL-from-centroid specificity score could rank WordNet/Wikidata concepts by generality, giving an explainable signal for hypernym ordering in the ontology.
  • I3 (tokens): The en morphology layer could store a predicted base category for noun–verb homographs, using a small explainable logistic model over frequency difference and specificity difference.
  • I4 (frames): When a converted verb lacks a VerbNet entry, prag could build its frame around the source noun (instrument, location, etc.), and treat deverbal nouns as event or result nominals linked to the verb's roles.
  • I5 (tokens): The tagger could add an inflection-based rule that resolves -ing/-ed noun/verb conflicts and logs the rule as the explanation for the chosen tag.

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

🎓Heidelberg UniversityHeidelberg · Germany27 works · 40,944 citations

 Fields, topics and concepts

📚 Shelves 2The field of science this work belongs to (OpenAlex fields).Arts and Humanities · 9,296 worksLanguage and Linguistics · 9,206 works

🧭 Topics 1Research subjects and languages the work is about.English · 1,373 works

Cite

Cite as Max Kisselew, Laura Rimell, Alexis Palmer and Sebastian Padó. 2016. Predicting the Direction of Derivation in English Conversion. Proceedings of the 14th SIGMORPHON Workshop on Computational Research in Phonetics, Phonology, and Morphology.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/w16-2015

@inproceedings{kisselew2016predicting,
  title = {{Predicting the Direction of Derivation in English Conversion}},
  author = {Max Kisselew, Laura Rimell, Alexis Palmer and Sebastian Padó},
  booktitle = {Proceedings of the 14th SIGMORPHON Workshop on Computational Research in Phonetics, Phonology, and Morphology},
  year = {2016},
  doi = {10.18653/v1/w16-2015},
  url = {https://doi.org/10.18653/v1/w16-2015}
}

References

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

  1. Herbert Koziol. 1937. Handbuch der englischen Wort-bildungslehre. C. Winter, Heidelberg. find
  2. Maya Arad. 2005. Roots and patterns: Hebrew morpho-syntax. Springer, Dordrecht. ↗
  3. Stela Manova and Wolfgang U. Dressler. 2005. The morphological technique of conversion in the inflecting-fusional type. In Approaches to Conversion/Zero-Derivation. Waxmann, Münster. find
  4. Harald R. Baayen, Richard Piepenbrock, and Leon Gulikers. 1995. The CELEX lexical database. Release 2. LDC96L14. Linguistic Data Consortium, University of Pennsylvania, Philadelphia, PA. find
  5. Hans Marchand. 1969. The categories and types of present-day English word-formation: A synchronic-diachronic approach, 2nd edition. C.H. Becksche Verlagsbuchhandlung, München. find
  6. Laurie Bauer. 1983. English word-formation. Cambridge University Press, Cambridge. ↗
  7. Leonard Bloomfield. 1933. Language. Henry Holt, New York. find
  8. David Pesetsky. 1995. Zero syntax: Experiencers and cascades. MIT Press, Cambridge. find
  9. Hagit Borer. 2005a. In name only: Structuring sense, Volume I. Oxford University Press, Oxford. find
  10. Ingo Plag. 1999. Morphological productivity: Structural constraints in English derivation. Mouton de Gruyter, Berlin and New York. find
  11. Hagit Borer. 2005b. The normal course of events: Structuring sense, Volume II. Oxford University Press, Oxford. find
  12. Ingo Plag. 2003. Word-Formation in English. Cambridge University Press, Cambridge. ↗
  13. Eve V. Clark and Herbert H. Clark. 1979. When nouns surface as verbs. Language, 55:767–811. ↗
  14. Laura Rimell. 2012. Nominal Roots as Event Predicates in English Denominal Conversion Verbs. Ph.D. thesis, New York University. find
  15. Jan Don. 1993. Morphological conversion. Ph.D. thesis, Utrecht University. find
  16. Jane Grimshaw. 1990. Press, Cambridge. Argument Structure. find
  17. Enrico Santus, Alessandro Lenci, Qin Lu, and Sabine Schulte im Walde. 2014. Chasing hypernyms in vector spaces with entropy. In Proceedings of EACL, pages 38–42, Gothenburg, Sweden. find
  18. MIT Ken Hale and Samuel Jay Keyser. 2002. Prolegomenon to a theory of argument structure. MIT Press, Cambridge. ↗
  19. Petra Maria Vogel. 1996. Wortarten und Wortarten-wechsel. Zu Konversion und verwandten Erscheinungen im Deutschen und in anderen Sprachen. Mouton de Gruyter, Berlin and New York. find
  20. Morris Halle and Alec Marantz. 1993. Distributed morphology and the pieces of inflection. In S.J. Keyser and K. Hale, editors, The view from Building 20, pages 111–176. MIT Press, Cambridge. find
  21. Heidi Harley and Rolf Noyer. 1999. State-of-the-article: Distributed Morphology. GLOT, 4:3–9. find
  22. Frank W. Harwood and Alison M. Wright. 1956. Statistical study of English word formation. Language, 32(2):260–273. ↗
  23. Jennifer Hay. 2001. Lexical frequency in morphology: Is everything relative? Linguistics, 39:1041–70. ↗
  24. Aurélie Herbelot and Mohan Ganesalingam. 2013. Measuring semantic content in distributional vectors. In Proceedings of ACL, pages 440–445, Sofia, Bulgaria. find
  25. Ferenc Kiefer. 2005. Types of conversion in hungarian. In Approaches to Conversion/Zero-Derivation. Waxmann, Münster. find
  26. Paul Kiparsky. 1982. Word formation and the lexicon. In Fred Ingeman, editor, Proceedings of the Mid-America Linguistics Conference, page 3–29. University of Kansas. find
  27. Andrew Koontz-Garboden. 2007. States, changes of state, and the Monotonicity Hypothesis. Ph.D. thesis, Stanford University. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/kisselew-2016-predicting-direction-derivation-english-conversion/