Structural Ambiguity and Lexical Relations
Summary
The authors take a classic case of structural ambiguity: where to attach a prepositional phrase that follows a verb and its object, to the verb or to the noun. They cite earlier studies showing that the structural principles Right Association and Minimal Attachment are poor predictors of how humans resolve this ambiguity. The authors therefore propose relying on the lexical preferences of specific words and obtaining these preferences from a corpus. To this end, they took 13 million words of Associated Press newswire from 1989 and parsed them automatically with the partial parser Fidditch. From the parses they extracted verb–noun–preposition triples and converted them into frequencies of word–preposition pairs. Clear-cut cases were assigned first, then the statistics from these cases were used to resolve some of the ambiguous ones, and the remainder were split between the verb and the noun. These frequencies are used to compute a lexical association (LA) score. It compares how strongly the preposition is associated with the verb and with the noun, and also indicates how confident the decision is. The method errs in roughly one fifth of cases. This is somewhat worse than humans who see the same word triple. At the same time, the corpus proved to be a better source of associations than the COBUILD dictionary. The work showed that word distribution statistics can approximately solve a problem that seemingly requires complex reasoning.
Key points
- Data: 13 million words of AP (1989), parsed with the Fidditch parser; parts of speech were assigned beforehand by Church's tagger. The corpus contains 2,661,872 noun phrases. Of these, 467,920 are verb objects, and 753,843 are followed by a preposition. There are 223,666 ambiguous verb–noun–preposition triples.
- Prepositions were assigned in stages. If there is no preposition, the symbol NULL is recorded. If the object is a pronoun, the preposition is unambiguously attached to the verb. Unresolved cases are split between the two hypotheses.
- The error rate is about 20%. The method performs worst on adjuncts, particularly temporal ones. Here the authors propose also taking into account the noun that follows the preposition.
- A fixed lexicon of 7,822 associations (f > 3, I(w, p) > 2) yields the same accuracy as LA with a threshold of 2, but lower coverage. From this the authors conclude that the relative strength of associations matters. Most associations in the corpus-based lexicon and in COBUILD do not coincide.
- For comparison: in the study by Whittemore et al., none of the structural principles accounted for more than 55% of cases.
Cite
@misc{hindle1993structural,
title = {{Structural Ambiguity and Lexical Relations}},
author = {Hindle, Donald and Rooth, Mats},
year = {1993},
url = {https://aclanthology.org/J93-1005/}
}References
- Altmann, Gerry, and Steedman, Mark (1988). "Interaction with context during human sentence processing." Cognition, 30(3), 191-238. find
- Church, Kenneth W. (1988). "A stochastic parts program and noun phrase parser for unrestricted text." In Proceedings, Second Conference on Applied Natural Language Processing. Austin, Texas, 136-143. find our card
- Church, Kenneth; Gale, William; Hanks, Patrick; and Hindle, Donald (1991). "Using statistics in lexical analysis." In Lexical Acquisition: Exploiting On-Line Resources to Build a Lexicon, edited by Uri Zernik, 115-164. Lawrence Erlbaum Associates. find
- Ford, Marilyn; Bresnan, Joan; and Kaplan, Ronald M. (1982). "A competence-based theory of syntactic closure." In The Mental Representation of Grammatical Relations, edited by Joan Bresnan, 727-796. MIT Press. find
- Frazier, Lyn (1978). "On comprehending sentences: Syntactic parsing strategies." Doctoral dissertation, University of Connecticut. find
- Gale, William A.; Church, Kenneth W.; and Yarowsky, David (In press). "A method for disambiguating word senses in a large corpus." Computers and Humanities. find
- Hindle, Donald (1983). "User manual for fidditch, a deterministic parser." Naval Research Laboratory Technical Memorandum 7590-142, Naval Research Laboratory, Washington, D.C. find
- Hindle, Donald (in press). "A parser for text corpora." In Computational Approaches to the Lexicon, edited by B. T. S. Atkins and A. Zampolli. Oxford University Press. find
- Jenson, Karen, and Binot, Jean-Louis (1987). "Disambiguating prepositional phrase attachments by using on-line dictionary definitions." Computational Linguistics, 13(3--4), 251-260. find our card
- Kimball, J. (1973). "Seven principles of surface structure parsing in natural language." Cognition, 2, 15-47. find
- Marcus, Mitchell P. (1980). A Theory of Syntactic Recognition for Natural Language. find
- Mosteller, Frederick, and Wallace, David L. (1964). Inference and Disputed Authorship: The Federalist. Addison-Wesley. find
- Sinclair, J.; Hanks, P.; Fox, G.; Moon, R.; Stock, P. et al. (1987). Collins COBUILD English Language Dictionary. Collins. find
- Taraban, Roman, and McClelland, James L. (1988). "Constituent attachment and thematic role assignment in sentence processing: Influences of content-based expectations." Journal of Memory and Language, 27, 597-632. find
- Whittemore, Greg; Ferrara, Kathleen; and Brunner, Hans (1990). "Empirical study of predictive powers of simple attachment schemes for post-modifier prepositional phrases." Proceedings, 28th Annual Meeting of the Association for Computational Linguistics, 23-30. find
This page in Markdown → https://md.research.elamur.ai/free/en/papers/hindle-1993-structural-ambiguity-lexical-relations/