Elamur Research⌕
☰
home / papers / 1977 / Computation of a Subclass of Inferences: Presupposition and Entailment

Computation of a Subclass of Inferences: Presupposition and Entailment

Paper · 1977 · Aravind K. Joshi and Ralph Weischedel

Summary

Joshi and Weischedel study a narrow, well-behaved slice of natural-language inference: entailment and presupposition. Unlike open-ended conjecture, these hold in every context. S entails S′ when S′ is true wherever S is. S presupposes S′ when both S and its internal negation entail S′. The authors argue that such inferences are tied to word definitions and syntactic structure, so they can be computed structurally rather than by general deduction, which tends to blow up combinatorially. Their key notion is the "subformula-derived" inference: pick a well-formed piece of the sentence's meaning representation and apply a finite-state tree transformation to it. The only requirement is that the representation language has an approximately context-free syntax, so the method does not depend on any particular set of semantic primitives. They built an ATN-based system that computes these inferences during parsing. Lexical entries carry the transformations, and certain constructions, such as clefts, trigger their own. Because each sentence yields a finite set of results, the approach also gives a principled answer to when inference should stop.

Key points

  • Presuppositions come from two sources: lexical items such as stop, manage, fail, force and appreciate, and syntactic constructs such as cleft sentences and only.
  • The worked examples show the scheme in action:
    • John forced us to leave entails We left through an identity transformation on a subformula.
    • John stopped beating Mary presupposes John had been beating Mary, which requires a non-trivial transformation.
  • manage and fail to presuppose an attempt. Embedding under "is false" flips the entailment (Dr Smith challenged John) but leaves the presupposition (attempted to challenge) in place.
  • The system sorts its output into three groups: non-NP presuppositions, NP-related presuppositions (each referent exists in the shared information), and entailments. Each result is shown both as a formula and as a regenerated English sentence.
  • Embedded contexts are handled. Under if…then, presuppositions come out conditionalized. Under says that, they are wrapped in the speaker's claim. Full details are in Weischedel (1976).

Interesting ideas

  • I1. Store inference triggers in the lexicon as small tree rewrites applied to a subformula. Inference then becomes local, explainable and cheap instead of open-ended theorem proving.
  • I2. Use survival under negation as the operational test separating presupposition from plain entailment.
  • I3. Inference stops naturally because each sentence licenses only a finite set of structurally derived consequences.
  • I4. NP-related presuppositions treat each referent as existing in the shared information, much like "given information".
  • I5. Every derived formula is regenerated into English, so each inference can be inspected as a sentence.

For Mova

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

  • I1 (frames): verbs like stop, manage, fail, force could carry rewrite rules over their frame arguments, so prag emits presuppositions and entailments with a traceable rule ID.
  • I2 (learn): generate negated variants of silver sentences and use them as negative controls, checking that inferred presuppositions persist while plain entailments flip.
  • I3 (world): the world builder could restrict automatic state updates to these bounded sets, which avoids runaway chaining when populating entities and events.
  • I4 (world): definite NPs could add existence assertions to the text world, and coreference could check them against prior discourse.
  • I5 (back): verbalizing each inferred graph fact through the generator would give a human-readable audit of the inferences, plus a round-trip check.

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

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 works

🧭 Topics 1Research subjects and languages the work is about.Implicature · 138 works

Cite

Cite as Aravind K. Joshi and Ralph Weischedel. 1977. Computation of a Subclass of Inferences: Presupposition and Entailment. eScholarship (California Digital Library).

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

@inproceedings{joshi1977computation,
  title = {{Computation of a Subclass of Inferences: Presupposition and Entailment}},
  author = {Aravind K. Joshi and Ralph Weischedel},
  booktitle = {eScholarship (California Digital Library)},
  year = {1977},
  url = {https://aclanthology.org/J77-1008/}
}

References

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

  1. Chard&, Eugene, l T a a r d Model of Children's Story e n s i o n v . Artificial Intelligence Laboratory Report AI TR-266. Cambridge, MA: Massachusetts I n s t i t u t e of 'rechnology , 1972. find
  2. Charniak, Eugene, "Jack and Janet in Search of a Theory of Knowledgen. In Pmceedings o f the T h i r d International Joint Conferynce on Artificial Intelligence. Palo Alto, CA: Stanford Research Xnstimte , 1973. . find
  3. Clark, H. M. and Haviland, S E. , ~Corrrprehension and the Given-New Con-tpaetIt, in Discome Production and Comprehension (ed R. Freedle ) , Lawrence Blbaum Associates, Millside, N.J., 1976. find
  4. F5.lImre, C . Y e , "Verbs of Judging: an Ekercise in S m t i c Desmi~tion". In F i l h r e and Langendoen; studies in L i n g u i s t i c . S m t i c s . NLW York: Holt , Rinehart , and Winston ,ml. Givon, Tdlmy , "The Time — Axis Phenomenontt. Language 49,411973 7 : 890-925. . . Harris, Z S , "Tho Systems of Gr-: Report and Paraphraseu. In Harris, Structural and Transformational Linguistics. Dordrecht , Holland : D. Reidel Publishing Co. , 19 70. . . Haviland , S E and Clark, H.M. , "What ' s New? Acquiring new information as a process in Comprehension. Journal of V e r b a l Lmrnhg and Vmbal Behavior, — 13, 1974. . find
  5. Isard , S , "What Would You Have Cone I f . . tics . and Weischedel, R.M. , ”Some Frills for PMal Tic-tac-toe : of Predicate ~6mplemkn-t Constru@ionst', IEEE Transactions on Joshi A, K. Electronic Computers Special Issue on Artificial Intelligence, April 1976. k t t u n e n , L a u r i , "On the Saxintics of Complement Sentences". Ih Papers f r o m the Sixth Region@ Meetkg of the Chicago Linguistic Society. Chicago: University of Chicago, 1970'. find
  6. Karttunen, Lami, llPresupposit ions of Compound Sentences". N(1973): 169-193. find
  7. KArttunen, Lauri, "Presupposition and Linguistic Context". Liin&stics 1, 1/2(1974):182-194: Kwttunen, Lauri and Petem, Stanley, "Conventional 3nplioa-tt.m in Montague Gmm~". Presented at the F h t Annual Meeting of Berkeley L,inguistic Society, FelxYary 15, 1975. Berkeley, Cfl. . . . ?" Unpublished, 1974. n q w L i n . tic I Tneoretid Keenan, Edward, L. , A Logical Base for Ehglish. Unpublished Doctorel Diss&atioh, University of. Pennsylvania, 19 69. ↗
  8. KeeMn, Ehyard L., Kinds of Presupposition in M t m l Iimguagesff. In F i l l m r e and Langendoen, Studies in Ihguistic Semntics. New Yak: Holt , Rineha&, and Winstan, 1971. find
  9. Keenan, Edward L., "On Semantically Based (2immtT. Linguistic Inquiry IIJ(1972) : 413-461. find
  10. Carol, "Factfl. h Steihbwg and J&bovits, Cambridge Univer~m press j 1971. find
  11. K,,iparsky, Paul a d Semantics. New Y&k: Wff, George, HPresupposition and Relative Well-fomednesstl. In Steinberg and Jakobovits, Semantics. New York: Cambridge University Press, 1971. find
  12. Lehnert, Wendy, "What Makes Sam Run? Script Based Techniques for Question h s w k i n g In . Proceedings of $he Wprkshcp on Theoretical Issues in Nabnil. Language Processing. Association $or C q u t i o n a Linguistics, l 1975. bsenschein, Stanley J. , S j Lexical Information and Event Int&pretaticm. F h ' .D LIissePtat ion, University 'of Pennsylvanla, 197 5. find
  13. Sager, Naomi, 'The S t r i n g Parser f o r Scientific Literature". In Rustin, Natural Language Pmcessing. New Yo&: AlgoritMcs Press, 1973. find
  14. Schank, R., '!The Fourteen Primitive Actions and Their Inferences". Starifon$ A. I. Mm-183. Palo Alto, CA: Computer Science Deprbnent , S t a n f o r d University, 1973. . find
  15. Schank, Roger C , and Reiger , Charles J. , 111. "Inference and the Computer Understanding of Natural Language". Stanford A. I. Memo-197. Palo Alto, CA: m u t e r Science Department, Stanford University, 1973. . , . ↗
  16. SchaTlk, Roger C et al, "Inference ind Paraphrase by Computern. Journal ACM 22, 3(July 1975): 309-328. - Schank, Roger C., "Using Knowledge to Understand". In Proceedings of the Workshop on Theoretical Issues in Natural Language Rxcessing. Association for Computational Linguistics, 1975. find
  17. Smaby , Rid-md, "Consequence, Presuppositions and Cmeferacetl. To appear in Theoretical Linguistics, 1975. find
  18. Weiscbedel, Ralph M. , ltCaputatifn of an Unique Class of Inferences : A?esuppusitbn and b-tw . Ph.D. Dissertaticn, Univemity of krru3y1&, 1975. . find
  19. Weischedel., Ralph M. "A New Semantic Computation While Parsing: Presupposition and Ehtaimt ,Iv Technical Report 676, Dement of M m t i o n and hputter Science, University of Cal'ifornia, I n h e , CA, 1976. find
  20. Wilks, Yorick, "A P r e f e r a t i d , Pattern-seeking, Semantics fcr Natural Language Inferenceu. Artificial Intelligence 6 (1975) : 53-74. , Winogmd, T. Understanding Natural Language. find
  21. "TMnsition Network Q?am~xs far Natwal Lan&uage A n d Z y i s , Cc~rm. A m 13 ,lo (October 1970) : 5 9 1 6 0 6 . find
  22. Woods? W.A., Woods, U.A., "An E k p e r h t a l P a m h g System f a r TMnsition Network GMmnarsrt. In Rustin, Natmal Language Processing. New Yo~k: A l g a r i c Press, 1973. find
  23. Y-ki, Ma-aki, l t L R x i c a l Decc~npositim and DrplLied PIK)position". Unptiblished paper. University of Michigan, 1972 find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/joshi-1977-computation-subclass-inferences-presupposition-entailment/