Multiple Simultaneous Inter pr etations of Ambiguous Sentences Peter Norvig Uni ve rsity - PDF document

Multiple Simultaneous Interpretationsof Ambiguous SentencesPeter Norvig *UnieleyIntroductionThis paper is concerned with the problem of semantic and pragmatic interpretation of ambiguous sentences.We start by offering a simple yet commonly adopted interpretation strategy:Strategy 1:Apply syntactic rules to the sentence to derive a set of parse trees.Next apply seman-tic rules to the trees to get a set of logical formulae, and discard anDo apragmatic interpretation of each formula, and give a score to each possibility based on consistencyor likelihood in the gixt. Finallyhoose the interpretation with the highest score.In this framexicon, grammarnd semantic and pragmatic interpretation rules determine a map-ping between sentences and meanings.xactly one interpretation is unambiguous, one withno interpretation is anomalous, and one with multiple interpretations is ambiguous.To enumerate the pos-sible parses and logical forms of a sentence is the proper job of a linguist; to then choose from the possibili-ties the one ``correct'r`eaning of an utterance is an eIntelligence.One major problem with Strategy 1 is that it ignores the difference between sentences that seem trulyambiguous to the listenernd those that are only found to be ambiguous after careful analysis by the lin-guist. For example, each of the following is technically ambiguous (withcould signal the instrument oraccompanier case, andportcould be a harbor or the left side of a ship), but only the third would be seen asambiguous in a neutral context.(1) I saoman with long blond hair.(2) I drank a glass of port.Zadeh (personal communication) has suggested that ambiguity is a matter of degree. Heassumes eachinterpretation has a likelihood score attached to it.entence with a large gap between the highest andsecond ranked interpretation has loed interpretations has highambiguity; and in general the degree of ambiguity is inrsely proportional to the sharpness of the drop-offin ranking.So, in (1) and (2) abogree of ambiguity is beloome threshold, and thus is notnoticed. In(3), on the other hand, there are twed interpretations, and the ambiguity is per-ceissuch. Manesearchers, from Hockett (1954) to Jackendof1987), have suggested that the inter-pretation of sentences lik3) behaer cube or theaces or duck/rabbit illusion.In other words, it is possible to shift back and forth between alternateinterpretations, but it is not possible to perceive both at once.This leads us to Strategy 2:Strategy 2:Do syntactic, semantic, and pragmatic interpretation as in Strategy 1.Retain only thehighest-ranking interpretation(s), according to some threshold function.If there is more than oneinterpretation remaining, alternate between them.roblem with Strategy 2 is that it assumes all possible interpretations will be considered and ranked.Hower, mentences have a ply large or inŒnite number of interpretations.Consider the fol-lowing:(4) He seems older nohecan refer to one of senowcan refer to one of an inŒnite number of timepoints. Thus,while syntax and semantics may be producing discrete lists of possibilities, it seems thatpragmatic interpretation must operate by proposing likely interpretations, rather than enumerating all possi-bilities and then choosing among them.Hobbs (1983) has argued that enumeration should be minimizedhe author has been sponsored by the Defense Advanced Research Projects AgencNo. 4871, monitored by Space and Naare Systems Command under Contract N00039-84-C-0089. ev ensyntactic and semantic analysis.Accepting that intuition, we get Strategy 3, which is similar to theapproachs used by seral recent authors, including Hobbs (1987), Stallard (1987), and Charniak (unpub-lished).Strategy 3:Do lexical, syntactic and semantic analysis to produce one (or occasionally more)neutral representation of the input, which can contain ambiguous and vague predications.Prag-matics then attempts to `solve' for the ambiguous predications and some of the vague ones.Solu-tions are generated in a roughly best-Œrst mannerge drop-ofnthe rankingof solutions, we stop and the Œnal interpretation alternates between the high ranking one(s).This might represent an efŒcient interpretation mechanism, but it doesn'mechanism particularly well.Sentences likve only one good pragmatic interpretation, whichuld be found easily by Strategy 3.But (5-9) are notoriously hard for humans to get right without at leastonscious sense of having to back up and re-parse the sentence.(5) The horse raced past the barn fell..(7) The rabbi was hit on the temple.(8) The landlord painted all the walls with cracks.(9) Ross was told what to do by the rigies likodor'terms of general syntactic preference principles, which appeal to performance issues such as limits on aable memory space.Schubert (1984, 1986) and Kurtzman (1984) argue convincingly that no simple syn-tactic preference will do.Rathergy 4:Strategy 4:Do lexical, syntactic, and semantic analysis on a word-by-word basis, identifyingpoints of ambiguity along the wayvidence to rank alternatis. Evi-dence for a particular choice can include lexical frequencreferences, pragmatic associations,and other factors outside of the simple logical form.(and its alternatiiscarded) before the parse is complete, if its score remains sufŒciently abothe alternatiŒcient amount of time.In addition, if at anoint there are more than amaximum numbern(nwest ranking alternatificlose to others.At the end, alternate between the highest ranking interpretations, as before.Mutually Compatible Interpretations and ConnotationsConsider the following quote from Richard Parsons, of the American Fur Industry Inc., on their neer-tising sloganFur is for Life:It has a good sound, a good connotation.s, they're a good product.s, furs support wildlife conservation.rsons (although not a professional linguist) is making a claim about language use: that the proper orintended meaning of a phrase can be a combination of a number of interpretations and connotations.gies 2-4 assume that the reader etaser-cube-likealternation between interpretations.But Parsons is saying that his sloganFur is for Lifeis different. Theslogan seems to have two primary interpretations, (10) and (11) below. But it also has importantconnota-tions,listed as (12-14), as well as another interpretation, (15), that Parsons presumably wants the public toignore.(10) Fur lasts a lifetime.ation.(12) Fur wearers are li.(13) The recipient of a fur may become indebted to the gi(14) Life is a good thing; hence fur is a good thing. (15) FurAlthough we would not be likely to say that an12-14) are good candidates for the Œnal interpretation,it seems that the intended effect of the slogan is for the reader to entertain some or all of these simultane-ouslyhile this is a radical departure from the Hockett/Jackendoff/Necker theory of `one interpretation attappears to be quite common in poetryertising (see Burli-Storz, 1980).The facts are admittedly slippery; I am suggesting that alternative parses can sometimes be combinedinto one interpretation, but it is hard to distinguish between distinct parses that have been combinedtogethernd a vague interpretation that has seral possible entailments.Also, it is notoriously hard tointrospect about the phenomenology of these cases.Perhaps the following example, from the last line ofGerard ManleGod'will be more compelling:(16) Becausethe Holy Ghost oarm breast and with ah! bright wings.The wordbroodsis lexically ambiguous between `to sit on eggs to hatch them' and `to think long anddeeply or resentfully.' This is clearly not a case of vagueness. Yet it seems that the most natural interpreta-tion is of a bird-likgg-likorld (or world-likand waiting for it to come to fruition.This interpretation clearly ines no Necker-likbetween senses forbroods;rathertinv oes a simultaneous synthesis of twNote that not just anay; if the world were at, or if eggs werecubical, the combined image would not work. Itispermissible to combine the images erld is quite a bit larger and composed of different material than the agg, and eprototypical image of God does not include wings.Poetryertising, seems to sanction this superimposition of distinct parses.To support thisclaim, I opened a poetry anthology at random, Œnding the opening line of to Dylan Thomas' poemIn theBeginning:In the beginning was the three-pointed starstes clearhe three-pointed star should be taken as referring to a stellar body in primordial space, to the light in God'-mative speech act ``Let there be light,'' the star of Bethlehem, and to the Holy Trinityseem to be a clear feeling of shifting between these referents; rather theously.Lakofnd Turner (1988) cite, but do not fully analyize, another Dylan Thomas poem,Do not gontle into that good night:(17) Donot go gentle into that good night,Old age should burn and rave alose of day;Rage, rage against the dying of the light.Understanding this passage requires knowledge of at least six metaphors for life and death.While thesemetaphors offer conicting views on the nature of death, there is no feeling of having to switch betweenthem in understanding the poem; theve ance. Inct, metaphors (18-23) are all used in theinterpretation of the six wordsgo gentle into that good nightgointoforgood nightnightordnightis being used simultaneously as a time, a destina-tion, a containerersaryer-lik.(18) Life is a journeersary.(20) Life is `here'; death is a another world.(21) Death is sleep.urns out.At this point let us try to modify Strategy 4 to account for these nendings. Thereare twwe can treat the combination of twnd try to shoow ian be sanc-tioned, or we can treat it as the neasic interpretation mechanism, and try to shoow ian be con- strained.Strategy 5a: The Conserve Simultaneity Strategy:Ammend Strategy 5 to alloneous amalgam of twrmed interpretations, but only when sanctioned bysome as-yet-unspeciŒed factors, and only when the result is a coherent combination of the two.Strategy 5b: The Radical Simultaneity Strategy:Alwaystry to combine top-ranking interpreta-tions into one image.When a coherent combination is impossible, alternate between interpreta-tions as in Strategy 5.To try to choose between the two, we will Œrst consider Strategy 5b, as it is applied to sentence (24), and itsinterpretation, the disjunction (24'):(24) The chicken is ready to eat.(24') chicken(x) & ready(x,e) & eating(e) & (agent(e,x) | patient(e,x))Using Strategy 5b, we could combine the twtion, yielding `the chicken is ready to eat the chicken.his is by no means a normal interpretation of (24),so we have aument against 6bev er, that argument only goes through if the proposed logical form(24') is accurate.Suppose we use the following logical form instead:chicken(x) & ready(x,e) & eating(e) &((agent(e,x) & aliyard) & patient(e,seed) & ...) |(patient(e,x) & not aliThen we have two interpretations that cannot be combined coherentlygy 5b nor 5a.Thus, we see that for 5b to be feasible, we need to insist on full frame-likplete with default assumptions.We need a rich set of defaults to rule out unwanted uniŒcation of the twointerpretations, eant to allorriding some of the defaults, as in ``Thechicken on the table is ready to eat her asparagus.'' Now lry an example that does not bring as muchbackground knowledge into play:(25) She opened the door with a k.The ambiguity is betweenwith a kseems to be nothing to stop 5b from accepting both interpretations for the phrase, whereas we knohat ifthis were the intended meaning, one would have tse something likwing:(26) She opened the door with the key that was in/near it.Thus, Strategy 5b as it stands is rejected.To eluate Strategy 5a, we need to desanctioningombined interpretation, which we will address in the next section.Consider the following advertisement for Flintstones brand Vitamins:(27) Wwing.The coordinateand growingcan attach to eitherareorten million strongve ithat the individual children are growing, or that the number of children is increasing.Most informants rec-ognize both alternatiut report an ability to fuse the twvidual child in an expanding group is growing. (Hower, nne interpretedstrongas possibly modifyingkids,perhaps because of the idiomatic nature of the phraseten million strong.) Myanalysis of this exampleis that the listener arritthe twgy 4, and in the process oftrying to choose between them, realizes that both were intended interpretations, and successfully superim-poses the twoiIn short, (27) is a kind of pun.In a regular pun, the main point of the utterance is that the speaker hasbeen cler, producing twecondary point is one of the meanings (and, for agood pun, both of the meanings taken separately).But in (27) we have a special kind of pun, where thepoint is that both meanings are to be taken simultaneouslyxample comes from another ad, for Michelin tires:(28) Because you've got a lot riding on your tires.Here the ambiguous phrasegot a lot riding onis ambiguous between `much depends on your tires' and`much rides in the car which is on the tires,ith the resulting combined interpretation `your family'while in the car depends on the tires.ere again the reader must recognize the `pun,effect of combining the twnterpretations, but here there is an added hitch: it is the combination of the twointerpretations that resolves the phraseto `your family'; neither of the twpoint to this interpretation singlyLet us compare these puns to the following example from Freud (1916):(29) I met Baron Rothschild, and he treated me quite as his equal­quite famillionairely.This is funny, Freud claims, because of the unexpected ease of combiningfamiliarlywithmillionaireto cre-ate a neamiliarly as is possible for a millionaire.amili ¨ar + Millionmilion ¨ar.) Freud also presents the standard deŒnition of joking as the ability to Œnd hidden similaritiesbetween dissimilar things.This is amended to alloor the discoferences, or just ``to bind into aunityith surprising rapidityev eral ideas which are in fact alien to one anothernother words, thecombination of disparate ambiguous interpretations is an unusual eut one that we have acapacity for.emaining problem is to explain why, while others are not.Whyis it that, to my ears at least,the rabbi was hit on the templeis funny, wthe plumber lit his pipeismerely confusing?Freud claims that the laughter response is illicited by the release of suppressed violentor sexual thoughts.That explains, perhaps, whwing is a fairly good joke, while other lexical andstructural ambiguities in this paper are not:(30) She criticized his apartment, so he knocked her at.Minsk1980) recasts Freud'otions into the terminology of mental agents acting as censors to violent orsexual thoughts.In Minsky'erms, certain mental agents are good at combining ambiguous interpreta-tions, but other agents notice that this is not the normal mode of operation, and act to censor them.Thelaughter response serves to `shakp' the mind, get it back on track, and post a warning to athoughts. Presumablyhe simultaneous combinations that sneak by uncensored are ones that do not repre-sent `dangerous' modes of thought.pretation in `Normal' LanguageThere are also cases of combined simultaneous interpretation which don'nv ooetic license or puns.Consider the use ofbookin (31).Bookis polysemous between a physical object, a string of words, and anabstract plot or sequence of situations.The use ofbeautifully boundrefers to the physical object,one newidearefers to the abstract content, and50,000 wordsrefers to a particular (abstract) instantiation of the con-tent. (Ifthe book were reprinted in paperback it would still have the same number of words, whereas if itwere translated into another language, it would have a dferent number of words, but the same number ofideas.) Allthree polysemous interpretations ofbookare used simultaneously.(31) This book, although beautifully bound, contains only one neords.(32) He is the author of oIt can'etbookis a single sense implying all these aspects, because in (32),bookmust refer only to the`plot or sequence of ideas' sense.One could not felicitously use (32) to describe someone who had writteningle book which has had a hundred copies printed, or a single book which was translated by others intoundred languages.Len Talmy (1977) provides a good example of image combination in non-ambiguous language.In(33), the single interpretation is `she tra,her path displaying a topology similar to a leaf wafting through air.' (33) forces the reader to combine theimage of a woman walking through a party with the image of a leaf wafting through the air (or somethingsimilar) to arrive ahe result.xplains just what properties of the verb are maintained, and which are taken from the complements.(33) She wafted through the party.Image combination is more obvious in the case of metaphors and cliches where the deriremohan the surface form.Compare (34), which is a consistent use of metaphorSentence(35) provides a topological clash that cannot easily be resolved into a single interpretation, emeaning of the tw(34) I've ays been 100% behind my husband, pushing him on as best I can.(35) I've ays been at my husband'(36) TheSentence (36), taken from a newspaper article on real estate prices, is an example of a mixed metaphor withrying effectibe striving to get out of a rat-hole, not out from under it.ConclusionIn this paper I have inv eated segies for pragmatic interpretation, and have presented a newstrategy which (1) accounts for the little-mentioned phenomenon of a simultaneous combination of ambigu-ous interpretations, (2) is not inconsistent with experimentally deriuses a combination mechanism that is needed for non-ambiguous language as well.yBurli-Storz, Claudia,Deliberate Ambiguity in Advertising,Bern, 1980.Frazieryn and Janet Dean Fodorgnition,6(4), December 1978, 291-325.Freud, Sigmund,nd Their Relation to the Unconscious,y, 1960, Nor-ton, NYHobbs, Jerry'oceedings of the Second West Coast Conference on Formal Lin-guitics,Stanford Uni,1983.Hobbs, Jerryaul Martin, ``Local pragmatics,oceedings of the International Joint Conference onArtiŒcial Intelligence,1987.Hockett, Charles F., ``Two models of grammatical description,'' Word, 1954, 386-399.Jackendoff, Ray,Consiousness and the Computational Mind,MIT Press, 1987.Kimball, John, ``Seace parsing in natural language,gnition,2(1), 1973, 15-47.rtzman, Hord S., `xperimental study'ingLakoff, George, and Mark Turner,Morto appearMinsky, Marvin, ``Jokes and the logic of the cognitive uAI memo No.603,MIT AI Lab,1980.Schubert, Lenhart K., ``On parsing preferences,oceedings of Coling84,Stanford, CA, July 1984.Schubert, Lenhart K., `re there preference trade-offs in Attachment Decisions?,'' Proceedings of AAAI,1986, 601-605.Stallard, David, ``The logical analysis of lexical ambiguity5th Annual Meeting of the ACL, 1987,179-185.eonard, ``Rubber-sheet cognition in language,go Linguistics Society,13