1、Proceedings of the ACL 2007 Demo and Poster Sessions, pages 3740,Prague, June 2007. c2007 Association for Computational LinguisticsDont worry about metaphor: affect extraction for conversational agentsCatherine Smith, Tim Rumbell, John Barnden, Bob Hendley, Mark Lee Hobbs, 1990; Mar-tin, 1990; Mason
2、, 2004; Narayanan, 1999; Veale,1998). The e-drama genre exhibits a variety of typesof metaphor, with a signi cant degree of linguisticopen-endedness. Also, note that our overarching re-search aim is to study metaphor as such, not just howit arises in e-drama. This increases our need for sys-tematic,
3、 open-ended methods.2 Metaphor and AffectConveying affect is one important role for metaphor,and metaphor is one important way of conveyingaffect. Emotional states and behavior often them-selves described metaphorically (Kcurrency1ovecses, 2000;Fussell 33K words across all transcripts).There are oth
4、er speci c, theoretically interestingmetaphorical phenomena arising in e-drama that areimportant also for discourse in general, and plausi-bly could be handled reasonably successfully in anICA using current techniques. Some are:1) Casting someone as an animal. This often con-veys affect, from insult
5、ingly negative to affection-ately positive. Terms for young animals (piglet,wolf cub, etc.) are often used affectionately, even37when the adult form is negative. Animal words canhave a conventional metaphorical sense, often withspeci c affect, but in non-conventional cases a sys-tem may still be abl
6、e to discern a particular affectiveconnotation; and even if it cannot, it can still plausi-bly infer that some affect is expressed, of unknownpolarity (positivity/negativity).2) Rather similarly, casting someone as a monsteror as a mythical or supernatural being, using wordssuch as monster, dragon,
7、angel, devil.3) Casting someone as a special type of human, us-ing words such as baby (to an adult), freak, girl(to a boy), lunatic.4) Metaphorical use of size adjectives (cf. Sharoff,2006). Particularly, using a little X to convey af-fective qualities of X such as unimportance and con-temptibility,
8、 but sometimes affection towards X, andbig X to convey importance of X (big event) orintensity of X-ness (big bully) and X can itselfbe metaphorical (baby, ape).Currently, our system partially addresses (1), (2) and(4).3 Metaphor Recognition like (including looks like) 81% (35 out of 43).In order to
9、 detect signals we use the GrammaticalRelations (GR) output from the RASP robust parserBriscoe et al., 2006 This output shows typed word-pair dependencies between the words in the utter-ance. E.g., the GR output for You are a pig is:|ncsubj| |be+_vbr| |you_ppy| |_|xcomp| _ |be+_vbr| |pig_nn1|det| |p
10、ig_nn1| |a_at1|For an utterance of the type X is/are Y the GRs willalways give a subject relation (ncsubj) between Xand the verb to be, as well as a complement re-lation (xcomp) between the verb and the noun Y.The structure is detected by nding these relations.As for you Y, Rasp also typically deliv
11、ers an easilyanalysable structure, but unfortunately the POS tag-ger in Rasp seems to favour tagging Y as a verb e.g., cow in You cow. In such a case, our systemlooks the word up in a list of tagged words that formspart of the RASP tagger. If the verb can be taggedas a noun, the tag is changed, and
12、the metaphoricitysignal is deemed detected. Once a signal is detected,the word(s) in relevant positions (e.g. the Y posi-tion) position are pulled out to be analysed. Thisapproach has the advantage that whether or not thenoun in, say, the Y position has adjectival modi ersthe GR between the verb and
13、 Y is the same, so thedetection tolerates a large amount of variation. Anysuch modi ers are found in modifying relations andare available for use in the Analysis Component.3.2 The Analysis ComponentWe con ne attention here to X is/are Y and You Ycases. The analysis element of the processing takesthe
14、 X noun (if any) and Y noun and uses Word-Net 2.0 to analyse them. First, we try to determinewhether X refers to a person (the only case the sys-tem currently deals with), partly by using a speci edlist of proper names of characters in the drama andpartly by WordNet processing. If so, then the Y and
15、remaining elements are analysed using WordNetstaxonomy. This allows us to see if the Y noun in oneof its senses is a hyponym of animals or supernaturalbeings. If this is established, the system sees if an-other of the senses of the word is a hyponym of theperson synset, as many metaphors are already
16、 givenas senses in WordNet. If different senses of the givenword are hyponyms of both animal and person, othercategories in the tree between the noun and the per-son synset may provide information about the eval-uative content of the metaphor. For example theword cow in its metaphorical usage has th
17、e un-pleasant person synset as a lower hypernym, whichheuristically suggests that, when the word is used ina metaphor, it will be negative about the target.There is a further complication. Baby animalnames can often be used to give a statement a moreaffectionate quality. Some baby animal names sucha
18、s piglet do not have a metaphorical sense in Word-38Net. In these cases, we check the words gloss to seeif it is a young animal and what kind of animal itis. We then process the adult animal name to seek ametaphorical meaning but add the quality of affec-tion to the result. A higher degree of con de
19、nce isattached to the quality of affection than is attachedto the positive/negative result, if any, obtained fromthe adult name. Other baby animal names such aslamb do have a metaphorical sense in WordNet in-dependently of the adult animal, and are thereforeevaluated as in the previous paragraph. Th
20、ey arealso agged as potentially expressing affection butwith a lesser degree of con dence than that gainedfrom the metaphorical processing of the word. How-ever, the youth of an animal is not always encodedin a single word: e.g., cub may be accompaniedby speci cation of an animal type, as in wolf cu
21、b.An extension to our processing would be required tohandle this and also cases like young wolf or babywolf.If any adjectival modi ers of the Y noun were rec-ognized the analyser then goes on to evaluate theircontribution to the metaphors affect. If the analysernds that big is one of the modifying a
22、djectivesof the noun it has analysed the metaphor is markedas being more emphatic. If little is found the fol-lowing is done. If the metaphor has been tagged asnegative and no degree of affection has been added(from a baby animal name, currently) then little istaken to be expressing contempt. If the
23、 metaphorhas been tagged as positive OR a degree of affectionhas been added then little is taken to be expressingaffection.4 Examples of Course of ProcessingYou piglet:(1) Detector recognises the you Y signal with Y =piglet.(2) Analyser nds that piglet is a hyponym of an-imal.(3) Piglet does not hav
24、e person as a WordNet hy-pernym so analyser retrieves the WordNet gloss.(4) It nds young in the gloss (a young pig) andretrieves all of the following words (just pig theanalysis process is would otherwise be repeated foreach of the words captured from the gloss), and ndsthat pig by itself has negati
25、ve metaphorical affect.(5) The input is labelled as an animal metaphorwhich is negative but affectionate, with the affectionhaving higher con dence than the negativity.Lisa is an angel:(1) Detector recognises the X is Y signal with Y =angel, after checking that Lisa is a person.(2) Analyser nds that
26、 angel is a hyponym of su-pernatural being.(3) It nds that in another sense angel is a hyponymof person.(4) It nds that the tree including the person synsetalso passes through good person, expressing posi-tive affect.(5) Conclusion: positive supernatural-being metaphor.Results from Some Other Exampl
27、es:You cow , theyre such sheep : negativemetaphor.You little rat : contemptuous metaphor.You little piggy : affectionate metaphor with a neg-ative base.Youre a lamb : affectionate metaphor.You are a monster : negative metaphor.She is such a big fat cow : negative metaphor, in-tensi ed by big (curren
28、tly fat is not dealt with).5 Concluding RemarksThe demonstrated processing capabilities make par-ticular but nevertheless valuable contributions tometaphor processing and affect-detection for ICAs,in e-drama at least. Further work is ongoing on thefour speci c metaphorical phenomena in section 3as w
29、ell as on other phenomena, such as the vari-ation of conventional metaphorical phraseology bysynonym substitution and addition of modi ers, andmetaphorical descriptions of emotions themselves.As many extensions are ongoing or envisaged,it is premature to engage in large-scale evaluation.Also, there
30、are basic problems facing evaluation.The language in the e-drama genre is full of mis-spellings, texting abbreviations, acronyms, gram-matical errors, etc., so that fully automated evalua-tion of the metaphorical processing by itself is dif-cult; and application of the system to manuallycleaned-up u
31、tterances is still dependent on Rasp ex-tracting structure appropriately. Also, our own ul-timate concerns are theoretical, to do with the na-ture of metaphor understanding. We are interestedin covering the qualitative range of possibilities andcomplications, with no strong constraint from their39fr
32、equency in real discourse. Thus, statistical evalua-tion on corpora is not particularly relevant except forpractical purposes.However, some proto-evaluative comments thatcan be made about animal metaphors are as fol-lows. The transcripts mentioned in section 2 (33Kwords total) contain metaphors with
33、 the followinganimal words: rhino, bitch, dog, ape, cow, mole,from 14 metaphorical utterances in all. Seven ofthe utterances are recognized by our system, andthese involve rhino, dog, ape, mole. No WordNet-based metaphorical connotation is found for therhino case. Negative affect is concluded for bi
34、tch,dog and cow cases, and affect of undetermined po-larity is concluded for ape and mole.The system is currently designed only to do rela-tively simple, specialized metaphorical processing.The system currently only deals with a small mi-nority of our own list of metaphoricity signals (seesection 3.
35、1), and these signals are only present in aminority of cases of metaphor overall. It does notdo either complex reasoning or analogical structure-matching as in our own ATT-Meta metaphor sys-tem (Barnden, 2006) or the cited approaches of Fass,Hobbs, Martin, Narayanan and Veale. However, weplan to eve
36、ntually add simpli ed versions of ATT-Meta-style reasoning, and in particular to add theATT-Meta view-neutral mapping adjunct feature toimplement the default carry-over of affect (see sec-tion 2) and certain other information, as well as han-dling more signals.Other work on metaphor has exploited Wo
37、rdNet(see, e.g., Veale, 2003, and panel on Figurative Lan-guage in WordNets and other Lexical Resourcesat GWC04 (http:/www.fi.muni.cz/gwc2004/.Such work uses WordNet in distinctly different waysfrom us and largely for different purposes. Our sys-tem is also distinctive in, for instance, interpreting
38、the contribution of size adjectives.AcknowledgmentsThe research has been aided by Sheila Glasbey andLi Zhang, and supported by ESRC/EPSRC/DTI Pac-cit LINK grant (ESRC RES-328-25-0009) and EP-SRC grant EP/C538943/1.ReferencesJohn Barnden. 2006. Arti cial Intelligence, FigurativeLanguage and Cognitive
39、 Linguistics. In G. Kristiansenet al. (Eds), Cognitive Linguistics: Current Applica-tions and Future Perspectives, 431 459. Berlin: Mou-ton de Gruyter.Ted Briscoe, John Carroll and Rebecca Watson. 2006.The Second Release of the RASP System. In Procs.COLING/ACL 2006 Interactive Presentation Sessions.
40、Sydney, Australia.Dan Fass. 1997. Processing Metaphor and Metonymy.Greenwich, Connecticut: Ablex.Susan Fussell & Mallie Moss. 1998. Figurative Lan-guage in Emotional Communication. Social and Cog-nitive Approaches to Interpersonal Communication.Lawrence Erlbaum.Andrew Goatly. 1997. The Language of M
41、etaphors.Routledge, London.Jerry Hobbs. 1990. Literature and Cognition. CSLI Lec-ture Notes, 21, Stanford University, 1990.Zoltan Kcurrency1ovecses. 2000. Metaphor and Emotion: Lan-guage, Culture and Body in Human Feeling. Cam-bridge University Press, Cambridge.James Martin. 1990. A Computational Mo
42、del ofMetaphor Interpretation. Academic Press.Zachary Mason. 2004. CorMet: A computational,corpus-based conventional metaphor extraction sys-tem. Computational Linguistics, 30(1), 23 44.Andreas Musolff. 2004. Metaphor and political dis-course: Analogical reasoning in debates about Eu-rope. Palgrave
43、Macmillan.Srini Narayanan. 1999. Moving right along: A compu-tational model of metaphoric reasoning about events.Procs. National Conference on Art. Int., 121 128.Serge Sharoff. 2006. How to Handle Lexical Semanticsin SFL: A Corpus Study of Purposes for Using SizeAdjectives. System and Corpus: Explor
44、ing Connec-tions. Equinox, London.Tony Veale. 1998. Just in Time analogical mapping, aniterative-deepening approach to structure-mapping. InProcs. 13th European Conference on Art. Intell.Tony Veale. 2003. Dynamic Type Creation in MetaphorInterpretation and Analogical Reasoning: A Case-Study with Wor
45、dNet. In Procs. International Confer-ence on Conceptual Structures (Dresden).Li Zhang, John Barnden, Bob Hendley & Alan Walling-ton. 2006. Exploitation in Affect Detection in Impro-visational E-Drama. In Procs. 6th Int. Conference onIntelligent Virtual Agents: Lecture Notes in ComputerScience, 4133, 68 79. Springer.40
