{ "paper_id": "J76-1007", "header": { "generated_with": "S2ORC 1.0.0", "date_generated": "2023-01-19T02:52:29.113277Z" }, "title": "", "authors": [], "year": "", "venue": null, "identifiers": {}, "abstract": "The first three l i n e s w i l l e v e n t u a l~y form me bulk of a Micro-Planner progrwhich, when evaluated w i l l seek an object X that is a block, is equidimenoPona1 (EQDIM) and is red (w h e r e \"red\" itself has a definition, the system that restricts its application to objects w i t h the feature PHYSOB) The l a s t line of the ftgure is a set of 'semantic features\" read off right to left from the following feature tree\" The semantic structure of \"the red kuheJ\"ccan be used by C:?e deductive \"&WING-c o a p n e n t of t h e system, hefore eb*aluatfon r e s u l t A n g i n the actual picking up, to see i f such an object 1s J E it wers I I L .~, (an '\"~y~idinestsiannl pyramid1\" would not be) t h e system souLd yo dncl try ta re-pitrse N r n t h e sentcnco. The meaning of verbs i n SHRDLU is mre mmplex. The seaantic c~w p n e n t has access to n definition fer \"pick-up\" just as it d~e s for \"tt5d1' a:~d \"klcck\" and this definition will enable SHRDLU to translate \"pick-up tateaents\" into Micro-planner in a mdnner analogous to that for noun phrases. These are two complications here. Firstly \"pick-up\", u n l i k e \"red\",", "pdf_parse": { "paper_id": "J76-1007", "_pdf_hash": "", "abstract": [ { "text": "The first three l i n e s w i l l e v e n t u a l~y form me bulk of a Micro-Planner progrwhich, when evaluated w i l l seek an object X that is a block, is equidimenoPona1 (EQDIM) and is red (w h e r e \"red\" itself has a definition, the system that restricts its application to objects w i t h the feature PHYSOB) The l a s t line of the ftgure is a set of 'semantic features\" read off right to left from the following feature tree\" The semantic structure of \"the red kuheJ\"ccan be used by C:?e deductive \"&WING-c o a p n e n t of t h e system, hefore eb*aluatfon r e s u l t A n g i n the actual picking up, to see i f such an object 1s J E it wers I I L .~, (an '\"~y~idinestsiannl pyramid1\" would not be) t h e system souLd yo dncl try ta re-pitrse N r n t h e sentcnco. The meaning of verbs i n SHRDLU is mre mmplex. The seaantic c~w p n e n t has access to n definition fer \"pick-up\" just as it d~e s for \"tt5d1' a:~d \"klcck\" and this definition will enable SHRDLU to translate \"pick-up tateaents\" into Micro-planner in a mdnner analogous to that for noun phrases. These are two complications here. Firstly \"pick-up\", u n l i k e \"red\",", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Abstract", "sec_num": null } ], "body_text": [ { "text": "To euntey an enewetic field like thie one is inevitably to laavo a great deal of excellent work unextiminad, a t least if one i a going to do more than give a paragraph to each research project.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "I have left out of cotasideration at least seven groups of projects:", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "(1) Early work in Artificial Intelligence and Natural Language that has been sumeyed by Wfnograd (1973) and Simmons (1970a) among others.", "cite_spans": [ { "start": 108, "end": 123, "text": "Simmons (1970a)", "ref_id": null } ], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "(2) Work by graduate students of, or intellectually dependent upsn that o f , people discussed in same detail here.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "( 3 )", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "Wxk that derives essentially from projects described i n detail here. This embraces several groups interested i n testing psychological hypotheses, as r e 1 1 as others constructing largescale systems for speech recognition. I have devoted no space to speech recognition as such here, for it seems to me to depend upon the quality of semantic and inferential understanding as much as anything, and so I have concentrated upon this more fundamental task.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "Work on language generators, as opposed to analysers and understanders.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "They are essential for obtaining any testable output, but are thearetically secondary. That would be wholly inappropriate i n the present s t a t e of things. A g r e a t deal o f work is being done a t the moment, and many of the p r i n c i p a l researchers change t h e i r views on very fundamental questions between one paper and t h e next without drawiw any a t t e n t i o n t o the f a c t . Cheap self-contradictions and changes of mind are a l l too easy t o f i n d , so c r i t i c i s m and smparisons are best drawn w i t h a very broad brush and a l i g h t stroke.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null }, { "text": "Winograd's systea runs a s a dialogue, i n r e a l time, between a human Winogradas parsing is top down, and depth f i r s t , with na automatic back up. The parsing progrim fur each griuxnatical c a t c q~r~~ is n functional definition in PRfXZWWR, which can be stated either a s &VP, fcr SEWPEHCE, or as a f l o w -c h a r t as below for VP:", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Winograd's Understanding System", "sec_num": "2." }, { "text": "DEFIWJ pxqram VP R E m f a i l u r e Yes L", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Winograd's Understanding System", "sec_num": "2." }, { "text": "Here is Winograd's own account of the start of this top-down parsing procedure for the sentence \"Pick up a red block\" (where t h e material i n C 1 is added explanation and not Winogradas o m ) :", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Winograd's Understanding System", "sec_num": "2." }, { "text": "\"The CLRUSE program looks at the first wrd, to decide what w r i t the", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Winograd's Understanding System", "sec_num": "2." }, { "text": "If it sees an adverb, i t assumes the sentence begins w i t h a single-word modifier tslowly, Jack l i f t e d the book] ; iP it sres a preposition, it looks .for an i n i t i a l PREP6 Con top of the h i l l stood a tree]", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "CLRUSE begins w i t h .", "sec_num": null }, { "text": "If it sees a BINOBR, it calls the CLAUSE program to look for a BOUND CLAUSE Since the next word is not: \"do\" it checks the n e x t word i n the input ( i n U~is case still the first word) : o see whether it is the i n f i n i t i v e form of a verb.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "CLRUSE begins w i t h .", "sec_num": null }, { "text": "C~efore", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "CLRUSE begins w i t h .", "sec_num": null }, { "text": "If SO, it is to be attached to the parsing tree, and given the additionql feature MVB (main verb). The current structure can be diagramad as:", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "CLRUSE begins w i t h .", "sec_num": null }, { "text": "(CLAUSE MAJOR) (VG IWER) (VB MVB INF TRANS ,WRT --------------pick", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "CLRUSE begins w i t h .", "sec_num": null }, { "text": "T W S AND VPRT cmfw fran the definition of the word \"pick\" when we zalled the function P M S q for a word.4'", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "CLRUSE begins w i t h .", "sec_num": null }, { "text": "Mter this syntactic parsing, a number of \"semantic specidlists'' attach cwbantic structures to specific syntactic dnes.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "CLRUSE begins w i t h .", "sec_num": null }, { "text": "I n the case of \"a red cube\", the f o l l o~i n g structure i s built up by an NP \"semantic specialist\"", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "A semantic d e f i n i t i o n of an", "sec_num": null }, { "text": "Secondly, there are two rrpes of verb definition, s e m a n t i c and inferential", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "A semantic d e f i n i t i o n of an", "sec_num": null }, { "text": "Winograd does not give the semantic definition for \"pick-up\", but here is t h e one f o r \"grasp\" which is a closelb-related verb. ", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "A semantic d e f i n i t i o n of an", "sec_num": null }, { "text": "SIZE LOCATlcW \\' corn MIbWTB ---------------------- M t w (--------- C ' 5 m RED U C ) f \\NITS CaEM bTAC # PifiSoB--( '~~~S~~~~------------------------ rrg& , - I N ( --I T~U FkW1Z.P", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "(CMEANS ( ( ( M A N I~T E ) ) (~# & w I P J ) )", "sec_num": null }, { "text": "which says e~sentiallp that grasping is scmathinq dorle by an animate entity to a oanlpulabla one (flret line).", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "(CMEANS ( ( ( M A N I~T E ) ) (~# & w I P J ) )", "sec_num": null }, { "text": "More of the real content of such actions is found in their inferential definition.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "(CMEANS ( ( ( M A N I~T E ) ) (~# & w I P J ) )", "sec_num": null }, { "text": "Here is the one fox '\"pick-up\":", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "(CMEANS ( ( ( M A N I~T E ) ) (~# & w I P J ) )", "sec_num": null }, { "text": "(CONSE TC -PICKUP I X ) (PJCKUP xr ( r n ( W S P ?X) T i i E o r n ) (C.XXAL (RAISEHAND THEOREMS)", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "(CMEANS ( ( ( M A N I~T E ) ) (~# & w I P J ) )", "sec_num": null }, { "text": "TSlis definition allows the program to actually carry out the \"pick-up\" caamand if it is possible to do so in the simulated world, as it would not be, for example, if -re were already a block dn top of the red one..", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "(CMEANS ( ( ( M A N I~T E ) ) (~# & w I P J ) )", "sec_num": null }, { "text": "such as GRASP and RAISEHAND each of which must be carried out in order that sameming may indeed be p i c k e d up.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "PICKUP is being d e f i l e d in edzrms of a number of more p r i m i t i v e s&-actions,", "sec_num": null }, { "text": "fesential definitions: the one given for GRASP, for example, i s somewhat differant from i t s \"C1(&EANS1' definition g i v e n above, although the inferential iPefini,tiona are aim, i n sane se-e, definitions of meaning as wall as prog r a~a s for actually carrying out the associated conmands.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "There s u b -a c t i o~ themselves have in-", "sec_num": null }, { "text": "One reason for the enormous impdct of this work was that, prior to its appearance, A 1 work was not very l i n g u i s t i c a l l y interesting, while the eystems of tho linguists had no place far the use of inference and real world knowledge.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "There s u b -a c t i o~ themselves have in-", "sec_num": null }, { "text": "was able to breed considerable results. where it is c l e a r that the answer is both definite, and that fhding i t requires some inferential manipulation of genaralirrtiens d w u t the world.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Thus a very limited union between the t w o techniques", "sec_num": null }, { "text": "3ha reader should ask himself at this j m i n t h o w he kwwe the nsrxact refemnt of the pmnaun in that sentanca.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Thus a very limited union between the t w o techniques", "sec_num": null }, { "text": "$ m e Discussion of S W L U S~J -fax, the reaction ts Winugqad \\s work has L m t k wbSrSrla ~ineri t heal. What would crAitics find to attack i f tthtzp M~S F 90 a9nde.I: F i r s t l y , that ~~i n e g r a d ' s linguistic system is highly w n s r~~? a t i w , and that UIQ distinction between 'syntax' and 'semantics' m y not L w necessary at all.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "3.", "sec_num": null }, { "text": "a way *.at m u l d make it fnextensihle, to any genexal, real tmrld, situatioh.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "#at h i s semantics is tied tu the s h p l e referential w r l d cf t h e b l c c k s i t ?", "sec_num": null }, { "text": "Suppose 'block' were allowed mean 'an obstrxxctidnl and ' a nental ir,k,i*i t i o n ' , as well as ' a cubic object'. It is dauktful whether Winograd's features and rules could express t h e ambiguity, and, nore i m~r t a n t l y ,", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "#at h i s semantics is tied tu the s h p l e referential w r l d cf t h e b l c c k s i t ?", "sec_num": null }, { "text": "whether the simple s t r u c t u e s he manipulated c o u l d decide correctly between t h e a l t e r n a t i v e meanings i n any given c~n t e x t of user Again, f a r more sophisticated and systematic case structures than those hg used might be needed to resolve the ambiguity of 'in1 in \"He ran the mile i n five minutes and Ire ran t h e m i l t . in a pawr kwg , as we11 as 'tllu ccmbination ~, l f case w i t h w r d sense a m h i p i t y i n 'Ho p u t t h o key in t h e l~r k ' (door lock1 and 'Be threw the key in the luck1 (river leek).", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "#at h i s semantics is tied tu the s h p l e referential w r l d cf t h e b l c c k s i t ?", "sec_num": null }, { "text": "The b l o c k s mrld is also strongly deduct-ive and l w i c a l 1 y closed, S f g r a v i t y were introduced into i t , then anything supkwrted t h a t war pushed i n a c e r t a i n way would have, logically have, t o f a l l . B u t t h e cazmon sense wlorld, of ordinary language, i s n o t like that: in t h~ 'waaen and soldiers'", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "#at h i s semantics is tied tu the s h p l e referential w r l d cf t h e b l c c k s i t ?", "sec_num": null }, { "text": "example given e a r l i e r , .the pronoun ' s e v e r a l 1 can be s a i d to be resolved u s i n g same generalisation such as ' t h i n g s shot a t and hurt tend t o fall'", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "#at h i s semantics is tied tu the s h p l e referential w r l d cf t h e b l c c k s i t ?", "sec_num": null }, { "text": "There are no logical 'have to's1 t h e r e , even though the meaning of the ~r onoun i s perfectly d e f i n i t e .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "#at h i s semantics is tied tu the s h p l e referential w r l d cf t h e b l c c k s i t ?", "sec_num": null }, { "text": "Indeed, it might be argued that, in a sense, and as r~a f i s its semantics, Winograd's system is n o t about n a t u r a l language a t a l l , but a b u t the -technical question of how goals and sub-goals are to be crganised in a problem-solving system capable of mani~ulating simple physical ehjects.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "#at h i s semantics is tied tu the s h p l e referential w r l d cf t h e b l c c k s i t ?", "sec_num": null }, { "text": "If n reeqber, for example, that tfre key problem that brought b w n the emrr#rur work on mchiae translation i n the Fiftiea and Sixties, was that of the eensa d i g u i t y of nattWi.l. language wprds, then we will look in vain to SHRDLU Lor any help w i t h t h a t problem. There seem? to be only one dear exmaple of aur aLnbiguous mrd in the whole system, namely that of 'curitairat as it appears in 'The b o x contains a red block' and rhe stack", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "#at h i s semantics is tied tu the s h p l e referential w r l d cf t h e b l c c k s i t ?", "sec_num": null }, { "text": "Again, i f on@ glahces back a t the definition of 'pick-up' quoted &ve, ana can see t)cdt it. i e in fact an e%pression rrE a prcrcedure f~r picking up an ob5ect i n the SHR~LU .yet=.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "#at h i s semantics is tied tu the s h p l e referential w r l d cf t h e b l c c k s i t ?", "sec_num": null }, { "text": "one, understand t h e p~r f e~t l v ordinary sant;ence 'I picked up my bags \u00a3tom the plstforn, and ran for the train', let alone any sentenco n o t &out a physical action performable by the hearer. T ' k difft!zeace i s t h a t Molods t a k e s a much more logico-sanantic i n t q r e t a t i u a of t h a t s l o g a n than does Winoqrad.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Nothing about it, for ekample, would help", "sec_num": null }, { "text": "In partisular, fbr W s t h e meaniq of an L n p~t utterance to h i s system is t h e procedures w i t h i n t h e system that raJ+ipulata tfr t r u t h conditions of t h e utterancfe and e s t g b l i s h its t r u t h value.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Nothing about it, for ekample, would help", "sec_num": null }, { "text": "To p u t t h e m t t e x crudely, f o r Moods an assertion has x i meanhq if hie.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Nothing about it, for ekample, would help", "sec_num": null }, { "text": "system cannot e t a b l i g h its t n t h o r falsity. should show, as it were how i t is a c t u a l l y t o be applied to l a n g u a~e , b u t that is n o t the s-e as demanding that it should be w r i t t e n in a ~r~c~d u r a 3 language,line PLANNER. I shall r e t u r n to this last pifit l a t e r . The i m p r t a n t wards these are 'lt-k f e y ' , which suggest that t 3 1 r > r~ may w e l l bo c o n f i r m i n g hints to be found in the :=tory and, i f t4lerc are, than this tentative, partial, inference is cursect, and TE a c h i n of demons can 'reacht one of the passiblo xaferente i n a story then there is a suct+ass registered and the ambiguity of the corresponding pronoun is resolved.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Nothing about it, for ekample, would help", "sec_num": null }, { "text": "It can be seen that the information encoded in the system is of a highly specific sortin the present case it is not about containers as such, and how to g e t their contents out, but about Piggy B a n k s in particular, and everything relies on that partfcular knowledge having been put i n . Thus, the apparently English words in the PB-OUT-OF Pound b e then tied directly, or via these inference rules, to response patterns which are generated. (Hendrix et a1 l73) , the ather apprdach is adopted, w i n g a primitive action B&CI.fANGG inetead of 'transfer'.", "cite_spans": [], "ref_spans": [ { "start": 447, "end": 466, "text": "(Hendrix et a1 l73)", "ref_id": null } ], "eq_spans": [], "section": "Second Generation Svstems", "sec_num": "5." }, { "text": "Enormous", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Second Generation Svstems", "sec_num": "5." }, { "text": "ThIe implementation under c~nstruction is a front-end ,oarsex of the Woodat augmented t r a n s i t i o n network type (see Woods '701, and a ganefation system going gram the s e a~t i c networks to surface strings described8in detail i n ( S W n s and S1ocu.m ' 7 2 ) .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Second Generation Svstems", "sec_num": "5." }, { "text": "SimPons has also given considerable an & take + book Here 'pf itldicatea past, and is the aepndendy symbol liking a PP to We ACT ( ' t a k e ' ) which i s the hub of the conceptualization, as w i t h Simon& ?'he ' 0 ' indicates the objective case, marking the dependence of the object PP on the central ACT.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Second Generation Svstems", "sec_num": "5." }, { "text": "There is a carefully constructed syntax of linkages between the conceptual categories* that will be describpd only in part in what follows.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Second Generation Svstems", "sec_num": "5." }, { "text": "The next stage of the notation involves an extended case notation and a set of primitive ACTS, as well as a n q e r of it:ems suoh as PHYSWNT which indicate ather stqtes, and items of a fairly simplified psychological t h e o r y (the dictionary entry for 'advise', for example, contains a subgraph t e l l i n g us that Y 'will benefit' as part of the meaning of ' X advises Y ' can consider t h i s entzy als an active 'frAxte-like~ object seeking f i l l e r itms i n any context In which it is activated. Thus, in the sentence 'John sho the girl w i t h h riflet, the variables w i l l be f i l l e d in frcm context and the case inference will be made f m the main act PROPEL, which is that its h s t r u u e n t is lkSOVEI GRASP or PFtOPRL, and so we w i l l arrive a t the whole conceatualieation: Sot in building a template for 'John drinks,wine', the whole of the above tree-formula for 'drinks' would be piaced at the central action node, another tree structure for 'John' at the agent node and so on.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Second Generation Svstems", "sec_num": "5." }, { "text": "John PROPEL 4 - bullet <-", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Second Generation Svstems", "sec_num": "5." }, { "text": "The complexity of the system comes from the way i n which the formulas, considered as active entities, dictate how other places hn the same template should be filled.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Second Generation Svstems", "sec_num": "5." }, { "text": "Thus, the 'drink1 formula above can be thought of as an entity that fits at a template action node, and seeks a liquid object, that is ~ say a f~rmula w i t h (FLOW STUFF) as its right-most bzanch, to put at the object noda of the same template. This seeking is preferential, in that formulas not satisfying that requirement will be accepted. but only if nothing misf a c t a n c a lXZ'-fotUEa. TIie -\u20acElliplate Uif ly esWIisned Tm 3 Tragm e n t of t e x t is the one in which the most formulas hive their preferences s a t i s f i e d . a general principle at work here, t h a t the r i g h t interpretation 'says the least1 in inforreation-carrying terns, T ) r h wry simple device is able to do much o f the work o f a syntax and wxdysnse wzibigukty resa1vi;tap pxagraa P n a r c u s q~e , LZ the a,mteme k d been 'John drank s whole pitcher1, the fomulr tor th. 'pitcher of klquidb w u l d hawe h e n pr.rsEerreCI to that for thar human, s f m the subf~mS;a (FLOW STUFF) could be apprtopriateAy located uithirr i t . The extracted templates express information a lready implicitly preser.t i n the text, wen though many of them are partial inferences: anes that may not necessaxily, be true. Whether or not such a, system can remain s-le with a WrutderabLe vecabulary. of say several thousand words, has y e t to be t r r t d .", "cite_spans": [], "ref_spans": [ { "start": 538, "end": 593, "text": "a general principle at work here, t h a t the r i g h t", "ref_id": null }, { "start": 773, "end": 963, "text": "P n a r c u s q~e , LZ the a,mteme k d been 'John drank s whole pitcher1, the fomulr tor th. 'pitcher of klquidb w u l d hawe h e n pr.rsEerreCI to that for thar human, s f m the subf~mS;a", "ref_id": "FIGREF12" } ], "eq_spans": [], "section": "Second Generation Svstems", "sec_num": "5." }, { "text": "ft will ba ovidrnt tso any reader that Zha laat t w a systems described, Bch.nktm ud my moun, share a great deal in cccpmon. whether we ahould expect advance in understanding natural language from those tackling the problems head on, or those coroncerned to build a 'fr8nt", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "There is", "sec_num": null }, { "text": "andv. It i~ cLtt,xly the case thata n p i e c e coulL bp esrcntial to the understanding of sane story. The question is, does it follow that the epehifict.tion, organieation and formalization of that knowldge l a &a studf oP l : . a g a , because i f it is then all human enquiry f t m physics and history to medicine is a linguistic enterprise. Yet clearly, any syatQIP OP C C E P g e K ) . n sense", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "There is", "sec_num": null }, { "text": "inferences that considered such a truth when reasoning about eating would be making a mistake.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "There is", "sec_num": null }, { "text": "One might say Ulat the phenoeenolqtcrl lrvrL of the anraly_sis was m n g even thourgh all the InF'amznces it: !!ad8 ware Uue", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "There is", "sec_num": null }, { "text": "The stme w u l d be true of any W . I . system that wade everyday inferences about physical objects by mnsiQaring their quantum structure.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "ones,", "sec_num": null }, { "text": "by uovirsg the hands to the mukh, and it might be argued that yven ulis is goisrg too far ftom the '@aaningl of eating, whataver that m y bar towsrds generally true information about ma act which, if always inferred &ut a U acts of qating, w i l l carry the systesrs nruamageably fax. But Chatniak's decoupling has the effect of completely separating these two closely related liniguistic phenomena i n what seems to me an unraallstic aanner. H i s system does inferencing to resolve pronoun ambig2 uttfes, while sense ambiguity is presumably to be done in the future by and spatial loeatioqw amow others. As wa saw earlier, tiha B1serhinatian rnvolved in actual analysis is a matter of a p c i E y i n g wry delioat. M h s Q suggests 'gunt as the default value of the instrument of %he action of shooting, but I would claim that, in an example like the earlier 'He shot her w i t h a c o l t ' , we heed to be able to see in the structure assigned whether or n o t what is offered as the apparent instrument is in fact an instrument and whether it 'is the default or riot.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Schank's analysis of eating rontaias t h e inf'matian &st i t l a done", "sec_num": null }, { "text": "In other words, we need sufficient structure of application to see not only that 'shcotlng1 prefers an instrument &at is a gun, but also why it will chaose the sense of 'colt1 thatcis a gun rather than the one which i s a horse.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Schank's analysis of eating rontaias t h e inf'matian &st i t l a done", "sec_num": null }, { "text": "ATtlx>ugh Schank sametinee writes of a system making 'all possible1 inferences a8 it p10ceBd8 though a textt this i e not in fact the heart o t tho dispute, since no one would want ta defend my atmng definittior oL the tom 'all poesibla infetences'.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Schank's analysis of eating rontaias t h e inf'matian &st i t l a done", "sec_num": null }, { "text": "Chacniakqs argument 4s that, unless certain fornard inferences w e made during an analysis o f r say, a e -r yforward inferencest that is, that are not problem-driven; not made in rerrpcnss ta any particular problem of analy.ysia then known to the ayrtsmthan, ar a matter of empirical fact, the system will not in general be able t o solva srPbiguity or rofetence ptoblems that arise later, because it will never in fact be possible t o locata (while looking backwards a t the text, as i t were) the points Ohere those forward inferences ought to have been made.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Schank's analysis of eating rontaias t h e inf'matian &st i t l a done", "sec_num": null }, { "text": "This i s r in very crude summary, Charniak's case against a purely ptoblw-driven inferencer in a natural language under-", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Schank's analysis of eating rontaias t h e inf'matian &st i t l a done", "sec_num": null }, { "text": "stander ,", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Schank's analysis of eating rontaias t h e inf'matian &st i t l a done", "sec_num": null }, { "text": "A ditficulty w i t h this.argument is the location of an axample of t e x t that c~nffrms the pofnt in a ncn-contentious manner.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Schank's analysis of eating rontaias t h e inf'matian &st i t l a done", "sec_num": null }, { "text": "Chatniak has found an excerpt ftcm a book describing the l i f e of apes in which it is indeed hard to locate the reference of a particular pronoun in a given passagQ. Chamiak's case is that it is only possible to do so i f one has made eertaln inon-prublm occasioned) inferances earliez in the story.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Schank's analysis of eating rontaias t h e inf'matian &st i t l a done", "sec_num": null }, { "text": "nuabet QE readers find it quite hard tb refer that particuXe pronoun anywayl which might s w e e t that, the t e x t was simply badly written. (iv) In terns of the l f n q u i s t i c and or p s y c h c l c g f c a~ plausibility of the proffered system of representation.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "But a", "sec_num": null }, { "text": "Oversimplifying considerablyr one might say that Charniakls system akpeals mostly to (ii) and somewhat to (i) and (iv); Winogr3S1s to {iii) and scmewhat to the other three categories; Colbyls (as r e g a d s i t s natural language, rather than psychiatric, aspects) appeals almost entirely to (iii);", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "But a", "sec_num": null }, { "text": "Simmons largely to (iv) , and Sthank's and my own to dif f ereng mixtures of (ii), (iii) and (iv) .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "But a", "sec_num": null }, { "text": "In the end, of course, only (iii) counts for enpiricistsu but there is considerable d i f f i c u l t y in getting all parties to agree to the terns of a t e s t . * A cynic might say Chat, in the end, a l l these systems analyse tho setltenres a t i t they analwe orl to put the same p o i n t a l i t t l e more W w t e t i c a l l y , these is a sense i n which systerms, those described here and tho st^ elsewhereb each define a natllcleal, languaqe, namely the one to which it applies.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "But a", "sec_num": null }, { "text": "The difficult question is the extent to'which those mnv and mall natural lacpages resemble E n g l i~h .", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "But a", "sec_num": null }, { "text": "The Last section ,atreased areas of cuzrsnt disagreement, but there", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Cdnclueiqn", "sec_num": "7." }, { "text": "w~u18, i f vote& mse M e n r be tmnsiderable agreement atmt~g A.X. workers on natwdl, language about where the large problems of the immediate future a : tt,e need for a g o 4 memory mdel has been stressoa by Schank (197-la) , and m y would add the need for an extended procedufbl theory of t e x t s , rather of individual example sentences, and'far a more sophisticated theory of reasons, causes, and motAFhs for use i n a thwry of understanding.", "cite_spans": [ { "start": 216, "end": 224, "text": "(197-la)", "ref_id": null } ], "ref_spans": [], "eq_spans": [], "section": "Cdnclueiqn", "sec_num": "7." }, { "text": "Many Ptight also be pezsuaded to agree on the need to steer between the ScylLa of t r i v i a l first generation Fmplementatfons and the Charybdis of u t t e r l y fantastic ones.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Cdnclueiqn", "sec_num": "7." }, { "text": "By the lattet, f mean projectfi that have oeen sericuly dfscussed, but never implemented for obvious reasons, that would, ray, enable a dialogue program to discuss whether or not a participant fn ufvlul o-ty '$%kt qllilt~', end if 80 why.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Cdnclueiqn", "sec_num": "7." }, { "text": "n i e last disease has socpatbes had as a lrajor syrpptoCll an extensive use t -f the w r d 'praqmat-cs' (though this ern also indicate quite benign condf tioas in other cases) , along wdth the implicit claim that lsemant ics has been salved, t~) w e should get on w i t h the pragmatics'. It s t i l l needs repeatthat there bs rn sense whatever in which the semantics of natural language has been solved. It is still the enoxmaus barrier it has always been, even if a feu dents in its surface are beginning to appear here and What is maant by 'stock1 i s clearly the stock piece of the gun, but any preference system like mine that considers w e two sansss of ?stockt, and sees t h a t an edible, soup, sans@ of 'stock1 is the preferred object of the action ' t a s t e , w i l l infallibly opt for the w n s n g sense, Any \u00a3 g a m e or expectation s y s t m is pmaa to the same general k i d df countex-exmphe,", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Cdnclueiqn", "sec_num": "7." }, { "text": "In particular cases like this it is easy to suggest what might be done: here we might suggest a preference attached to the formula for any; thing t h a t was essentially pareof aslother thing (stock = 'part of gun' in @is casej, so that a local search was made whenevex the 'part-of1 e n t i t y was mentioned, and the satisfaction ofmt search w u l d always ", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Cdnclueiqn", "sec_num": "7." }, { "text": "' b m~p ----------", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "", "sec_num": null } ], "back_matter": [ { "text": "sapre other, ulti.mately remupled, s y s t e '~. *", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "annex", "sec_num": null }, { "text": "Madularity concerns the deccwposability af a firogran or system i n t o (interacting) parts, and fhe nature of the relationship between t+e parts.", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Wodulari ty", "sec_num": null }, { "text": "BegmentJ which interact in a way he describes as 'heterarchicl (as oppo$ed to 'hierarchic8) which means that different w e n t s can be in controlaat d i f foreht tiswc .Qn the other hand, S c W and W i l k s have argued that it is n o t necesaaty t a absarve efther the syntactic-semantic, or the semantic-deductive, dlatfnctlon i n an understanding program. On that view there 0 no part;icular,virtue i n integrating syntax and semantic rbutlnes, since +here was no need tm separate them. A suitable envir-ent in which to consider tke question is t h a t of translation from one language to another: suppose we are analyging a sentence containing the word 'nail1 meaning a physical object.It is clear t h a t t h e translation of that word i n t o ~r e n c h should not be the same * '~n i s p o i n t is to some extent hypothetical s i n c e , as we s a w , Schankls conceptualizations s t i l l do c c n t a i n , cr aspear t~ c~z t a i n , 3aF.y surface items; in particular nouns, adjectives an3 adverbs. Iizwever, tLis is a transitional'natter and L l e y a r e in the course of r'epkace.zezt, as noted, by non-superficial items.as the translation for 'screw or 'peg1. Y e t is it plausible that any dascription of the function of these three entities entirely i n terms of arawmtf~ prWtima1 ma without any explicit mention of the wrb name and its connection to its French equivalent, will be sufffcienG t43 ensure t h a t only the r i g h t match is made? ", "cite_spans": [], "ref_spans": [], "eq_spans": [], "section": "Winograd's program, as we saw, contains syntactic, semantic and deductive", "sec_num": null } ], "bib_entries": {}, "ref_entries": { "FIGREF0": { "type_str": "figure", "uris": null, "text": "in his report t o the Science Res~arch Council, on the s t a t o aE A r t i f i c i a l Intelligence, Sir S u e r , Lighthi11 (1973) qaw &oat cf tho field a eather bad propnosis. One of the few h o p f u l igns ho a m war Winwredls (1972) natural language wderstanding system. Yet, only e ye-car later, Winograd had s t~p p d wrk on a e system he mnstruuted, and had kwgurr A nnlar one on e n t i r s l x different principles.** He went SJ far, in a survey lecture (Winograd '737 o f extrnordinnry modesty Ln a f i e l d not krtawtl for i t s mall cwputer systerns designed to understand natural languaga, and \\rent on t c describe others second generation' systems. I shall xeturn later t o this metaphor of generations, b u t what is cne to say in general terns of a field where yesterday's brightest spots are today's f i r s t generation systems, even though they have not been criticised i n p r i n t , nor shown i n any generally acceptable yay to be fundamentally wrong? Part of the answer lies i n the profound role of fashion in Artificial Intelligence i n its present pre-scientific phase. A cynital American professor remarked recently that Artificial ~ntelligencd (AX) had an affair w i t h someonels work every year or two, and that, just as there were no reasons for galling i n love, s o , l a t e r , there were no reasons for f a l l i n g o u t again. In tho csse of Winograd's work it is i m p r t a n t now t o resist this fashiony and re-emphasize what a good piece of research i t was, as 3 shall inl a m~ment. Another part of t h e answer l i e s ,in t h e still fundamental role ofmetaphysical criticism i n A I . I n the f i e l d of computer vision things are bad enough, i n that anybody who cansee f e e l s e n t i t l e d t o c r i t i c i s e a system, on the ground t h a t he is surehe does not see using such and suck principles, In the field of natural language understanding things are worse: n o t only does anyone w h m o can speak and write feel free to criticise on the correspandina grounds, but i n addition theze are those trained i n disciplines p a r a s i t i c upon n a t u r a l language, l i n g u i s t s and logicians, who often know i n addition how things bIUST BE DONE on a priori groundsa. I t i s this metdphysical aspect of the s u h j e c t that gives its disputes their characteristically a c r b n i o u s In this paper I w a n t to sort out a little what is agreed and what is noti what are swre of the outstanding disputes and how testable are the claims being made. If what follows seems unduly philosophical, it should be remetsabered t h a t Uttleis agreed, and almost no achievements are beyond question. To pretend otherwise, by concentrating only tm the d e t a i b o f established programs, ~u l d be meretricious and misleading.", "num": null }, "FIGREF1": { "type_str": "figure", "uris": null, "text": "All t h e many and varied reasoning schemzs now a v a i l d l e i n AI, hcluding PLANNER (Hewitt 19691, QA4 fRulifson et a1 19721, MERLIb (*ore and Newell 197a as well as automatic programming (Balzer e t a1 1974) (tleidor-174) and debugging (Sussman 1974) projects, many of which are producing formalisas that appear increasingly l i k e natural l w u a a e . Conservative reasoning schames, such as f i r s t order predcicatc! calculus, thab have boen appliud ts, or i l d~~~u a have also ignored, as one musk in oldor to write a t a l l fro a rapidly changing field, Uaa i n t e r p r , e t n t i o n given t~ d4inskjw' a (1975) n o t i a n b f l\"'frmelq auring 1975 by Chsmiak and ScEaadc. During this year Lmth have product4 s k c t c h~a Eer a rrprasanta t i c n UP knuwladqrs sn a larger scallr than any 3ls;usss.l l n this survey: roughly speaking, they have each produced a schmd Pok a whole s t o r y . The value and function of sucll a larger-scale representation i s much under discussion a t the manent, bdt acne of it invalidates what! is contained here, except for sane q u a l i f i c a t i o n to the position of Schank which is noted under the heading \"Centrality\"' i n section 6. The e x c l u s i o n s under ( 2 ) above a r e p a r t i c u l a r l y u n f a i r t o the wark i n the unpublished theses of Rieger 11974) and McDesmott ( 1 9 7 4 ) , and can only be justified, l i k e those of sections (5) and (6) above, by space, bias and the considerations advanced i n a later section of this survey concerning what it is f o r A 1 research to be nlmut natural l m q u a q e , ra'tht3r than slwut. sanething else, l i k e human p s y c h~l~~~y , or logic, as the bargaining h h a v i o u r of chil-dsen under s t r e s s . One f i n a l remark &wut the s t y l e of t h i s survey. I have not indulged the n~r m a l critical luxury of c a r e f u l and extensive quotation f r o m authoys f o l l o w e d by d e t a i l e d and meticulous c r i t i c i s m .", "num": null }, "FIGREF2": { "type_str": "figure", "uris": null, "text": "who types i n m e S q e s , and t h e s y s t e m proper, called SHRDLU, which displays on a screen pictures of a closed world of coloured blocks and p y r a i d a , a b o x inwhich they can be put, and an arm or crane khat can move che objects about. SHRDLU displays its replies to the operator by writing on the samq screen. The f o l l o w i n g i s a discontinuous sample of t h e dialogue Winograd giws in his book. SHIIDLU1s replies are in the upper case r eick up the big red block, OK find a block which is taller than t h e one you are holding and put it into the box. BY IIT', I ASSUME: YOU MEAN TKE BLOCK WHICH IS TALfwER THRN Tm ONE I AM HOLDING what does the box contain THE BLUE P Y M I D RND THE BLUE BLOCK had you touched any pyramid before you put the green one on the Lkttle cube YES, TEE GREEN ONE when did you pick it up WEIILE.1 WAS STACKING tlP THE RED CUBE, A LARGE RED BLOCK AND A EARGE W E N CUBE. why TO GET R I D OF IT. Thia, ingenious program was also ahle to accept new definitions.& words, such as steeple'; to make use of what it knew about the p~operties o f t h e objects described, such as that a block could have i t s top cleared off, but a pyramid could not; and also ta remember what it had done before, as in t h e sample above. The syn-tactic analySis program was written in PR%RAMMAR, a procedural language related to PLANNER (see Hewitt ' interpreted: a sentence consists of a noun phrase followed by a verb phrase) would be expressed as: (PDEFIME SENTENCE ( ( ( PARSE NP) NIL FAIL) ( ( PAFSE VP) FAIL FAIL R E T U R N ) ) ) The d e t a i l s of the n o t a t i o n need not detairius at this p o i n t ; what Ls important is that Winograd's gramuax 4s not tile cmnventirrnaL l i s t o f culas, but small sub-programs Like tha lines above, t h a t actually xaysooant x-o~iiures for iulposing t h e desirad grstmalieA1 structure, The f i r s t leva1 of linguistic p m & u r~s in the system applies a systemic grammar' , dur to M.A. K. Mallidal* (1970) , which inapses a hierarchical structure of clauses on tihe input senterac&s\\, which secb tc' b~ dram from a vocabulary of about 175 ims\\ls.", "num": null }, "FIGREF3": { "type_str": "figure", "uris": null, "text": "you get there, w e l e f t ] . In English (and possibly a l l languages) the firsb word of a construction o f t e n gives a very goad clue as to what that construction will be. fn this case, \"pick\" is a verb, and i n d i c a t e s that we may have an IHPERATIW CLAUSE. The program. s t a r t s the VO program w i t l a the i n i t i a l VG feature. l i s t (VG IWER), looking for a VG of this type. This must either begin w i t h some form of the verb \"do\" [Do not call me!] or w i t h the main verb itself [Call me!].", "num": null }, "FIGREF4": { "type_str": "figure", "uris": null, "text": "Before Winograd &ere were few pmgrams i n A 1 that could take a reasonable complex English s e n t a c e and ascribe any s t r u c t u r e whatever to i t . In early classics of ' n g t u r a l language understanding' in AX, such as Bobrow's STUDENT (1968) problem solver for simple algebra, input sentences had to be short and of stereotyped form, such as \"what is the sum of .... ? \" Conversely, in linguistics, there was, u n t i l very recently, l i t t l e speculation on how we understand the reference of. pronouns in such eleqentary sentences as \"The s o l d i e r s fired at the woslen.and I saw several f a l l \" ,", "num": null }, "FIGREF5": { "type_str": "figure", "uris": null, "text": "like very general analogies may have.noticgd that Wittgenstain (1953 para. 2ff) devated considerable space to the construction of an e1sentaz-y language of blocks, heaxus and slabs; one postulated on the as tipn that the words of language were basically, as is supposed in modal theory, the names of items. But, as he showed of the enterprise, say that an extension' of this -st i s kmir~g s in t h e aanee nf saction 4 blow ''.+mad p o u a t l m t , , a base analyser Lrolln Bruce ' a ClUWWS amtem (1971) rhlch is e rl+zo r~rw ---late first genesat~un---systPmp i n k m soaw wnvr & W b t q~d * u . Otbch in t h e l a s t c a t w r y t h a t should be-mtentAwm3 Blaavhw bmd fsmdgs'419'?2) e~p3sration of k b c s n w p t s of luustl ahd ,,cmi8dq kn r mt-ldl & tic-tac-to!&, and Joshils anfcnsion ob it (k , m o t recsz~thy a p y l Y d to a r&cm-world d l t m e~ SOCBt m m , is not discussed i n t h e detail i t deserves i n this papex. '%be b w d on an augwnted s t a t e t r a n s i t i o n n e b o r k gs-, is udmabtmdly of 't;ba", "num": null }, "FIGREF6": { "type_str": "figure", "uris": null, "text": "rxaeast m b u s t in actual u s e , i n t h a t i t i s l e s s s,ensiUve to 9.4hRTICmS&t i n p u t guestions i t e n c o u n t e r s thdn its r i v a l s . 'Thq reasam fir tmdtbg it in depth is t h a t bath W a s and Wimq~ad 'nave an~u.ed 3n print t h t thsLs twlo s y s t a s are e s s e n t i a l l y eqiivaLent [Wisrsgrad 1971) 19731, a d so, if they are right, there i s no need Its dise:ss b t h , and Wixmgxwil's is, w i t h i n t h e WI camiaunity at l e a s t , t h e b e t t e r known of the t w o , Their ec&ivalenc,e arguments are proLdly mrxect: b t h ur g rbased deductive systems, operating w i t h i n a guestim-anmriqq b~~i -f i n a h i g h l y l i m i t e d d a a i n uf disccurse. Wixqsad'3 syst-QX h$aw a a hou to psoceed, w i t h i n his P-PM gramas, is, as he h h s e h f + p i n t s mt, fomallj. equivalent t~ an augmented s t a t e transition network, M in p r t i t v l a r t c~ the ordering of cbictes a t n d e s i n PIJodsl system, There i s a s i g n i f i c a n t diffexencfe 2n theis mtaphysicaf -, presuppositions about meaning which, howevex, has m inf flu err ace ca the aamL operation of t h e i r respective s y s t e a~. This d i f f e m e i s disguised by t h e allegiance both givd ta a 'prolcedwal view of m e a a i q '", "num": null }, "FIGREF7": { "type_str": "figure", "uris": null, "text": "d o f f ' , the l a s t ' i t t can only be referred correctly to the box, rqther than the t & l e , on the basis of some knowledge quite oth#r than that i n a conventional, and implausible, linguistic s o l u t i o n s m h as the creation of a class of 'level nouns' sb that a box would not be considered as being or n o t being level. These points would be generally conceded by those who believe there isam AX paradigm of language understanding, but there wdlrld be Ear less agreement over the p s f k i v e content of the paradigm, The txouble begins with the d e f i n i t i o n of 'understanding' as applied t o a computer. A t one extreme are those who say the word can o n l y rqfer t o the performance of a machine: to its ability to, say, sustaih dome farm of dialogue long enough arad s e n s b l y enough for a h w n interrogator to be unsure whether what he is conversing w i t h i s a machine or not. On t h e other hand, there are =my, a h s t c e r t a i n l y a majority, who argue that more is required, in t h a t the msthde and representations of knowledge by which the pexformance i s achieved must be of the r i g h t formal sort, and that mere performance based on ad hoc methods does n o t demonstrate understanding. This issue i s c l o s e l y related to that o f the r o l e of dqduction i n natural language understanding, simply because deduction is often the structure mant when 'right methods' are mentioned. The dispute between those who argue for, or, like Winograd, use deductive methods, and those who dvocate othex inferential systms closer to cammon sense reasoning, is i n m y ways a pseudo-issue because it is so d i f f i c u l t t~ define cltarly w h a t a mn-deductive system is, (if by t h a t i s meant a system that cannot i n principle be lnodellea oy a deductive system) since almost any set of forrpal'procedures, including 'invalid inferences1, can be s o displayed.", "num": null }, "FIGREF8": { "type_str": "figure", "uris": null, "text": "und'ewstand what was meant when Winograd contrasted h i s own w i t h what he called second generation systems, we have to remember, as always in this s u j e c t , that the generations are of fashion, n o t chronology or inheritance u f i3eas. He dedcribed the work of Simmons, Schank and myself among others in h i s s-ey of new approaches, even though t h e foundations and terminology of those approaches were set out i n p r i n t i n 1966, 1968 and 1967 respectiyely, What those approaches, and o t h e r s have in mmon is the b e l i e . t h a k understanding systems must be able to manipulate very complex linguistic ohjects, or semantic s t r u c t u r e s , and t h a t no simplistic approaches to understanding language w i t h computers will wrk.In a very influential recent paper, Minsky (1974) has drawn together strands in the work of Charniak (1972) and the authors above using a teminolagy of 'frames': tvA frame is a data-structure for representing a stereotype situation, like a certain kind of l i v i n g room, or going to a children's birthday party, Attached tea each frasne are several kinds of information. Sbme of thia is information about how to use the frame, Somc is about what cne can e x p c t to happen next, Some is about what to do if those axpectatione are not confimed. W e can think of a frame as a network of nodes and relations. The top levels of a frame are fixed and represent things t h a t are always true about the supposed situation. The lower levels have many terminals ---'slots1 t h a t must be filled by s p e c i f i c instarices or data. Each terminal can specify conditions its assignments must meet .... Simple conditions a r e s p e c i f i a by markers that might require a terminal assignment to be a person, an object of sufficient value, e t c . . , , II The key point about, such stxuotures is that they attempt to specify in advance what is going to be said, and how the world encountered is goihg to be structured. The structures, and the inference r u l e s that apply to them, ate also expressions of 'partial information' (in MKarthyts phrase) t h a t are not present in f i r s t generation systems. As I showed aarliar, with the 'women and soldiers' example, such loose inductive information, seeking confirmation Erom the surrounding context, is required for very s h p l e sentences. In psychological and v i s u a l terms, frame approaches ~n v i s a g e cnn undezstander as at ieast as much a looker a? a seer. Thus, w e might, very tentatively, begin by i d e n t i f y i q what Winograd called 'second generation' approaches with those making use 05 very general notions akin to what Minsky called 'frames'. But this is no more than a temporary device, for convenient initial classification of the field, because later we shall have reason to question t h e first-second generation distinction, and, as noted earlier, Minsky's notion of 'frame' is itself a highly f h i d one in the process of definition and refinemgnt. L e t us now turn briefly to five approaches that might be called sewnd generation. Charniak The new work which owes most to Mirrsky I s advocac,y is Ckarnink s . Hea studied what: sorts uf i n f o r e z i t i i z l infomation Charnlak 7 ' 7 3 , ''741 would be needed to rrasulm pronoun W i y u i ties i n ciai3dren l a s torjCr::, and AJ% that sehso tu understirrail t.htxu. O r l a of h i s ex,mgic, ' otouicls ' is: 'Jane was invited tu Jack's b i r t l~d a y p a r t y . She w n d e r d if he w u l d like a kite. A friend tcld June t h a t Jack already had a k i t e , an3 that ha would m k u her take it back it refers to the f i r s t k i t s uasrti~ned c?r tha s e~w n d . Charniakts analysis begins by p i r a t i n g out that a great deal cf what is required to understand t h a t story is implicit: Iun~ilrledge aLwut the g i v i n g of p~esents, knoililedge that if one possesses one of a certain s o r t of thing then one may well not want another, and *so on, Charniak's system does n o t actually run as a preyram, but is a theoseti c a l s t r u c t u r e of rules called 'demcnsl k h a t correspond roughly to what Minsky later called frames. A demon for this exanple would b e , If we s o p that a person might not like a present S , then lock Pclr S being r e t u r n e d t o the store where it was bought.Zf we st?e t h a t hngfening, or even b e i n g suggested, zssert t h a t thp rcasor why is t h t P docs not like S 1 .", "num": null }, "FIGREF9": { "type_str": "figure", "uris": null, "text": "we have a d e f i n i t e and c o r r e c t answer. \"I'his approach, of using partial { n o t n e c e s s a r i l y true) inferences, i n o r d e r to a s s e x t a definite answer, is highly c h a r a c t e r i s t i c of 'second generation' systems. The demons are, as with Winoyraa's work, expressed i n a procedural language which, on running, will seek for a succession of inter-related 'goals'. Here, for example, is a demon concerned w i t h another s t o r y , about a child's piggy bank (PB) and a child shaking it, looking fcr money b u t hear-324 K N E Y ) qDEDWE) (GOIIL (?NOID S W ?PERSON 3PB) $TRUE) (ASSERT (? HAVE: ?PERSON ?M) (ASSERT ('3 RESULT ?N WOLD) ) ) Again, it i s not necessary to explain the notation i n detail. to see that conditions are being stated for the contents of a piggy bank having k e n emptied. The pattern being sought by the demon i n operation is tire third l i n e .", "num": null }, "FIGREF10": { "type_str": "figure", "uris": null, "text": "the knowledge is of this general sort: in a recent paper (Charniak '74) whws tha ' k i t e ' s t a r y is reconsidered there are rules ~f considerable ganrrrsuty snd interest. One such is that Charniak calis a R+SSA rule: 'XE t h e .tory give8 information which would make it plausible to infer mt PERSON is favourably inclined towards action A, and PERSON does S I a signif i c m t subaction (SSA) of A, then f nfer that PERSON is doing A An important azsumption of Charniakls is what I s h a l l c a l l his 'decoupling assunaption', namely that this work on the knowledge-based detemahation of reference can be effectively decoupled from any partichlar syntactic or semantic representation, or of its apgdication, arid can be explored Fn isolation.", "num": null }, "FIGREF11": { "type_str": "figure", "uris": null, "text": "inyenyity has gone into the heuristics of this system, as its popularity t e s t i f i e s .The system has also changed considerably: it is n o w calltad PARRY2 and contains the above pattern-aratchinq, rather u a n earlier key Wrkt heuriatica. It has the partialt or what scme would call 'pragm t f~~, rule& about wpctation and intention, 9ndvthese alone might qualify it as 'swmn8 genarqtion' on some interpretations of the phrase.A genexator i+ alro being instaXLd to avoid the production of only *lcannedl remponsas .Colby and his associates have put considerable energy i n t o actually w i n g to find o u t whether or not psychiatrists can distinguish PARRY'S re8ponaes fran those of a patient (Colby and Hilf '73).This is probablythe f i r s t attempt actually to apply Turing's t e s t og machine-person distingu l s h b i l i t y . There are statistical difficulties about interpreting the results but, by and laxye, the result is that the sample questioned cannot distinguish the two, Whether or not this w i l l influence those who still, on principle, bel~eve that PARRY is not a simulation because it 'does not understand', r a i n s to be seen. It might be argued that they are in danger of falling into a form of Fapext's 'human-superhuman fallacy1 of attacking machine s h u l a t i o n s M a u s e *Aoy d not perfom s~p m k -n talkse like trcnslate m t x y , taakr that arcma p~p l a c @ r b h n i y cand.cr but #1i7 anajority cannot. Whdn such aceptlcr say Ult PARRY dws mt u n d~t s t a d they hava i n m i n d e lave1 of uderstsndLng that is mrtalnky highcna could e x t d their case iyonLcaJCly by pointing out t h a t m y EQN px?pl W a r s t a d the content af oantences i n the daptn am detafl that an and,~cic philosophar does, and a vszy g d thing toe. But t P I~c e can be +&t #at Paany p~g l a on many omasian% !XI oaem to W r r x s t a n d in ti:@ way that P M Y does. S m n s The remaining three s y s t m s differ fs\\# the t:w in Uaa&r attempt to provide soma repressntationrtl structure quite different frc~a L!it of t h e English input, This mans t h e use of cases, and of cxazpl~x structures that allow inferences to be &am from the attsibution of case in ways Z shall axplain, There as also, in the remaining syst=sr same attwzpt ta construct a primitive, or reduced, .-c&ulaxy irito which the lampage represented is squeezed. Simmons1 w x k is o f t e n thought of as a ' r~e a o r y Wed', though he dees in fact w y tsbore attention to wrd sense &iguitla, and ts actual recogn i t i o n in t e x t than do many other authors. For ! : i n the E u n h c n t a l ncscian is that of a 'soarantic networkq, defined aussntialfy bp the sratxmnt of relational triples qf f~m r aRb, where R is the name ut a relatlou and s and b are the names of nodes in the network. Shmorts' w r k w i t h this general formalism goes back to at least ( S h a m s et al, ' 6 6 ) but, in its never fern w i t h case foxmalisn, it has been reported since 1973 (Shamans '?Obi, ( S h a o n s and Bruce 1 , (Simmns and S&ocum '72) , (Sicmons ' 7 3 ) , ar,d (Hexdxix et a1 '73) may reasonably be considered a further implementation of Simmns' methods. Simmons considexs the example sentence 'John broke +the window w i t h a h a m e r ' . This is analysed into a network of nodes C1, C2, C3, Cd corresponding to the appropriate senses of 'John1, 'Bread', 'Windcv' and 'Hammer' respectively. The linkages between the nodes are labelled by one of the following 'deep case relations ' : CAUSAL-ACWUtT (a. , CA2) , ' EENS, =US, S03JRC6 and GOAL. Case relations-are specifications oT thd way dependent parts of a sentencer or concepts corrtesponding to parts o f a sentence, depend on the main action. SO, i h this example, John is the first causal actant (CAI) o f the breaking, the hammer is consideted the second causal actant (mi?) of that breaking, and the window is the theme of the breaking. Thuss the heaxt o f the analysis could ba repzesented by a diagram as followd: ' John C2 OF by a s e t of relational triples: (Cl CAS C2) (Cl CAZ C4I C C~ l?HE&E. C3) Huwwer, thisis not the f u l l representation, and my addition of the word 1-1s to the diagram is misleading, since the nodes m e intended to be nar~ea of senses of wozds, related ta the actual occurrence of the corxesporufing wad in a t e x t by tho relation TOK (for token), In an inrplemntatfon, a node would have an arbitrary name, such as L97, which would then rumit a stared sense definition. %, for a sense of 'apple Shmanrr suggests an ao~otiated set of featwes: NBR-singular ( S ) , SHAPE-spherical, COLOR-red, PRINTTHAGS-apple, THEME-eat, etc. If the name of the node tied to t h i s s~t of features was iweed L97, then t h a t W e might becane, say, 6 5 on being brought i n t o s a a sentence representation during parsing. Thus the diagram 1 gave must be thought of supplemented by othez relational ties fram the nodes; so that the.ful1 sentence about John would be represented by the larger set of triples: (Cl TOK break1 (C1 C A l C2) (Cl TEIEf3E C3) (Cl -2 C4) (C2 TOK J o h ) (C2 DET Pef) (C2 NBR S) (C3 'ToK Window) (C3 DET Def) (C3 NBR S; (C4 TOK Bamer) (CC DET Indef) (C4. NBR S) (C4 PREP With) Wxd eense ambiguity i~i taken account of in that the node for one sense of 'hamaex' would be different fram that corresponding to some other sense of the same wzd, such as that meqaing Mwam3, s l i g h t l y strained alternatiw for this ssntenca, The network above i s slro o represantrrtion aC t)U Collouing rsnt,nger which can bb .Ihouoht of as surface vaqlmta of p s h q l e ' u I * l l~t l y L q~~ S~N G t w e : John broke the w i n d~w with a hakmmr John broke the whnaow The hammer b r~k e the widow Thw window broke, k t all parts of t;harknetwrk w i l l be s a t up each QP these sentences, ~i caurse, but tha need f e x same i t e m tO f i l l an appmpriate s! t can 3Ye infer-~'dj i , e . of the f i r s t c~u s~~c W~ [John) in the last tw sentences, The sentences a h~v e are recognised by mans of t h e 'ergative paradigm2 of orderec matching patterns, of which t h e Eollowing list is a part: each match, a s left-rigw ordered itarns, one sf t h e dwve sentenceq. It w i l l b2. s clear t h a t Simmons' method of ascribing a node to each word-sense i s mt in any way n p r h l t i v e s y~t~~ by which t m e a n a system of classifiess i n t o which a l l word sanses,are mapped. S h m n s is, however, cansidexing a system of paraphrase rules that wlould map from one network to another j31 a way that he claims is equiwlent to a system of primitives. Thus in (Simmons '733 he considers the sentence: John bought t h e boat from Mary Mazy sold the boat to. J o b which would noracally be considered approximate paraphrases of each other. He then gives 'natural' representations, in his system, as fallews in the same order as the sentences: * S h o n s ' normal example o f word sense ambiguity does not apply to the sentence above: he distinguishes ' p i t c h e r l ' , a pouring containerr froea 'pltcher2 ' , in the U .S . sense of 'one who h o w l s a! ball ' . Ci TOK buy, SOURCE ( W s q ) , GQRt \\John), (ha$], cl TOK e;okl, S c T m C e tllary) ? COAL (John), T M ( h a t ) , and also the bLngla ,gapreselltation for both sentences, as below, using a prlnitfve action ' tranatsr' 9sae description of Schank's work in next sec-, SWRCX (Nary) , GOAL (John) , THE B (boat) S L~w n a opts for the f i r s t .\u20acom of representation, given Ule poalibi l i t y of a transfer rule g o h g Ptopl either clT the shalkower representations tb I h c other, while i n", "num": null }, "FIGREF12": { "type_str": "figure", "uris": null, "text": "SFrPPJons and Bruce '7 1) to the auloPPatic ttanslation of the networks i n t o a cmxrespontling Efrst-order predicate calnilus format of the sort developed by Sandenall. (1971). This last i s particularLy valuable because,LZ penatalirablr, lt rhwr that any linguistic d i , n g i n network Ecm -can b translated in-some form of the predicate calcults, i f that Formalism and i t # crssvclated proof kachnigurs oan be shorn t o be app?!!priate for carb i n p #~~b l a a $ La the area of natural language analysis.SehankSchanScqa is a rich system of semantic representation, developed aver a psiad of six yeaxsf w i t h the collaboration af a number of talented students.Its graph system of notation has influenced psychologists like&belson (1973) , amony others. Schankls c o n t r i b u t i o n has been the n o t a t i o n a l s y s t e '~ representing the structure of natural lancruage sentences, and this has &ern progr-ed by various collaborators over #e years. In i t s present version, caJled W G f E (schank et a1 ' 7 3 ) it has an ana'lyser of I\";nal&h t %~to Zdesbeck (19741 s seiixmtic memory E w p n e n t due to RiegW (1974) , and t genetatox of English due t;o GoLban (1974). MaRGIE produces cxutput in two modes, demonstrating the sort of wnceptual FnShrencfng t h a t goes on at -the Bevel s'f the stmmUc repkesenbUwB: Samples of 'input wd outwt tm rad ho t& ba rPdw e m h shm thus: q b w rtirisc s k u wmts a kiodt. 8511 wmts to OWM CQ t~w a m. BAAL Moats saawsle to ~~~Q W haw 198, book, B ' i h l wmts %aEZ ram3 a book, OUWIPP: JQhn s~a . n q l e d ,~a r y . Joksn c h o W Wxy and she died because &e c=add mt kmae%aaa, h s y d i d because she was unable t~ i-le air azla she was u&Ls to inhale s m e air because John g r a h k d her D I F & The a h of Schadc's aystem has ahways h e n t xa pmvida a --Urn of meaning in terns of which these aLsdl ether tasks, such sa &i,m indeprsdent o f any pasticullax language, and of syntax, /ityq fS'sbrca, Oe a l l surface structure whatever. The f c m l s t r u c t u r e of Schank's g r a n s is that of (Bays ' 6 4 ) , a d t h e items i n gsaw axg p f f~u r typesq ar cmtegories. They are symblisd as PP, A m , PA asld M, vhfdr a m mzzuqms, h t which correspond closely (for t h e purpose of um3e.r~-th.ix f e a m = t k d tb those of a noun, verb, adjective and adverb, r e s p c t i d y . e + T'Ik basic --Schank distinguishes 'ccnceptuall and 'searnantic' z e p e s e m a t i a &~ im r w that is important fox h h within his own syttep. Bowever, Z obrll tmm t h e terms i n d i f f e x e n t l y since, in this brief a d stprfichl W-, nothing hangs upon t h e distinction. **This is a considerable o v e r s~p l i f i c a~~n~ in cx&w to gim r 4 self com.taine3 ,descrSigrion. ;Butl in fact, aany Ex@ km represented as A m ' s : chair, pen+ honesty, am3 ' t r w I t f c m , structtare is called a conceptualiaation, and is normally intrbduced w i t h a etraightfomard dependency structure such as, for the sentence 'The man teok a book':", "num": null }, "FIGREF13": { "type_str": "figure", "uris": null, "text": "Phere are at present fourteen* basic actions forming the nubs of the graphs, as well as a default action DO. They are: PROPEL, .MOVE, INGEST, EXPEL, GRMP, PTRANS, MTRANS. ATRANS, SMELL, SPEAK, LOOK-AT, LISTEN-TOI C O W and MBUZLD. The notions of case and primitive act are related by rules i n the develOpment of conceptualizations. So, for example, the primitive act INGEST has as its instrument the act PTRANS. mere are also other infer-Since the publicabion of (Schank 73a) their number has been reduced to ezeven (plus DO) by the elimination of SHELL, LISTENID, IXK)KAT and COQC, and the adation of ATTEND, ances fram any ACT classified as an INGEST action, such as that the thing ingested changes its fomt that M t A e U A~G Awp&ac3 is&LUe k b In gestet becomes lmrs nourishodl s t e m (oao S c b n k '73, pp. 38tf . I . T h i s will all k c a m clsaxar i f w e consider tho trmsitl~n Zrcm a dietiwary entry Pox asr action to a f i l l e d -i n mnceptuahirati~n. Hem Is tthe d~ccionary entry for the action ' s h w t ' :", "num": null }, "FIGREF14": { "type_str": "figure", "uris": null, "text": "This case inference muSF b~ made, according to Schank, i n order to achieve an a d m a t e zepresentation. There i s , in the last diag~am, a cextain redundancy of expression, but as w e shall see tn the next section this often happens w i t h deeper semantic notations.More recently, Schank, together with Rieger, has developed a new class of causal inferences which deepen t h e diagrams s t i l l further.So, in the analysis of 'John's cold improvPQ1 because I gave him an v p l e l (Lrt Scfrank '74a) the extended diagram contains at I h~t four yet lower levels of causal =rowing, including one corresponding L.ht the notion of J u h can-s+Ncting the idea (WBUTLD) that he wants to ea% an apple.So we can see that the undexlying explication of mean* here i s not only i n the serlso of Iinpulistlc prLmltLws, but i n tern of a theory of mental acts as w e l l .N o w Ulsra a t e a number of genuine ~p o s i t i 6 n a ] . difficulties here for the euml&tator faced w i t h a epstm of this complexity, One aspect of t h i~ is the atages of d e v e l o p n t pf the ~y s t a m i t s e l f , which can bc seen ae a consimtcntpmcesa of produrlng what was argued for i n advance. For ekampla, Schank claimed early on to be a constructing system of semantic mtructures undatlyirrp the 'surface of natural language', alehaugh initially them were no primitives at all, and qa late as (Schank et a1 '70) there was only a single primitive TRANS, and most of the entries in the dictionary conmisted of the Bnylish wards coded, together with subscripts. Since than the primitive system has b&ossomed and &ere are now twelve primitives tor hCTS including three Ebr the original TRANS i t s e l f . Each axposition of the system recounts its preceding phrases, from the original primitive-free one, throuqh to the present causal inference form; rather as each human foetus is said to relive in the womb a l l the evolutionqy stages of t h e human race. The only trouble w i t h this, f x m an outsider's p o i n t of view, is khat at each stage the representation has been claimed, to be the correct one, while at the same ti-S c h d admits, in moments of candor (Schank '731, that there is no a d to the conceptual diagrrVbraing oE a sentence. This d i f f iculty m y well reflect genuine problems in language itself, and, in its acuteat form concerns a three-way confusion between an attractive n o t a t i o n for displaying the 'meanings of wordsv, the course of events i n the real world, a d , f i n a l l y , iibtual procedures for analysis to be based on the diagrams, This raises t h e , to me, inrportant question of t h e application of a semantkc system, that I shall touch on again later. Schank, for example, dues mention i n passing the questions of w d -s e n s e ambiguity, and t h e awful ambiguity of English prepositions, but there are in no way central for him, and he assumes that w i t h the availability of 'the correct repxesentationWt h i s s y s h when UnpleIm?nted must i n e v i t a b l y soive thtraditional and vexing questions. M procedures are hinted at along w i t h the graphs as to how tnrs is to be done. A d i s t i n c t i o n cof importance may be becoming ap-parent here batween Schank s work and Riegerls: in Riagar's thesis ( R L q e r '74) the rules of inference appear to craatg clclparata and new rubprrphr wnicn may s w d in an lnfersntial celatlon to each other so as tsl produce cloncluaiona &but: ~~~~~a of, gayt pronoun rabr~ence, etc. But i n Schmk's cormsponUng papers the sihfctrsncss u r n not applid ke actual problems (Schank '74a) but only beto amplaxity th. conceptual graphs yet further. Closely connected w i t l a this raattex i s the quaation af the survival of the mlvface sr'tructure in tho diagrams. U n t i l very recently p~~h , i t i u i sation applied only to verbs, t h t of nouns being Left to Mehr [Wekwr 272) Most recently, though, noun w d s have been disappearf ng from dfagraps and been replaced by.categorfes such as +PZIYS0=* But A t is cleax that the swface is only slowly disappearing, rathex than having been abhorred a l l along. In a mra recent publication CSchank '74b) there are signs that this. trend of infinitely p r o l i f e r a t i n g diagrams (for i n d i v i h a l sentences) is feversing. I n i t Schank considsrs the application of h i s approach to the repraqentation of t e x t , and concl :des, c o r r e c t l y i n my t P 4 e t q , that the representations of p a r t s of the t e x t must be interconnected. by causal arrows, and that, i n l o r d m to present@ 1rv.cidity, the conceptual diagrams for individoal sentences and their partvi must be abbreviatedr as by triples such as POEPLE P p l W S PEOPLE. her^ indeed, the surface simply has to s u r v i w i n the representation un1esF one is prepared t o c a m i t oneself to the axeme view that the ordering OF sentences i n a t e x t is a purely s u p e r f i c i a l and arbitrary matter. The F e n s e i n wnich this is a welcome reversal of a trend should be clear, because i n t h e 'causation inference' development, mentioned earlier, all the consequences and e f f e c t s oE a conceptualization had t o be drawn within i t s e l f . Thus, i n the extreme case, each sentence of a t e x t 7 should have been represented by a diagram containing most o r a l l of We t e x t of which it was a part. Thus-the representation of a t e x t would haye been impossible on such prihciples. Pay own system constructs a semantic representation f o r small natural language texts: the basic representation i s applied d i r e c t l y t o the text and can then be 'massaged' by various forms of inference t o became as deep as is necessary for well defined tasks demonstrating understanding. It is a uniform representation, in that information that might convenionall~ be considerea as syntactic, s-ntic, factual ox rnrerencial LS w e u axpressed within a single type o f struwra. The fundamental u n i t i n the mnatructioe of th2s beaning representation is the template, which is intmded to correspond to an intuitive notion of a basic message of agent-action-object fom. Templates are rigid format networks of more basic butldlnp blocks called fomulas, which correspond t o senscln of individual worde. In order to cohskruet a cctnplete t e x t representation templates a t e bound b y e t h a t by t w o kinds of higher level structures called p a p l a t e s and inference rules. The templates themselves are built up as the construction o f the representation proceeds, but the. formulas, paraplates and jlnference rules are all present in the system,at the outset and each of these three types of pre-stored s t r u c t u r e is ultirpately constructed frm an inventory of eighty semantic primitive elements, and from functions and predicates ranging over those elements. The system runs on-line as a package of LXSP, k&ISP and MLISP2 program, W i n g as input small paragraphs of English, that can -be made up by the uber from a vocabulary of about 6QO word senses, and p r d u z f n g a good French t:ransL&tion as output. This environment provides a pretty clear t e a t of lmguage undarstding, bcauaa E'rench translations for everyday. pmse are either right or wmng, and can be seen' t o be 910, while a t the same titPe, the mfot difficulttee of understanding ptogtamsword sense ambigratty, case anbigufty, difficult pronoun reference, etc.can a l l be represented within a machine translation environment by, for example, choosing the words of Lhe input sentence containing a pronoun reference difficulty so t h & E the potsible alternative references have different genders in French. In that way the French output mdkes quite clear whether or not t h e program has made the correct inferences in order to understand what it is trans-la_ting. The program i s reasonably robust in agtual prformance, and w i l l even tolerate a certain amount ob bad grammar in the input, since it does not pzfbm a syrkax a~L y s I ; s &=%he sense, hut snnkn message forms representable &XI the semantic s m c t u r e s employed. v p i c a l input would ble a sentence such as 'John l i v e s out: QE tQtm h i &inks his wine out of a bottle, Mo than throws the h t t l a s out uf Uae for each of tke thraa occurrences of 'out o f t , since i t raslisarrs that they diffexenco must be reflected in the Fwonch, A sentonce such as, * G i v e Use monkeys b;mands although they are not ripe because they a y~ very I~wQry' produces a translation w i t h d i f f e r e n t equivalontd fur t)ra tm eccurrmwr of lUaey1, bocause the s y s l m oorr@ctlp realiasst iraQlSI wlrat 4 s h a l l describe below a t preference considerations, that tlae most sensible i n t c t~r e t a t i c n is one in which the f i r s t they' refers to t h e bananas and the second tc the monkeys, +and bananas and monkeys~have different genders in French. These two e x m p h are dealt w i t h in: t h e 'basic d e ' of the system. (Wilks 73a) Inmany cases i t c m o t x e s o l v e pronoun ambiguities by the sort of straightforward 'preference considerations1 used in t h e last exaaple, where, roughly speaking, 'ripeness' prefers to &-predicated of plant-like things, and hunger of animate things, Even in a sentence as s b p l e as 'John drank the wine on the table and it was g d t t such considerations aye inadequate to resolve the anbigtlity of 'it' between wine and table, since both m y be good things. In such cases, 02 i n a b i l i t y to r a s~l v e w i t h i n its basic moder the program deepens the xepresentatio~ of the t e x t so as to t q and s e t up chains of inference that will reach, and su prefer, only one of the possible referents. I w i l l return to these pzocesses in a n m e n t , but first I s h a l l give sane brief description of the basic representation s e t up for English. For each sense of a word in its dictionqxy the program sees a f o h u l a . This is a tree structure of semantic primitives, and is to be interpreted formally using dependency relations. The main element i n any fonnula is the rightmost, called its head, and that i s the fundamental category to which the fonnula belongs, In the formulas for actions, for example, the head will always be one of the primitives PICK, CAUSE, CHANGE, FEEL, HAVE, PIXME, PAIRl SENSE, USE, ,WANT, TELL, BE, 5X), FQRCE, W , -THINk, FLOW, W, DROP, STRZK, FUNC or HAPN. Here 18 t n t s eLee stxuctufe for the. action of drinking: ( m u PART) Qace again, it is n~t necessary to explain the formalism in any d e t a i l , to see that this sense of Idrink* is being expressed as a causing to mve a liqyid object (F'WH m) by an animate agent, into that saine agent (containment case indicated by IN, and formula syntax identifies SELF w i t h thB +gent) and via (direction case) an aperture (TLIRU PART) of the agent. Template structures, which actually represent sentences and their parts are built up as netwcrks of formulas like the one above. Templates always c o n s i a t o f an agent nude, and action node and an object node, and other: nodes ttat laay depend on these.", "num": null }, "FIGREF15": { "type_str": "figure", "uris": null, "text": "tonsidarable tamwnt a f squeezing af this sbapl~ eansnkcal Lorn of template is necessary to W e it f i t tha mmplexfty of language: t e x t s have to bt Eraymented initfalLy? then. in fragments which a m . say, grapositional phrases there is a daaay agent Lapasad, and the prepsitions1 phrases thexe is a dummy agent imposed, and the gremsiticmai LomuLa functions as a pseuda-action.There are special 'less preferred1 oaliers to deal w i t h fragments not in agent-acti~n-object order, and so on.men the local inferences have been done that s e t up the agest-action object templates for fragments of input text, Shd system attempts tm t i e these templates together so as to provide an overall initial s t r u c t u r e fox the input. One form of this is the anaph,oxa t i e , o-f the sort discussed f o g the monkeys and bananas example above, but the =re general \u00a3 o m is t h e case tie. Assignment of these would result in t h e template far the last clause of 'He ran the mile in a paper bag' being tid to the action &e of t h e template for the first clause ('He ran t h e mile'), and the tie k i n g l~~~e CONTaiment. These case ties are made with the aid of-another class of ordered s t x u c t u e s , essentially equivalen* to FMlrPore s case f ruses, called p r a p l a t e s and which are a t t a h e a t~ the formulas for English prepositions8. SO, for 'outof', for =aaple, there h-ould be at Least six ordered paraplates, each of which Is a string of functions that seek inside templates for information. In general, paraplates range across t w o , nat necessarily contiguous, templates. So, in analysing 'He put t h e n u h r he thought of i n the t a b l e ' , the successfully matching paxaplatz would p i n down t h e dependence OP the template for the last of the three clauses a s DIREctior., by W i n g a s ampment only that particular template for the l a s t clause that contained the formula for 'a numerical t a b l e ' , (andnot a template repxesenting a kitchen", "num": null }, "FIGREF16": { "type_str": "figure", "uris": null, "text": "elaborate o q m a t t c rraprssantation than is necessary. fox the t a s k i n hand and, FE the i n i t i a l block can be constructed and a generation of F-;rich &one, rto 'deepening* of the representation w i l l , be attempted, H O W~V~X , wmy exmples cannot be resolved by the methods of this 'baeic mode' and, in particular, if a ward sense arPbiguity, or pronoun reference, i~ still unresolved, then a unique semantic block o f templates canrnot be constructed and the 'extended mode' will be entered.\" In this &a, new template-like forms are extracted fran existing ones, and then added to M e template pool ftom which further inferences can be made. So, i n Ula t m~l a t a derived earlier for 'John drinks w i n e ' , the system enters the Loarula for 'drinks', and draws inferences corresponding to coach case sub-Eorrmula.In this t~xtmple it will derive template-like forms equivalent to, in o m J f~r y English, 'The wine is in J o b t , 'The wine entered John via an aperture' and so on.", "num": null }, "FIGREF17": { "type_str": "figure", "uris": null, "text": "inference rules a r e then brought down, which attempt, by a s-fe strertegy, ta construct the shortest possible chain of rule-linked tmmplate fo-8 from one containing an ambiguous pronoun, say, 50 one c2nt a i n h g one of its ~s s i b l e referents. Such a chain then constitutes a solution ta the ambiguity problem, and the preference approach assumes that the shartest chain is always [the right one.So, Yn the case of 'Jahn drank tha wine /on the table/ and it was good', (in three temprate-matching fragmenb as shewn) the camact chain t~'wine' uses the t w o rules * Wibo '73b, and @n preera) ~n L y p t h a & , that k r cu sky, tlwy corrraFnd 4 w l y to w h t w a may xeao~rmbPy I w k out tor i n a gkvrn r t t u t k o n , net? to u b t WSP happn. Tha hypotl,.sir irere btrt wderrtahl&np can only t r k r plracs oh the baefs of .akpptble rufoo that: are mafixred by the eunlext af ap~lication. I n this axample the chain constructed u y ba expressed as ( w 8 s r i n q the &ova sguam bracket rmtaatio~~ to cont&bn nut a representation, but sisxxply an indiedtion, in BngldsA, of the template contents): The a s s m p t i o n here: is mat t w ehain u s h q ather inference niles wttl,d have reached t h e ' t&1q1 s o l u t i o n by using less ~.aa t w o s u l e s , The chief drawback sf this sp!irm is that d i n g s consisting e n t i r e l y of primitives have a considerable amount of bo'eh vagueness and redundancy For example, ns reasonable coding in terms of structured primitives could be expected to distinguish, say, ' h m e r l and 'mallet'. That m y n e t matter provided the cdings can distinquish i a p o s t a n t l y d i f f e r e~' : of words. Again, a template for t h e sentente s he sheperd tended his f l o c k ' would contain considerable r e p e t i t i o n , each node sf the template t r y i n g , as it were, to t e l l , the whola story by i t s e l f , again, t h e ~r e f a s e n c e c z i t e r i a are not in any weighted, which m i g h t seegn a dxahcack, and t h e prefexential c h a d LET@I c r i t e r i a for h f e r e n c e chains m i q h t v~f f seem too crude.", "num": null }, "FIGREF18": { "type_str": "figure", "uris": null, "text": "In notation is reduced if one see$ the topological similarity mat rorults from mnrlderfng the head of a formula as functioning rather lLko a Schwk bait action.If one thinks of khe dependencies of the case e u b p r t e of a fornula, rot &ranged 1 lneargy along the. .bottom of a tree, but radiating out Exthe head i n the centre, then the t w o diagtsms actually have identical topologies under interpretation. A difference vises in that the 'filled-in entity' for Schgnk is the conceptualization centred on the basic action, though for me it is the network oE formulas placed in relation La a t~l o p L a t e , whexe there is indeed a basic action, the he& of the action formula, but there is also a basic entity i n the agent formula and SO on.OX, to put FL another way, both what-is and what-is-expected are represented in the templates: the agent formula represents the agent, Pox exampla, but the left-hand pact of the action formula alsp represent3 what atgrant was expected or sought, as in the (*ANT S U M ) sub-formula of the *&inkt formula, A~t h o u~h developed i n isolation i n i t i a l l y , these twlo systems have also influenced each other in more recent years, probably unconsciously. For eatample, conceptual dependency now emphasises the agent-action-object far~rcrt -re than befoxe, and is less iverb-cent.red' and t~h e l e s s while, ronvez-sely, rrty own system n o w W e s much more overt uuc= of ~l e s of w t -fa1 LnfcanaaUun than in its earlier versions. Again, b~t h systems have intellectual conneotions that go back before either generation of A 1 systeam. I n my view, both these systems have roots in the better parts of the Computational Linguistics movement of the F i f t i e s : in the case of Scfrank's s y s t m , cane may think of the earlier systems of (Hays '64) and (Lasrb '661, and the arkow-structured primitive system of (Farradene '66) ~~E B & e~~O L a 3 s -~~~p r c c e a e n t s . F n t h p .Parkex-&odes '61) system of classiffc&ion awl the early seamtic structures of (Richens '61) and (Casrsterman '61). In 1961 the last author was arguing that 'what is needed is a disoiplina that w i l l study suantlc nsolrge camaction in a, way malogous to that in which r&nrba#omatfes n o w rtudiro ~~t h m t L c a l connaetion, and to that in which ~U l m~t L historical p i n t x&fsrso s final bns that is, I feel, of prgsing i n t e r e s t . T h e n seem t~ bs two rsscarch styles i n this field: one is what sight b9 callled the ' f u l l y f i n i s h e d style1, is whish ma w x A e x i s t s only in one ccmphte fow, and is not issued i n iaerly ar dove1 v e t s iodns, The best example of this i o Winoptad's w r k . The other type, examplified by a l l the other authots di8scussad hate, to same e x t e n t , is the det-eloying style: work which appears in a n-r of vlersiens over tke y e a r s , one b p s w i t h gxadual hprovt?ments, perkraps i n attmpts ta t a c k l e a wider range of lfncpistic or other inferential phenomena. There are & v a tages to both styles, but even in the latter one h w s t h a t any proposed stxuctura ox system will, in the end, be found wanting! in Lhe balances of language, so it can only be a question of when one w i l l have to abandon ~t . The interesting question, and one to which no answer could possibly h given here, is just how far is it worth pushing any given structural approach before starting again fram scratch? 6 , Sane Cormpisans~, CVkd,. -,l'qn-$r-ass In t h i s sactivn X shall -para and c o~~t r a s t , under some nine interconnected headings, t h e projects ae.;cribed in b e M y o f the papex, This is not easy to do, particularly when, the pxesent author is among the writers discussed, though that is easily m r d i e d by b e reader's W i n g an appropriate discount. A more serious problem is t h b , at this stage of research in artificial intelligence and r:ztural language, the most at-Ctractive #istinctions dissolve on more d s~a i l e d scrutiny, laxgely because of the lack of any p~ecise theoretical statement in =st, if not a11,'tha major prn jects. There are those w h o think that it therefore follows that t h i s i s not m e lzrwrent for any form of c r i t i c a l camprison in this field,and that no more is needed than a 'psitive attitudey towards all possible p x w j e e b .", "num": null }, "FIGREF19": { "type_str": "figure", "uris": null, "text": "r t Z y t h s e w h o feel tht,.-wt theemkrary, any k b e is a s g o d @Len arms at2 u~ @puo *Xqoa ayr7 aaoqapuey a m a-m m s X s s , auo axaw qua;EPdstrea aq pvcm q~a 3 aqq p anbv -yaaq bu~~ozd-asxoaqq pzapu~qs X x a~ awzs 73123 q axaw sanljmwtfazapq s,euo qew zea3 am aq pTn= 'aldmoxa ro3 &uossaz peq B :asam azo~dbm a e a q d am qcm sdeqrad araq put? uo~qqolu p r a m s e bqp~oha xoz stmmaz eren smq9Xs sqq yo ; I~T xaysaa qanm aq X~x w l 3 ' . p~n a~ &pd s~w 30 3saqq 316511 &XPB 03 SF 1-.pssnasTp s i a y r m y a m e m ftm 49 pasn smzbalp pua rpuopqou aulsla33fp so q a~e w s~ 3apzw-q s 2 s w x a u o~ gua suomrrodans wyqe0~130.)38,03 raqqoyy 'pax.dB03 aq 03 suarcud atp Xluo pup naantm t p~y r rAm! 30 qasqno am qle # n a p S'J q j 0s tsaq3euzdd~ ~t o~~? r i r u o 65~2~3 saw0 UPTTarJI sta \" y r m UMO s , P z ) Z~~~U~ ran-X~upqxa;r pTnm qupkl pumas oy7 *raho a6pa~n~t.q pix-Teaq plro ~snqdaaucm bu~quasoz8ez o r m q s~r awm-6~~3 U'flpqUUJ (t) PUP 'r)ndu~", "num": null }, "FIGREF20": { "type_str": "figure", "uris": null, "text": "m a r , $ to 31e Ln w e -t o w cxxrorpwdbam vlsh ropLM m. The s l s a q e s t how-kewl appmach fs -y h t ollrcm;t$J dm that this dispute is ultimateby one oE degree, simx no a a e clnilr that every lccution recognized by an $nltelligaat rryl-a r t br i.Qpd h t o a 'deep' representation. To taka an extmme auab amy -tam tikt uag@ 'Gxd W8rninqT into a deep sewm%hc represemgltkm that the carsect sespsns8 was also rGcad &arnhxjt wauld $Ch & seirfms thesretical mistake. Hawever, t h e m s t serious arqument f a ira m n -t q p a r % f s h r $ -tmUm Ss mt in kerns 09 the av~Hdaqam of ammbm~ d % g % % d U w , & c l o s e l y t i e d to t h e defence of se&anthc pahb2hwm %EI m, u&%& Ps a large subject not to be unde~taken here. Cbe a% the t m x U e s & tic p r h , i t i v q s is t h a t they are open kabad &e%eaces, representation have declared t h e m to h v e scme d -WU existence and have implied that there is a 'right sett ~W w e s b p q to ernpiriqal discovery. On t h a t view the essentially h q d . s U c of structures sf p r b i t i v e s is lost, because %t is an u s m w k L d fe&are af a language that we can ch-e i t s vocabuhry as function ui& dLt&matf.a vocabularies. B u t if there is a z i g h t set of prLmiUwm, u t m t~ m k~~% are t h e a w e s of brain-items, then t h a t essential SF = -uxald km h&t-What is the is t h a t there is a considerable amount of psychologicill evidence that Geople a2e able to recall. the content of uhat they hear and understand without being able to recall either the actual words or the syntactic structure used, Thare is large literature on this subject, from which two sample references would be [Wettler '73) and (Johnson-Laird '74) . Thesc results are, of course, no proof d f the existence of semantic primitives, but they are undoubtedly supportkng evidence of their plausibility, a o is, on a different plane, the remlt from the encoding of the whole Weboter's Third International Dictionary a t S y s t m v Develagmefit Corporation, where it was found that a rank-ordered frequency count of the words usad to define other words in that vast dictionary was a l i s t (omitting 'the1 and a which corresponded almost item-for-item to a plausible list of ssmantic primitives, derived q ptioxi, by those actualhy concerned to.codel the structure of w m d md sentence meanings. Zt is important t o d i s t i n e i s h t h e dispute Ibout level from the, closely connected, topic that I s h a l l call the centrality of khe #nowledge required by a language understanding system. Centrality What X ari a calling the centrality of certain kinds of information concerns n o t its level of representation but its non-specifidty: again a contrast can be dram between the sorts of infomiltion required by Charniakls s~s t~, 0 x 1 on@ hand, and that required bySchankls* and my o m on the obhar. Charniak's examples suggest that .the fundemental form of information is highly spacific** t o particular situations, Like parties and the giving of presents, while the sorts a\u00a3 information central to Schank's and my own systems are general partial hssertions a b u t human wants, expectations, and scr on, m y of which' are so general as to be almost vacuous which, one misht argue, is why their zple i n understanding has been ignored for so long. -Though as noted earlier, Schank in 1975 has adopted Rbelson9s (1973) notion of 'script', as a largar-scale 'frame1, in such a way as to incorporate much less 'central1 knowledge. **In a recent paper (1974), Charniak gives much more general-rules, such as his 'rule of significant sub-action', mentioned earliw.If I were a reasanably Eluent spaker of, say, G8man, 1 might we13 not understand a G e m conversation about birthday presents unless Z had tietailed 8factuaS. information &but: h w Gens~~~ns organixa the giving of! presents, which Plight be considerably different the way w e do i t, Converselys aP course, 3 migl.rt u m i a r s t d much og a twkmlcal artLcle abut a subject in which I was an expert, even t h~~r g h1 knew w r y Ifttka af the language in which it was written,Theseaz'e certainky wnrL3exatians that t a l l Lox Charniak's approach, and it La perhaps a paradox that the s~r t of r~aturrl language understalridler t h a t w u l d tend to C O J I P~~ his apswp,t tons ~muld be one concerned w i t h disooursa &L.r;)ut, say, the details aL reyapking a t$Otor CU, where factual. infsmathon is what is centsalt y e t r imnically, Charniak has concehtrated on something as general as childtents stories, w i t h their need of deep assumptions about hurwn desires and khaviour. In the end 'this difference may again turn out to be one of enphasisj and of what is most appropriate to diSSerent subject areas', though there niay be a vexy general issue l u r k i w somewhere here. It seems to me not a fuolish question to ask whether much of what appears to be about natural language in A.I. research is in fact about language a t all, Even if it is nbt that may in no way d e t r a c t fran its value. has argued that A . 1 . work is in fact 'theoretical psychology', in which case it ceul8 hardly be researchon natural language, When describing Winograd's work earlier in the paper, Z raised this question in a weak farm by asking whether his definition o f Ipickup1 had anything to da with the natural language use o f the word, or whether it was rather a description of how his system picked samething up, a quite different matter. Suppose we generalize this query samewhat, by asking the apparentky absurd question of what would be wrong w i t h c a l l i n g , say, Charniakls work an essay on the 'Socio-Economic Behaviour of American Children Under Stress? I n the case of Charniak's work this i s a facetious question, asked only in order to make a point, but with an increasing number of systems in A . I . being designed not essentially to do research on natural language, but i n order to have a natural language ' f r o n t end' to a s y s t m t h a t is essentially intended t o predict chemical spectra, or play snakes and ladders or whatever the question becomes a serious one. It seems to me a good time to ask", "num": null }, "FIGREF21": { "type_str": "figure", "uris": null, "text": "And, of cowrr, t h a t poaskbility has actually been entertained within certain strains of d a r n philosophy. itowaver, I am not w i n g hefa, to breathe fresh life into a philosophical distinction, batween being aLuut lmpunge andnot being about language, but tather introducing a practical distinction, (which is also a consideration in favour of optiqg, a3 I have, to work on very general and central areas of howledge) between specific knowledge, and central knowledge without which a syartem could not be said to unilexsttind the language at a l l . For 1 example, I might know nothing of the arrangement of American birthday parties, but could not be accused of not understanding English even though I failed understand s m e pazticular 'children's story. Yet, if I d i d not have available acme very general partial inference such as the ane people bainq hurt an8 fallingr or one about people e*avouring to possess things that they w a n t , then it quite possible that my lack of understanding of quits airtple aentencee would cause observers to think that I did not underr W Englbsh. An interesting and difficult question that then arises i s whether those who concentrate on central and less central areas of discouse could, i n principle, weld their bodies of inferences together in such a :gayas to create a wider system: whether, to put the matter another way, natural language is a whole that can be built up fxm parts?Pken-noloqica levelAnother distinction that can be confused w i t h the central-specific one is that of the lphencmenological levels1 of inferences in an understanding system.", "num": null }, "FIGREF22": { "type_str": "figure", "uris": null, "text": "mean nothing daunting by the phrase: consider the action eating which is, as smatter of matmica1 fact, quite often an act of bringing the bones of my ulna and radius (in my arm) close to that o f my lower mandible (my jaw).", "num": null }, "FIGREF23": { "type_str": "figure", "uris": null, "text": "Therq is no denying that t h i s sort of infomatioar might be useEul to have around somewhere; Wt, in Minsky's terms, the 'default1 value of the instrument for eating i s t h e hand brought to t h e mouth, so that, if we have no contrary infomation, then that is the way to a s s u e that any g f v m a c t of eating was performed.Nonetheless,there c l e a r l y is a danger, and t h a t is a l l X am drawing attention to here, of taking inferences to a phenolnencw ldgical level beyond that of uammn sense. A clearer case, in my view, would be Schank's analysis (1974a) of mental. a c t i v i t y in which a l l actions, such as kicking a b a l l , say, are preceded by a rsrenta9 action af conc~iving or deciding to kick a b a l l . This is clearly a level of a n a l y s i s untrue to c a u m n sense, and which can have only harmful effects in a systea intended to mimic corxlaPon sense reasoning and understanding. Demupling Another general issue in dispute concerns what I shall c a l l demupling, which i s whethex ox n o t the actual parsing of t e x t or dialogue into an 'understanding system.' is essential. Charniak and Minsky b e l i e v e that t h i s inmy view, that is n o t so, because many of the later inferences would actually have to be done already, i n order to have achieved the initial parsing. For example, i n analysing 'He shot her with a colt', we cannot ascribe any structure at all until we'can make the infexences that guns rather than horses are instruments f o r shooting, and so such a sentence cannot be represented by an 'inference-but-no-parsing' structure, without aremithat language doas not have one of its esgential charaeteristlca, namely ayptemrrtie ambiguity. The essence of decoupling is allowing roprersntational etructures to have significance q u i t e indtlpendant of theirapplication, and that may lead one to a eituatMh lot essentially ditfstont frm that of the logician who simply asserts that ouch-and-much ie the 'right structuxel of sme sentence. The inferences required t o resolve word aense ambiguities, and those ad tb reaolva pronoun reference pxobletast are not of different typos1 oftan the two pmblaas occur i n a eingle sentence and must be resolved together.", "num": null }, "FIGREF24": { "type_str": "figure", "uris": null, "text": "view the c o n t r o l structure of an undkrstanding progran is itself of theoretical significance, for only in t h a t way, he believes, can natural l aone should m a k e massive forward inferences as one goes through a t e x t , keeping a l l one's expectations intact, as Charniak and Schank hold, 0s whether, as I hold,, one should adopt some 'laziness hypothesis1 &ut understanding, and generate deeper inferences o n l y when t h e system is unable to salve, say a referential problem by mre superEicia1 methods. Of, in other terns, should an understanding system be ~roblern-, or data-, driven. This i s not meant to be j u s t bland assertion. I have written at same length on the relations between application and the theoretical status of linguistic theories in (Wilks ' 7 4 ) . **The differences between Minsky's (19741 notion of 'default value' and what I have called 'prefexence' can be pointed up in terms of application.", "num": null }, "FIGREF25": { "type_str": "figure", "uris": null, "text": "Another d i f f i c u l t y 16 that it is not always clear whether the argument is about what p p l e are thought to do when they understand, or about how one should mnstruct an wdexstandLng system. This is a d i f f i c u l t matter about which to be precise: it would be possible, for example, to agree with Charniakts argurnentmd still construct a 3urely problem-driven inferencer on the ground that, at the mment, t h i s is the onlv way one can cope with t h e vast majority of inferences for understanding, since any system of inferences made Fn response to no particular problem in me t e x t is too hard to control in practice. Indeed, it i s noticeable t h a t the mst recent papers of Schank (1974a and 1974b) and Charniak (1974) have been considerably less forwardinference oriented than earlier ones. This Bispute fs prhaps m l y one of degree;&nb about tha posalbili t y of befining a degree af forward inf~rlene~e that alds the a d u t i s n o f later semantic problem without going Lntta w w e~s r r y depth. might useZuLly, W w p A bslaf?Ly, curntxast ths Shffarent d e s of JustiPicathan iolplieftly appealed t a by t h e s y s t : .~~~s d e u c x i k d earlier hn this paper. These seem to FIM? to &u: e to EOW t ( i ) Tn terns of the p w e x of the inferentiak syskea enpisyPd. This fama of j.ustifhcation has underlain the-earLy predicate caXculusbased language programs, and is behind Hayesr (1974) r e c e n t deaand that any foxmalism for natural language analysis should admit of a s e t thecttetic s~t i c s , in the Tarskfan sense, so as to gain 'Intellectuai respecta b i l i t y 1 , as he puts it. The same general type of justificatim is appealed to in sane degree. by systems8 with BLMNER-type f~xtmlikas. (ii) In terms of the provision and formalisation, in any terms including Elqglish, of We sorts of knowledge recpirea LO understand a r e a s oe ~~S W U S~. (iii) In terns af #e actual performance of a s):st~m, i~p l e m e n t d on a ckmguter, at a task agreed tu demonstrate understanding.", "num": null }, "FIGREF26": { "type_str": "figure", "uris": null, "text": "if we s t i c k to the simplest examples, that present no d i f f i -Though an interesting, potentially revolutionary, distinctlon seems to have bean introduced by a recant reviewer of many of the systems d i scmsed here, heteen the functionirig of a program and la 'program in itself1 : 'Only Winograd describas a program that is sufficiently impressive in itsell tc force us to M e h i s ideas seriously. The t6chniques of U k i others havc t a get by an whatever Fatuitive appeal t h e y can muster1. ( I s u d ' 7 4 ) culllty to the human p a d s ranb it must h ad$sittad that t c bdan M+ of th. p r a i a k r n t faults of t h o A . I . paradigm oC l m g u p a that it h.r vprnt too much t.ha M puzs1es U I I P~~~~Sthrm a n r t L U gnat b i~~l c u l t h s both ayst:-tfc axl e3 l i n q~I $ t I~. A n example of the f o m r wwld be UH devq10pmnL OL a =yakof undarstandictg texts or storias'that had ,my cdpaclty to r w r c r after having its expactations satisttad a h then, subsequently. Xructrrtul, A t prssent no systtm QC the B Q Z~ dascrhbed, uhethor of 9 ox whatever, has any such o a p e i t y t o rcrovsr, The sitrwrtiian is quite digferent fmp that in a dialogue, as i n Winograd's o ) r s t a~, where, on Being given each new piece of inEcmtlon, the s y s t e~l checks it against uhat i t ~W S~ to see if it is baing contradicted, and then behaves i n an appropriately puzkled way if it is. In fraroe or lexpectationl systems it is a11 too easy to mnstruct apparently trick, but LxesicaEliy plausible, examples that satisfy what was king l m k d for & then o v e r t u r n it. That p s s i b i l i t y is already b u i l t into the notation oE f r m or expectatian. A n example of Phil Hayes against my o m system will same: mnsider \"The hunter licked his gun all over, and the s t o c k tasted especially gwd\"", "num": null }, "FIGREF27": { "type_str": "figure", "uris": null, "text": "the overriding pxeference. Gut that is m t t h e same as a general solution to thq problem, which used to be called that of 'topic' in the cmputational semantics of the Fifties. There are no solutions to L1i$ problem available here and haw, though some suggestions have been made by Abelson 91974) and M~D e m o ' t t (1974) . nay 30 q e~ sp mepcpord 'a~cpaoreauf XTT-I w ' WeTw x~aao'l;3", "num": null }, "TABREF1": { "text": "", "type_str": "table", "html": null, "num": null, "content": "" }, "TABREF2": { "text": ", and his advocacy of the latter, have isolated modern generative linguistics from any effecti-test of the systems of ruhes it proposes.", "type_str": "table", "html": null, "num": null, "content": "
be 'Lntellectually respectable' a knowledge syptem must have n a t u r a l d i s t i n c t from everyday language) . However, i f Woods ' ' internal ' interpretation of tho 'meanings are procedurest slogah has certain drawbacks, so too does Winograd's, or what one might c a l l the 'external' i n t e r p r e t a t i o n . By Ghat I mean 'concentrate', ' c a l l ' , 'have', ' i n t w p r e t 1 , e t c , are -f o r a r o b t . received co-n-sense view of t h e i r work, Consider the following three assertions: (1) s system is an implementation of a transformational grammar (2) Winograd's scale revision in systematic thinking, where the,pa.radigru r e v i s e is ,the 'generati-paradigm1 of t h e Chmskyan linguists fChomsky 1 9 5 7 ) . F r a the A 1 pintof view, the generative linguistic tmrk of the last fifteen years has three principal defects. Firstly, the generation of sentences, with whatever attached structures, is not in anv interesting sense a dem-onstration of human understanding, nor is the separation of khe well-formed from the ill-fomed, by such methods-for understanding requires, at the very least, b t h the generation & sentences as parts of coherent discourse and some attampt to interpret, rather than qer5ly zeject, what seen to be ill-farmed utterances. Neither the transformational grammarians following Chomsky, nor Whether or not m e distinction was intended to hdve t h i s effect, it has meant that a y test sxtuation necessarily involves performance, which is wnaLdered vutsfde the province of serious linguistic stugy. And any embdiment of a svstein of rules in a computer, and assessment of its out put, would be perf~rmance. AI, too, is much concerned with the structure of linguistic processes, independent of any particular implementation,** ** Vide: \"Artificial Intellige~ce is the s.tudy of intellectual mechanisms W b u q x a d I t i s interesting 'to notice that Woods' i s , i n h a t t o apart from applications and apart f r a how such mechanisms are realised in the human or in animals. ' ' (McCarthy 1974)
" }, "TABREF4": { "text": "U n i t e d S ...*'", "type_str": "table", "html": null, "num": null, "content": "
A B a n l W Q p l t E ;~.
&my headachara?
NEI llEALT\"H
I shall r e t u r n
" }, "TABREF5": { "text": "", "type_str": "table", "html": null, "num": null, "content": "" }, "TABREF6": { "text": "", "type_str": "table", "html": null, "num": null, "content": "
) B Q P~~B X * m T V 2
-auob puaaoo a u~3 a p aq q u e~ +o~dwpx* 303 *?qb~m, w T~P ; L~u~& q s 3 t~j ,
s , p p x b o t q~ l o as*~7 s apnloxe p w 4 ~0~3 3 0~e 8 am y a j~.~o r f i m w A n W T W~
-a 4 5pax-f\"qszt3uponaaq uo~qau~aun? a , WX-THWTT 'qroqs pesew
ST^ sqaaf.wd 3 0 \"~0 1 \" 3 3 b~a~ am 3 q AWMBF27T-mW 3snm 3 1
*uer
" }, "TABREF7": { "text": "", "type_str": "table", "html": null, "num": null, "content": "" } } } }