| { | |
| "paper_id": "1991", | |
| "header": { | |
| "generated_with": "S2ORC 1.0.0", | |
| "date_generated": "2023-01-19T07:36:08.562530Z" | |
| }, | |
| "title": "PARSING WITHOUT PARSER", | |
| "authors": [ | |
| { | |
| "first": "Koiti", | |
| "middle": [ | |
| "Tsuda" | |
| ], | |
| "last": "Hasida", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "Institute for New Generation Computer Technology (ICOT)", | |
| "institution": "", | |
| "location": {} | |
| }, | |
| "email": "basida@icot.or.jp" | |
| }, | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "Hiroshi", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "Institute for New Generation Computer Technology (ICOT)", | |
| "institution": "", | |
| "location": {} | |
| }, | |
| "email": "" | |
| } | |
| ], | |
| "year": "", | |
| "venue": null, | |
| "identifiers": {}, | |
| "abstract": "In the domain of artificial intelligence, the pattern of information flow varies drastically from one context to another. To capture this diversity of information flow. a natural-language processing (NLP) system should consist of modules of constraints and one general con straint solver to process all of them; there should be no specialized procedure mo dule such as a parser and a generator. This paper presents how to implement such a constraint-based approach to NLP. Dependency Prop agation (DP) is a constraint solver which transforms the program (=constraint) represented in terms of logic programs. Co nstraint Un ification (Cui is a unification method incorporating DP. cu-Prolog is an extended Prolog which employs CU insteaq of the standard uni fication. cu-Prolog can treat some lexical and grammatical knowledge as constraints on the structure of gram matical categories\ufffd enabling a very straightforward im plementation of a parser using constraint-based gram mars. By extending DP, one can deal efficiently with phrase structures in terms of constraints. Computa tion on category structures and phrase structures are naturally integrated in an extended DP. The computa tion strategies to do all this are totally attributed to a very abstract, task-independent principle: prefer com putation using denser information. Efficient parsing is hence possible without any parser.", | |
| "pdf_parse": { | |
| "paper_id": "1991", | |
| "_pdf_hash": "", | |
| "abstract": [ | |
| { | |
| "text": "In the domain of artificial intelligence, the pattern of information flow varies drastically from one context to another. To capture this diversity of information flow. a natural-language processing (NLP) system should consist of modules of constraints and one general con straint solver to process all of them; there should be no specialized procedure mo dule such as a parser and a generator. This paper presents how to implement such a constraint-based approach to NLP. Dependency Prop agation (DP) is a constraint solver which transforms the program (=constraint) represented in terms of logic programs. Co nstraint Un ification (Cui is a unification method incorporating DP. cu-Prolog is an extended Prolog which employs CU insteaq of the standard uni fication. cu-Prolog can treat some lexical and grammatical knowledge as constraints on the structure of gram matical categories\ufffd enabling a very straightforward im plementation of a parser using constraint-based gram mars. By extending DP, one can deal efficiently with phrase structures in terms of constraints. Computa tion on category structures and phrase structures are naturally integrated in an extended DP. The computa tion strategies to do all this are totally attributed to a very abstract, task-independent principle: prefer com putation using denser information. Efficient parsing is hence possible without any parser.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Abstract", | |
| "sec_num": null | |
| } | |
| ], | |
| "body_text": [ | |
| { | |
| "text": "The information-processing capacity of a cogmt1ve agent is severely limited, whereas the world in which it finds itself contains a vast amount of information which might be relevant to its survival. A cognitive agent is thus destined to face partiality of information. That is, information processing by a cognitive agent is limited to a very small part of the potentially relevant information. In the domain of artificial intelligence in general and natural language processing in particular, therefore, the pattern of information flow varies very *The order is not significant. 1 tsudafYicot.or.jp drastically from one context to another. This is nec essary in order for a cognitive agent to have chance of access to the entire domain of potentially relevant information across various different contexts.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "Due to this diversity of information flow, it is prac tically impossible to stipulate which pieces of informa tion to process in which order. Consider the case of comprehension of natural language sentences . for in stance. Parts of phonological information might be missing due to noise. and it may well be impossible to predict which part would be missing. Similarly. parts of syntactic information could be insufficient. gi,\u2022ing rise to syntactic ambig11ity. Semantic information also would be partially abu11dant or missing due to familiar ity or ignorance to the topic. and so on. It is therefore utterly implausible to :-uppose that all phonological in formation is processed prior to syntactic informat ion. or that syntactic information is processed before se mantic information.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "Accordingly. it is not at all a promising approach to AI or NLP to stipulate information flow totallv. as procedural programs do. In particular. a hierarcl1ical architecture consisting of mo dules of procedures fa ils to capture very complex. multi-directional. informat ion flow in the domains su'ch as :>iLP. because procedures stipulate what is input and what is output. se,\u2022erely restricting the global information flow across the entiw system. This is what happens in the prevalent architPc ture of NLP systems consisting of a sequence of proce dure modules such as, say. syntactic analyzer. semant ic analyzer, pragmatic analyzer. generation planner. and surface generator.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "The design of AI systems should abstract awa\\\u2022 in formation flow in accordance with its diversitv. 1 \u2022T his is where constraint paradigm [1-l] comes in. Si\ufffdce co n straint, or declarative program, does not stipulate pro-cessing order, it \u2022 does not restrict information flow \u2022 so severely as procedures do . and thus can capture the diversity of information flow. 2 In the constraint paradigm. a NLP system involves mo dules of linguistic (syntactic. semantic, pragmatic, and so on.) and extralinguistic constraints. Whether there are different constraint solvers for different mod ules of constraints is not a light question. but we strongly suspect the answer is no. If yes . the communi cation between different modules would be too cumber some to allow the massive interaction required in NLP. For instance. it would not be a very good idea to have a constraint solver specialized for processing syntactic in formation. Thus we employ a radical constraint\ufffdbased viewpoint: just one very general constraint solver deals with all the different constraints. giving rise to diverse communication across them. 3 The task of NLP is hence divided into modules of constraints rather than mod ules of pro cedures as has been traditionally done. As a matter of course, a NLP system should include no parser.", | |
| "cite_spans": [ | |
| { | |
| "start": 364, | |
| "end": 365, | |
| "text": "2", | |
| "ref_id": "BIBREF1" | |
| }, | |
| { | |
| "start": 1109, | |
| "end": 1110, | |
| "text": "3", | |
| "ref_id": "BIBREF2" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "In the rest of the paper. we will concentrate on pars ing. Efficient parsing will be shown to emerge from our constraint solver. which is a general constraint trans formation metho d employing se\\\u2022eral heuristics derived from the following very abstract. task-independent principle:", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "( 1) Prefer computation using denser information.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "That is. efficient parsing is attributed to this princi ple. This is regarded as an impressive demonstration of the feasibility of our constraint-based approach, be cause parsing is almost the only subproblem of NLP where there are endorsed efficient algorithms, mainly for dealing with phrase structures.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "Our constraint solver. called Dependency Propaga tion [8, 7] , deals with constraints in a combinatorial domain. unlike the constraint solvers embedded in most constraint logic programming (CLP) languages [2, : J , 9] . Section 2 describes how to parse ambiguous sentences with a CLP language, caUed cu-Prolog, which embe_ds an early version of DP. Although the ambiguity treated in Section :2 concerns only the structures of grammat ical categories. Section :3 applies DP itself to pars ing phrase-structure. typically formulated in terms of context-free grammars. It will be shown that efficient parsing procedures such as Earley's algorithm simply emerge from general processing strategies employed in a revised version of DP. Section -1 demonstrates that 2 Constraint is not the only approach to diversity of informa tion flow. For instance, blackboard arc hitecture is also regarded as aiming at the same thing. Coroutine implemented in lan guages such as CONNIVER [13] is another example. The reason why we employ constraint paradigm is twofold. First. it comes with intuitive declarative semantics. Secqnd. it implements the diversity of information flow at fint:>r-grained levels than might be captured in the other approaches . . 3 Thus we consider that General Problem Solver was basically on the right track. Its alleged failure was simply due to the immaturity of programming technologies.", | |
| "cite_spans": [ | |
| { | |
| "start": 54, | |
| "end": 57, | |
| "text": "[8,", | |
| "ref_id": "BIBREF7" | |
| }, | |
| { | |
| "start": 58, | |
| "end": 60, | |
| "text": "7]", | |
| "ref_id": "BIBREF6" | |
| }, | |
| { | |
| "start": 205, | |
| "end": 217, | |
| "text": "[2, : J , 9]", | |
| "ref_id": null | |
| }, | |
| { | |
| "start": 970, | |
| "end": 974, | |
| "text": "[13]", | |
| "ref_id": null | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "both types of information. about category structures and phrase structures . are processed efficiently in a naturally integrated manner by very general heuristics. Finally. Section . j concludes the paper. ", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "For the sake of expository simplification. in this paper we restrict ourselves to Horn clauses . although DP is not actually so limited.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Dependency Propagation", | |
| "sec_num": "2.1" | |
| }, | |
| { | |
| "text": "Dependency triggers constraint transformation in DP. Two occurrences of the same variable in a clause constitutes a dependency when both occurrences do not occupy any vacuo\u2022 us argument place. An argument place of an atomic fo rmula is said to be vacuous when a variable filling that argument place is never instanti ated by evaluating that atomic formula. For instance. the first argument place of predicate member defined below is vacuous.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Dependency Propagation", | |
| "sec_num": "2.1" | |
| }, | |
| { | |
| "text": "In the following clauses. ( : 3 ) has no dependency. and (-1) has a dependency because it is equivalent to (-S).", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "(:2) a. member(E ,[ EI-]) . b. member (E , [_ IS] ) :-member (E, S) .", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "(:3) : -member (a, X) . In DP. computation proceeds so as to eliminate de pendency. Note that this is a more general control schema than Earley <;!eduction (10] , which executes the body of each clause in the fixed left-to-right order.", | |
| "cite_spans": [ | |
| { | |
| "start": 156, | |
| "end": 160, | |
| "text": "(10]", | |
| "ref_id": "BIBREF9" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "(:2) a. member(E ,[ EI-]) . b. member (E , [_ IS] ) :-member (E, S) .", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "Basically, fu sion replaces one or more literals \\V ith another: so as to eliminate dependency. Fusion is a sort of unfold/fold transformation for logic programs (1-5]. For example, member(X, [a,b] ) is replaced by cO(X) . where cO is a new predicate defined as follows.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "(:2) a. member(E ,[ EI-]) . b. member (E , [_ IS] ) :-member (E, S) .", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "That is, two atomic formulas, member(X , Y) and Y= [a, b] 4 have been fused to one atomic formula cO (X) .", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "(:2) a. member(E ,[ EI-]) . b. member (E , [_ IS] ) :-member (E, S) .", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "The principle ( 1) provides us some heuristics for con trolling fusion. For example, the elimination of de pendency involving a variable binding, as in p (a, X) , should have higher priority than the elimination of de pendency between two ordinary atomic formulas, as in p (X) , q (X) . We will discuss heuristics further along this line later.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "(:2) a. member(E ,[ EI-]) . b. member (E , [_ IS] ) :-member (E, S) .", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "A program of cu-Prolog is a set of Constraint-Added Horn Clauses (CAHCs), A CAHC is a Horn Clause followed by constraints: The prolog part ( the head plus the body) of a C AHC is processed procedurally just as in Prolog, whereas the constraint part is dynamically transformed with a sort of unfold/fold transformation during the execution of the former part. The following is the inference rule of cu-Prolog:", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "cu-Prolog", | |
| "sec_num": "2.2" | |
| }, | |
| { | |
| "text": "A, K;C. ,-A': -L:D., 0 = mgu(A, A'), C' = dp (C0 1 D0) L0,K0; C'", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "cu-Prolog", | |
| "sec_num": "2.2" | |
| }, | |
| { | |
| "text": "A and A' are atomic formulas. K, L, C. D, and C' are s_ equences of atomic formu las. mg u(A, A' ) is the most general unifier between A and A'. dp ( C) is a modular constraint that is equivalent to C.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "cu-Prolog", | |
| "sec_num": "2.2" | |
| }, | |
| { | |
| "text": "If C is inconsistent 1 the application of the above infer ence rule fails because dp( C) does not exist.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "cu-Prolog", | |
| "sec_num": "2.2" | |
| }, | |
| { | |
| "text": "The following holds if Ci and C j share no variable:", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "cu-Prolog", | |
| "sec_num": "2.2" | |
| }, | |
| { | |
| "text": "(7) dp( C) = dp(Ci), \u2022 . . , dp(C n )-", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "cu-Prolog", | |
| "sec_num": "2.2" | |
| }, | |
| { | |
| "text": "For example,", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "cu-Prolog", | |
| "sec_num": "2.2" | |
| }, | |
| { | |
| "text": "(8) d p (m ember(X, [a, b,c]), member(X, [b ,c,d]), app(U,V) ) returns a new constraint cO (X), app (U , V)", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "cu-Prolog", | |
| "sec_num": "2.2" | |
| }, | |
| { | |
| "text": ", where the definition of cO is (9) cO (b) .", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "cu-Prolog", | |
| "sec_num": "2.2" | |
| }, | |
| { | |
| "text": "cO (c) .", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "cu-Prolog", | |
| "sec_num": "2.2" | |
| }, | |
| { | |
| "text": "but (10) d p (member(X,[ a,b;c]),member(X, [k, l,m]))", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "cu-Prolog", | |
| "sec_num": "2.2" | |
| }, | |
| { | |
| "text": "is not defined . ", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "cu-Prolog", | |
| "sec_num": "2.2" | |
| }, | |
| { | |
| "text": "In cu-Prolog. unification-based grammar such as HPSG or JPSG can be implemented naturally by treating the constraints formulated in those theories almost as they are. Figure l shows an example session of the .J PSG parser when it processes an ambiguous sentence.:; Be low we discuss two examples of C AHC in the .JPSG parser in cu-Prolog [18] . The first example concerns how to pack lexical am biguity. The following is the lexical entry of a Japanese polysemic noun \u2022 \u2022hasi\" that means bridge. chopsticks. or edge depending on contexts. and predicate has Lsem is defi ned as follows .", | |
| "cite_spans": [ | |
| { | |
| "start": 338, | |
| "end": 342, | |
| "text": "[18]", | |
| "ref_id": "BIBREF18" | |
| } | |
| ], | |
| "ref_spans": [ | |
| { | |
| "start": 167, | |
| "end": 175, | |
| "text": "Figure l", | |
| "ref_id": null | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "JPSG parser in cu-Prolog", | |
| "sec_num": "2.3" | |
| }, | |
| { | |
| "text": "(12) hasi_sem(structure ,bridge) .", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "JPSG parser in cu-Prolog", | |
| "sec_num": "2.3" | |
| }, | |
| { | |
| "text": "Constraint hasLsem (TYPE , \u25a1BJ) represents various meanings of 1 'hasi\" and the ambiguity may be resolved during the parsing process when other constraints are imposed. Be_ cause such ambiguity is considered at one time, instead of divided into separate lexical ent ries. parsing process can be efficient.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "hasi_sem(tool , chopst icks) . hasi _sem (place ,edge) .", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "In the second example, various feature principles of unification-based grammar are embedded in a phrase structure rule as constraints. The following clause shO\\vs the foot feature principle of JPSG: the foot fea ture value of the mo ther unifi es \\V ith the union of those of her daughters.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "hasi_sem(tool , chopst icks) . hasi _sem (place ,edge) .", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "(", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "hasi_sem(tool , chopst icks) . hasi _sem (place ,edge) .", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "1:3 ) p sr( [ff (MS)] , [ff (LDS)] , [ff (RD-S)] ) ;", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "hasi_sem(tool , chopst icks) . hasi _sem (place ,edge) .", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "un ion (LDS , RDS , MS) . p sr(Mother ,LeftJ)aughter,Head) is a phrase structure rule followed by the constraint un ion{LDS ,RDS ,MS) which represents the foot fea ture principle. MS.LOS, and RDS are foot features of mot her. left daughter, and right daughter respectively. The constraint is flexibly processed 1, v ith the const raint transformation mechanism with a heuristic.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "hasi_sem(tool , chopst icks) . hasi _sem (place ,edge) .", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "In traditional Prolog, these principles are supposed to be implemented in the following procedural way: (14) ", | |
| "cite_spans": [ | |
| { | |
| "start": 104, | |
| "end": 108, | |
| "text": "(14)", | |
| "ref_id": "BIBREF14" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "hasi_sem(tool , chopst icks) . hasi _sem (place ,edge) .", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "I I I l __ v[vs2 , SC{p [ga] , p[wo]}J : [love ,ken , 0bj 0_415] ---[ai] I l __ v[Form_675 , AJA{v [vs2 , SC{p [wo]}]}, AJN{Adj _677} , SC{SubCat _679}] :SEM_681---[suru] cat cat(v, Form_675 , [] , Adj _677 , SubCat_679 , SEM_681) cond c7 (Form_675 , SubCat_679 , 0bj", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "hasi_sem(tool , chopst icks) . hasi _sem (place ,edge) .", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "The JPSG parser discussed in Section 2, however, can not handle ambiguity on phrase structures because the parsing algorithm is written only in the Prolog part of CAHC. This section shows that chart parsing is natu rally derived from a very general control strategy of an extended version of DP.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Processing Phrase Structure", | |
| "sec_num": "3" | |
| }, | |
| { | |
| "text": "Let us consider the following extremely simple context free grammar. This clause has a dependency concerning Y. Then, the parsing process continues .", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Context-Free Parsing by Fusion", | |
| "sec_num": "3.1" | |
| }, | |
| { | |
| "text": "This transformation process is exempt from the infi nite loop due to left recursion. unlike DCG of the stan dard type, because fusion includes some sort of tabu lation technique [16] . If we had A 0 = [b I A 1 J instead of A 0 = [a lA 1 ], for instance, we would have the following instead of (16) . (19) with j = i. then ri+l will be created by \u00a3usion of ri (Y) and p(Y, Z). whichever literal might be unfolded. and a definition clause of ri+l will be ( 19) with j = i + 1. (19) ", | |
| "cite_spans": [ | |
| { | |
| "start": 178, | |
| "end": 182, | |
| "text": "[16]", | |
| "ref_id": "BIBREF16" | |
| }, | |
| { | |
| "start": 293, | |
| "end": 297, | |
| "text": "(16)", | |
| "ref_id": "BIBREF16" | |
| }, | |
| { | |
| "start": 300, | |
| "end": 304, | |
| "text": "(19)", | |
| "ref_id": "BIBREF19" | |
| }, | |
| { | |
| "start": 453, | |
| "end": 458, | |
| "text": "( 19)", | |
| "ref_id": "BIBREF19" | |
| }, | |
| { | |
| "start": 475, | |
| "end": 479, | |
| "text": "(19)", | |
| "ref_id": "BIBREF19" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Context-Free Parsing by Fusion", | |
| "sec_num": "3.1" | |
| }, | |
| { | |
| "text": "Note that fusion of rJ (Y) or p (Y , Z) with any other lit eral never takes place, because Y is constrained now here else and the second argument place of p is vacuous .", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "rJ(Z) :-rJ (Y) , p(Y ,Z) .", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "Since (29) is (19) ", | |
| "cite_spans": [ | |
| { | |
| "start": 14, | |
| "end": 18, | |
| "text": "(19)", | |
| "ref_id": "BIBREF19" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "rJ(Z) :-rJ (Y) , p(Y ,Z) .", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "As indicated here, the first argument of q must always unify with X in the first clause. whereas its second ar gument has no such restriction.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "The information of X=f (Y) makes X penetrate through p (X, Y) . creating new predicate q. as follows: (23 ) : -q(X ,Y) , p(X,Z) , X=f (Y) , Y=g (Z) . q ( X , A) : -X =f (A) , q (A) . q(X , a) .", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "In the following discussion, a penetrated variable is written with a superscript like X 1 \u2022 and called a transclausal variable. which roughly corresponds to the global variable of programming languages such as Pas cal and C. A transclausal variable may be treated as if it were a constant. Accordingly, a penetrated ar gument place are omitted for the sake of expository simplification. For instance, (23) may be rephrased as follO\\vs:", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "The information of X=f (Y) makes X penetrate through p (X, Y) . creating new predicate q. as follows: (23 ) : -q(X ,Y) , p(X,Z) , X=f (Y) , Y=g (Z) . q ( X , A) : -X =f (A) , q (A) . q(X , a) .", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "( 2 4 )q (Y) , p(X 1 ,Z) , X 1 =f (Y) , Y=g (Z) . -X 1 =f (A) , q(A) .", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 50, | |
| "end": 68, | |
| "text": "-X 1 =f (A) , q(A)", | |
| "ref_id": "FIGREF3" | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "The information of X=f (Y) makes X penetrate through p (X, Y) . creating new predicate q. as follows: (23 ) : -q(X ,Y) , p(X,Z) , X=f (Y) , Y=g (Z) . q ( X , A) : -X =f (A) , q (A) . q(X , a) .", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "Just as fosion. penetration has two cases: unfolding and fo lding. An unfolding, such as this case, introduces a new predicate, whereas a folding does not. Binding X 1 =f (B) in the second clause unifies with X 1 =f (Y) . and the resulting binding, X 1 =f (Y 1 ) . is shared by the first and the second clause:", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "q(a) .", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "(2. S ) :-q(Y) , p(x 1 ,z) , X 1 =f (Y 1 ), Y 1 =g (Z) . q ( 1 y) X 1 =f ( y 1 ) , q ( y 1 ) .", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "q(a) .", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "q ( a) .", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "q(a) .", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "X 1 may penetrate through p (X 1 , Z) as well:", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "q(a) .", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "(26) :-q(Y) , q(Z) , X 1 =f (Y 1 ), Y 1 =g(Z) . q(Y 1 ) X 1 =f (Y 1 ), q(Y 1 ).", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "q(a) .", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "This is a folding case of penetration. A typical pattern of penetration is shown in Figure 2 . p ( \u2022, \u2022 ) s in the left-hand side of the figure all \u2022have the same sign, and those in the right-hand side all have the opposite sign. That is, either p ( \u2022 ,e) s in the left are all body literals and those in the right are all head literals, or vice versa. o represents a penetrating variable. We say that this penetration is downward in the former case, and upward in the latter. The penetration to get (23) and (26)' is downward.", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 84, | |
| "end": 92, | |
| "text": "Figure 2", | |
| "ref_id": "FIGREF13" | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "q(a) .", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "For \ufffd :=:; i :=:; n, W\ufffd is a duplication of W i except that p ( \u2022 , \u2022 ) has been replaced by q ( \u2022 , \u2022 ) . When <I> i and \\JI j are the same clause for some i and j, the situation will be more complicated in the sense that the duplication increases not only the right-hand half of the figure but also the left-hand half. The example shown in the next subsection includes some such cases.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "q(a) .", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "As shown in the lower part of the figure, second or later penetration of o through the first argument of p is a folding, reusing q without introducing a new pred icate. Corresponding unfolding and folding must be in the same direction: upward or downward. Otherwise the original combinations of clauses are not preserved. Like fusion, penetration is also triggered by depen dency. In penetration, however 1 dependency may be transclausal. In ( 26 ), for instance, the dependency between Y=g (Z) and q (Y) could trigger penetration. This dependency is transclausal and involves a binding Y=g(Z) . In the case of upward penetration, the depen dency in question involves a head literal.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "q(a) .", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "To control computation. we must decide which de pendency to trigger a penetration into which direction. The general principle ( 1) suggests the following heuris tic in this respect. (276) guarantees that the resulting structure should have more homogeneous information distribution. in creasing the entropy of the entire system. For example, a binding in the top clause is consid ered to have much more information than bindings in the other clauses, in the sense that the atomic formulas in the top clause should primarily hold; if they do not, then we do not care whether the atomic formulas in the other clauses hold or not. The downward penetra tion occurring twice in the above example is motivated accordingly, because it is based on the information of X=f (Y) in the top clause.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "q(a) .", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "Now we demonstrate that Earley's algorithm natu rally emerges from penetration controlled by the above heuristic. We consider the simple CFG example (13) again.", | |
| "cite_spans": [ | |
| { | |
| "start": 149, | |
| "end": 153, | |
| "text": "(13)", | |
| "ref_id": null | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Emergence of Chart Parsing", | |
| "sec_num": "3.3" | |
| }, | |
| { | |
| "text": "(13) , _ _ p(A 0 ,\u2022B) , A 0 =[a l A 1 ] , \u2022 \u2022\u2022; A n -,l =[alA n ]. p( [a IX] ,X) .", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Emergence of Chart Parsing", | |
| "sec_num": "3.3" | |
| }, | |
| { | |
| "text": "The following is obtained \u2022 by downwa: rd penetration of A 0 through p(A 0 , B), which is u\u2022 nfolded. Now we have a non-vacuous dependency concerning Y. because p0 says something substantial about the in stantiation of its argument. The head p0 (A 0 ) of the first definition clause of p0 has transclausal variable A 0 7 as the argument. Since A 0 has been introduced in the top clause. upward penetration is applied here, so that the first definition clause of p 0 is replaced by Po.1 .. and a new definition clause is introduced. as follovvs.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "p(X,Z) :-p(X,Y) \ufffd p(Y\ufffdZ) .", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "(30) Po.1. Po (Z) po ( Z ) Po.1 , p (A 1 , z) . po (Y) , p(Y,Z ) .", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "p(X,Z) :-p(X,Y) \ufffd p(Y\ufffdZ) .", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "The last clause of ( 28 ) has been replicated vvhile p 0 (Y) therein has been replaced by p0, 1 plus Y = A 1 . giving rise to the second clause in ( : 30) above. );\u00b0ote that p(Y) no longer imposes any restriction on the instantiation of Y. The dependency concerning Y in the third clause here is vacuous and left untouched for the time being. A problem here. incidentally. is that another top clause as below is created.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "p(X,Z) :-p(X,Y) \ufffd p(Y\ufffdZ) .", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "( :3 1) : -Po.1 , 8=A 1 , A 0 =[a l A 1 ], An-l =[a l A n ].", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "p(X,Z) :-p(X,Y) \ufffd p(Y\ufffdZ) .", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "To avoid two top clauses. we could introduce a new predicate q by which to mediate between the top clause and the locus of upward penetration:", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "p(X,Z) :-p(X,Y) \ufffd p(Y\ufffdZ) .", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "(32) : -q, A 0 =[a lA 1 ] , \u2022 \u2022 \u2022, A n -l =[alA n ] . q Po.1 , B 0 =A 1 . q : -Po (B 0 ).", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "p(X,Z) :-p(X,Y) \ufffd p(Y\ufffdZ) .", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "Next. p (A 1 , Z ) in the second clause of ( :30 ) is un folded and a new predicate p 1 is creat\ufffdd. A 1 penetrating downwards:", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "p(X,Z) :-p(X,Y) \ufffd p(Y\ufffdZ) .", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "( ; 3; 3 ) Po (Z )", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "p(X,Z) :-p(X,Y) \ufffd p(Y\ufffdZ) .", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "Po.1 , P1 (Z) .", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "p(X,Z) :-p(X,Y) \ufffd p(Y\ufffdZ) .", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": ").", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "P1 (A 2", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "Operation prnceeds similarly. yielding the clauses f\ufffdt- ", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "P1 (Z ) : -P1 (Y) , p(Y,Z ) .", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "Part of ( :3, j ) amounts to a well-formed substring ta ble. as in CYK algorithm. Earley's algorithm [4] , chart parser. and so on. For instance. the existence of clause Pi.k : -Pi.; , P;.k . means that . as illustrated in Fig. : J . the part of the given string from position i to posi tion k has been parsed as having category P and is subdivided at position j into two parts. each having category P. Note that the computational complexity of the above process is 0( n 3 ) in terms of both space and time. Moreover, the space complexity is reduced to 0( n 2 ) if we delete the literals irrelevant to instantiation of vari ables. which preserves the semantics of the constraints in the case of Horn programs. That is i the resulting structure would be:", | |
| "cite_spans": [ | |
| { | |
| "start": 101, | |
| "end": 104, | |
| "text": "[4]", | |
| "ref_id": "BIBREF3" | |
| } | |
| ], | |
| "ref_spans": [ | |
| { | |
| "start": 223, | |
| "end": 233, | |
| "text": "Fig. : J .", | |
| "ref_id": null | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Computational Complexity", | |
| "sec_num": "3.4" | |
| }, | |
| { | |
| "text": "(:36 ) :-q , A 0 =[alA 1 ] ,. \u2022 \u2022 \u2022 , A n -l=[a l A n ]. q : -Po (B o ). q : -B 0 =A i \u2022 ( 0 < i \ufffd n) Pi(Z ) Pi(Z }. (0 \ufffd i < j < n) P-i (Z) : -P_i (Y) , p(Y ,Z) . (0 \ufffd i < n )", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Computational Complexity", | |
| "sec_num": "3.4" | |
| }, | |
| { | |
| "text": "Some sort of clauses listed here might be generated more than once in general cases where the grammar is less trivial than ( 1:3). For example. clause ( 37) may be derived from both (: 38 ) and ( : 39). If (37) is generated twice. then of course we are able to collapse the two instances to one, so that the space complexity should be 0( n 2 ). Needless to say, this col lapsing operation is totally domain-independent in its nature.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Computational Complexity", | |
| "sec_num": "3.4" | |
| }, | |
| { | |
| "text": "The process illustrated above corresponds best to Earley\u2022s algorithm. Our procedure may be general ized to employ more bottom-up control. so that the resulting process should be regarded as chart parsing in general. including left-corner parsing. and so on.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Computational Complexity", | |
| "sec_num": "3.4" | |
| }, | |
| { | |
| "text": "Section 2 treats linguistic constraints on category struc tures as constraint transformation. and Section : 3 pro cessed linguistic constraints on phrase structures. This section discusses how to handle various types of con straints mentioned in the previous two sections. Some heuristics will be needed to determine which constraint to process earlier than the others.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "4 Integrat ed Processing", | |
| "sec_num": "8" | |
| }, | |
| { | |
| "text": "In the following discussion, we consider two types of linguistic constraints: constraints on category struc ture and those on phrase structure. For simplicity. the former constraints are represented only by pred icate c. and the latter p. Accordingly. we intro duce two types of dependency: inter-dependency and intra-dependency. Inter-dependency is a double oc currence of a variable in both types of constraints , such as X in p(X) , c(X , Y) . Intra-dependency arises with non-variable arguments or a variable that occurs only in one type of constraints such as c(a,X) or Y in c(a, Y) ,c(Y, b) .", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Heuristics", | |
| "sec_num": "4.1" | |
| }, | |
| { | |
| "text": "By applying the general heuristic (27) to this do main. we get the fo llowing heuristic:", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Heuristics", | |
| "sec_num": "4.1" | |
| }, | |
| { | |
| "text": "\u2022 Eliminate intra-dependencies earlier than inter dependencies.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Heuristics", | |
| "sec_num": "4.1" | |
| }, | |
| { | |
| "text": "\u2022 Eliminate intra-dependencies in category struc ture earlier than those in phrase structure.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Heuristics", | |
| "sec_num": "4.1" | |
| }, | |
| { | |
| "text": "\u2022 In eliminating inter-dependencies. the literal that has the fewer OR-alternatives should be un folded ( penetrated downward).", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Heuristics", | |
| "sec_num": "4.1" | |
| }, | |
| { | |
| "text": "That is. constraints on category structures generally has more information quantity than those on phrase\ufffd structure. because the former are called by the latter.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Heuristics", | |
| "sec_num": "4.1" | |
| }, | |
| { | |
| "text": "In the case of a dependency between two argument places of ordinary atomic formulas , moreover. pene tration operation should take place at the one that has fewer alternatives of unfolding. because it is supposed to have more information quantity:", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Heuristics", | |
| "sec_num": "4.1" | |
| }, | |
| { | |
| "text": "The following is an ambiguous context free grammar that parses ''I see a man with a telescope.\" Apply upward penetrc1: tion to (44). Here p 0 ,1 is equiv alent to p 0 (A 1 ,v) . Unfold the category constraint of (49). 8 7 From unification-based point of view, suppose each category has the form [pQs/X] and c ( ) represents the pos fe ature prin ciple:", | |
| "cite_spans": [ | |
| { | |
| "start": 218, | |
| "end": 219, | |
| "text": "8", | |
| "ref_id": "BIBREF7" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Example", | |
| "sec_num": "4.2" | |
| }, | |
| { | |
| "text": "The combination of the values of pos fe ature of mother, left daughter, and right daughter cate gory is (vp ,n,np) , (np ,np ,pp) , or (vp , vp ,pp). 1 ,p(A 1 ,B,RC 1 ) ,co(Cat) .", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 150, | |
| "end": 168, | |
| "text": "1 ,p(A 1 ,B,RC 1 )", | |
| "ref_id": "FIGREF3" | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Example", | |
| "sec_num": "4.2" | |
| }, | |
| { | |
| "text": "However, c 0 has only one definition clause:", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Example", | |
| "sec_num": "4.2" | |
| }, | |
| { | |
| "text": "c 0 (vp) :-RC 1 =np .", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Example", | |
| "sec_num": "4.2" | |
| }, | |
| { | |
| "text": "So c 0 is reduced and you get (50) .", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Example", | |
| "sec_num": "4.2" | |
| }, | |
| { | |
| "text": "( . j0) Po (B ,vp) :-po,1 ,p(A 1 ,B,np) . Now the remaining clauses are (43), ( 47 ). (48). (46) and (. jQ ). Apply downward penetration in terms of A 1 to ( . SO ). p1 (B) is equivalent to p(A 1 ,B,np) .", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "9", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "( -Sl) p0 (B,vp) : -po,1 , p1 (B) . (-52) P1 (A 2 ).", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "9", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "( 53) p 1 (Z) :-p (A 1 ,Y,np) ,p(Y,Z,RC) ,c(np,RC , np) .", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "9", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "Cnfold the category structure constraint of (. j: J). ", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "9", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "In this paper, we have shown that various parsing tech niques are subsumed in a general procedure of con straint transformation, whose control heuristic is at tributed to an abstract, task-independent principle ( 1 ). Thus our conclusion is that no parser at all is needed in natural language processing, It is both desirable. as is discussed first in the paper, and possible, as we have so far demonstrated, for an NLP system to have no particular module for parsing sentences, just as a car has no particular part for driving towards the east or turning to the left .", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Concluding Remarks", | |
| "sec_num": "5" | |
| }, | |
| { | |
| "text": "Our approach will capture sentence generation as well, if we employ a more adequate control heuristic. which could also be derived from ( 1 ). In this connec tion, Shieber [12] , among others, has also proposed a computational architecture by which to unify sentence parsing and generation, but his method is primarily specific to phrase-structure synthesis. A significant merit of our approach is that, as shown above, it is not in any way restricted to parsing or generation of context-free languages. Also, no additional mechanism is required to extend the underlying grammatical for malism so that grammatical categories may be com plex feature bundles, as is the case with GPSG. LFG. HPSG, and so on.", | |
| "cite_spans": [ | |
| { | |
| "start": 172, | |
| "end": 176, | |
| "text": "[12]", | |
| "ref_id": "BIBREF11" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Concluding Remarks", | |
| "sec_num": "5" | |
| }, | |
| { | |
| "text": ". At any rate, heuristics play the most important role in our approach. As this paper only gave a.n intuitive ration\ufffdle on some heuristics in terms of information quantity, more formal. account of them is yet to be worked out. A promising direction seems to be to define some sort of potential energy over constraints, which should capture information density, providing not only processing control but also preference of conclusion. In troducing hierarchies in the constraint is regarded as along the same line.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Concluding Remarks", | |
| "sec_num": "5" | |
| }, | |
| { | |
| "text": "Of course, there are some aspects of cognitive process where information flow is rather restricted. Typical examples are found in low-level aspects of perception and motor control. Informa tion fl ow may be stipulated to some adequate extent in the design of those subsystems. Nevertheless. diversity of information flow must be captured across different dimensions even in these cases. as is indicated by R.Brooks [1]; In his robot. although informa tion flow in each module may be regarded as uni-directional and there is only a little interaction between different modules. input information is not restricted to flow all the way through t>vPry module before output information is tailored.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "cu-Prolog is implemented in C language on l: \ufffdIX 4.2/38S0This example is on SYM\ufffdlETRY machine[19].", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "A i represents the constant list of length (n-i) whose ele ments are ''a\" s.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| } | |
| ], | |
| "back_matter": [], | |
| "bib_entries": { | |
| "BIBREF0": { | |
| "ref_id": "b0", | |
| "title": "telligence without Represen tation, technical report", | |
| "authors": [ | |
| { | |
| "first": "R", | |
| "middle": [], | |
| "last": "Brooks", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1988, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Brooks, R. ( 1988) In telligence without Represen tation, technical report, AI Laboratory, MIT.", | |
| "links": null | |
| }, | |
| "BIBREF1": { | |
| "ref_id": "b1", | |
| "title": "An Introduction to Prolog III", | |
| "authors": [ | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "Colme", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "A", | |
| "middle": [], | |
| "last": "Rauer", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1987, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Colme . rauer, A. (1987) An Introduction to Prolog III, unpublished manuscript.", | |
| "links": null | |
| }, | |
| "BIBREF2": { | |
| "ref_id": "b2", | |
| "title": "Solving a Cutting-Stock Problem in Con straint Logic Programming", | |
| "authors": [ | |
| { | |
| "first": "M", | |
| "middle": [], | |
| "last": "Dincbas", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "H", | |
| "middle": [], | |
| "last": "Simonis", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "P", | |
| "middle": [], | |
| "last": "Van Hentenryck", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1988, | |
| "venue": "Proceedings of the -5th International Conference of Logic Programming", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "42--58", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Dincbas, M., Simonis, H. and Van Hentenryck, P. (1988) 'Solving a Cutting-Stock Problem in Con straint Logic Programming,\u2022 Proceedings of the -5th International Conference of Logic Programming. pp. 42-58.", | |
| "links": null | |
| }, | |
| "BIBREF3": { | |
| "ref_id": "b3", | |
| "title": "An Efficient Context-Free Pars ing Algorithm: Co mmunications of ACM", | |
| "authors": [ | |
| { | |
| "first": "J", | |
| "middle": [], | |
| "last": "Earley", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1970, | |
| "venue": "", | |
| "volume": "13", | |
| "issue": "", | |
| "pages": "94--102", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Earley, J. (1970 ) 'An Efficient Context-Free Pars ing Algorithm: Co mmunications of ACM. Vol. 13, pp. 94-102.", | |
| "links": null | |
| }, | |
| "BIBREF4": { | |
| "ref_id": "b4", | |
| "title": "Japanese Phrase Structure Grammar", | |
| "authors": [ | |
| { | |
| "first": "T", | |
| "middle": [], | |
| "last": "Gunji", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1986, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Gunji, T. (1986) 'Japanese Phrase Structure Grammar', Reidel, Dordrecht,1986.", | |
| "links": null | |
| }, | |
| "BIBREF5": { | |
| "ref_id": "b5", | |
| "title": "Conditioned Unification for Natural Language Processing", | |
| "authors": [ | |
| { | |
| "first": "", | |
| "middle": [ | |
| "K" | |
| ], | |
| "last": "Hasida", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1986, | |
| "venue": "Proceedings of th c 11th COLING", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Hasida. K. ( 1986) 'Conditioned Unification for Natural Language Processing, \u2022 Proceedings of th c 11th COLING.", | |
| "links": null | |
| }, | |
| "BIBREF6": { | |
| "ref_id": "b6", | |
| "title": "\u2022Dependency Propagation: A Unified Theory of Sentence Com prehension and Generation", | |
| "authors": [ | |
| { | |
| "first": "K", | |
| "middle": [], | |
| "last": "Hasida", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "", | |
| "middle": [ | |
| "S" | |
| ], | |
| "last": "Ishizaki", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1987, | |
| "venue": "Proceedings of the 10th I.JC AI", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "664--670", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Hasida, K. and Ishizaki. S. (1987) \u2022Dependency Propagation: A Unified Theory of Sentence Com prehension and Generation.' Proceedings of the 10th I.JC AI. pp. 664-670.", | |
| "links": null | |
| }, | |
| "BIBREF7": { | |
| "ref_id": "b7", | |
| "title": "Sentence Processing as Con straint Transformation: Proceedings of ECAJ'.90", | |
| "authors": [ | |
| { | |
| "first": "", | |
| "middle": [ | |
| "K" | |
| ], | |
| "last": "Hasida", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1990, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Hasida. K. ( 1990) \u2022 Sentence Processing as Con straint Transformation: Proceedings of ECAJ'.90.", | |
| "links": null | |
| }, | |
| "BIBREF8": { | |
| "ref_id": "b8", | |
| "title": "From Unification to Constraints", | |
| "authors": [ | |
| { | |
| "first": "J", | |
| "middle": [], | |
| "last": "Jaffar", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "J", | |
| "middle": [], | |
| "last": "Lassez", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1988, | |
| "venue": "Lecture Notes in Computer Science", | |
| "volume": "", | |
| "issue": "5", | |
| "pages": "1--18", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Jaffar, J. and Lassez, J. (1988 ) 'From Unification to Constraints.' Logic Progra mming 0 87, Lecture Notes in Computer Science, No. : 3 1. 5, pp. 1-18.", | |
| "links": null | |
| }, | |
| "BIBREF9": { | |
| "ref_id": "b9", | |
| "title": "198: 3) 'Parsing as Deduction", | |
| "authors": [ | |
| { | |
| "first": "F", | |
| "middle": [ | |
| "C N" | |
| ], | |
| "last": "Pereira", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "D", | |
| "middle": [ | |
| "H D" | |
| ], | |
| "last": "Warren", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": null, | |
| "venue": "Proceedings of A CL '88", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "1--37", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Pereira, F. C. N . and Warren, D.H. D. (198: 3) 'Parsing as Deduction,\u2022 Proceedings of A CL '88, pp. 1 37-144.", | |
| "links": null | |
| }, | |
| "BIBREF10": { | |
| "ref_id": "b10", | |
| "title": "Information Based Syntax and Semantics", | |
| "authors": [ | |
| { | |
| "first": "C", | |
| "middle": [], | |
| "last": "Pollard", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "I", | |
| "middle": [ | |
| "A" | |
| ], | |
| "last": "Sag", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1987, | |
| "venue": "CSLI Lec ture Notes", | |
| "volume": "", | |
| "issue": "1", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Pollard, C. and Sag, I. A. (1987) Information Based Syntax and Semantics. Vo lume 1, CSLI Lec ture Notes No. 13.", | |
| "links": null | |
| }, | |
| "BIBREF11": { | |
| "ref_id": "b11", | |
| "title": "A Uniform Architecture for Parsing and Generation", | |
| "authors": [ | |
| { | |
| "first": "S", | |
| "middle": [ | |
| "M" | |
| ], | |
| "last": "Shieber", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1988, | |
| "venue": "Proceedings of the 12th COLING", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "61--619", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Shieber, S.M. (1988) \u2022A Uniform Architecture for Parsing and Generation.' Proceedings of the 12th COLING. pp. 61-t-619.", | |
| "links": null | |
| }, | |
| "BIBREF13": { | |
| "ref_id": "b13", | |
| "title": "\\-fa nual. \ufffdlemo 2 . 59. -AI Labo ratory", | |
| "authors": [ | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "Siver Refere Nn", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": null, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "SIVER Refere nN .\\-fa nual. \ufffdlemo 2 . 59. -AI Labo ratory. MIT.", | |
| "links": null | |
| }, | |
| "BIBREF14": { | |
| "ref_id": "b14", | |
| "title": "\u2022 Con straints -A Language for Expressing Almost Hierarchical Descriptions ,\u2022 Artificial Intellige ncf", | |
| "authors": [ | |
| { | |
| "first": "", | |
| "middle": [ | |
| "G" | |
| ], | |
| "last": "Sussman", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "", | |
| "middle": [ | |
| "G" | |
| ], | |
| "last": "Steele", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1980, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Sussman. G. and Steele. G., .Jr. ( 1980) \u2022 Con straints -A Language for Expressing Almost Hierarchical Descriptions ,\u2022 Artificial Intellige ncf:. Vo l. 14.", | |
| "links": null | |
| }, | |
| "BIBREF15": { | |
| "ref_id": "b15", | |
| "title": "Pmcudings of the Seco nd lntun atio nal Co nfert na on Logic Progra mming", | |
| "authors": [ | |
| { | |
| "first": "", | |
| "middle": [ | |
| "H" | |
| ], | |
| "last": "Tamaki", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "", | |
| "middle": [ | |
| "T" | |
| ], | |
| "last": "Sato", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": null, | |
| "venue": "", | |
| "volume": "198", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Tamaki. H. and Sato. T. ( 198: 3) T nfold/Folcl Transformation of Logic Programs.' Pmcudings of the Seco nd lntun atio nal Co nfert na on Logic Progra mming. pp. 127-1:38.", | |
| "links": null | |
| }, | |
| "BIBREF16": { | |
| "ref_id": "b16", | |
| "title": "Proceeding.s of thE Th ird International Conff l'rn ct on Log-ic Programming", | |
| "authors": [ | |
| { | |
| "first": "", | |
| "middle": [ | |
| "H" | |
| ], | |
| "last": "Tamaki", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "", | |
| "middle": [ | |
| "T" | |
| ], | |
| "last": "Sato", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1984, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "84--98", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Tamaki. H. and Sato . T. (1984) \u2022OLD Resolu tion with Tabulation.' Proceeding.s of thE Th ird International Conff l'rn ct on Log-ic Programming. pp. 84-98.", | |
| "links": null | |
| }, | |
| "BIBREF17": { | |
| "ref_id": "b17", | |
| "title": "Parsing as Constraint Transformation -an extension of cu-Prolog' Proceedings of the Seo\u2022 ul ln ff rna tional Co nf ere nee on Natural Languagt Proct.s.s ing", | |
| "authors": [ | |
| { | |
| "first": "", | |
| "middle": [ | |
| "H" | |
| ], | |
| "last": "Tsuda", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "", | |
| "middle": [ | |
| "K" | |
| ], | |
| "last": "Hasida", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1990, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "325--331", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Tsuda. H. and Hasida. K. (1990 ) \u2022Parsing as Constraint Transformation -an extension of cu-Prolog' Proceedings of the Seo\u2022 ul ln ff rna tional Co nf ere nee on Natural Languagt Proct.s.s ing, pp. 325-331.", | |
| "links": null | |
| }, | |
| "BIBREF18": { | |
| "ref_id": "b18", | |
| "title": "Proceedings of the European Ch apter of A CL \u2022 89", | |
| "authors": [ | |
| { | |
| "first": "H", | |
| "middle": [], | |
| "last": "Tsuda", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "K", | |
| "middle": [], | |
| "last": "Hasida", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "H", | |
| "middle": [], | |
| "last": "Sirai", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1989, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "9--102", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Tsuda, H., Hasida, K .. and Sirai, H. (1989) \u2022JPSG Parser on Constraint Logic Programming.' Proceedings of the European Ch apter of A CL \u2022 89. PP \u2022 9 . s-102.", | |
| "links": null | |
| }, | |
| "BIBREF19": { | |
| "ref_id": "b19", | |
| "title": "\u2022 ( 1990 ) \u2022cu-Prolog V2 system", | |
| "authors": [ | |
| { | |
| "first": "", | |
| "middle": [ | |
| "H" | |
| ], | |
| "last": "Tsuda", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "K", | |
| "middle": [], | |
| "last": "Hasida", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "H", | |
| "middle": [], | |
| "last": "Yasukawa", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "", | |
| "middle": [ | |
| "H" | |
| ], | |
| "last": "Sirai", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": null, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Tsuda. H., Hasida, K.\ufffd Yasukawa,H. and Sirai . H. \u2022 ( 1990 ) \u2022cu-Prolog V2 system', !CO T TAI-9-52.", | |
| "links": null | |
| } | |
| }, | |
| "ref_entries": { | |
| "FIGREF0": { | |
| "uris": null, | |
| "type_str": "figure", | |
| "text": ":-member(X, [a, b,c]). (5) : -member(X ,Y) ,Y=[a,b , c] .", | |
| "num": null | |
| }, | |
| "FIGREF1": { | |
| "uris": null, | |
| "type_str": "figure", | |
| "text": "1 , B 2 , \u2022 \u2022 \u2022, B n ; C1 , C2 , \u2022\u2022\u2022,C m .", | |
| "num": null | |
| }, | |
| "FIGREF2": { | |
| "uris": null, | |
| "type_str": "figure", | |
| "text": "lexicon(hasi , [ ... sem (TYPE ,OBJ)]); has i_sem (TYPE ,OBJ) .", | |
| "num": null | |
| }, | |
| "FIGREF3": { | |
| "uris": null, | |
| "type_str": "figure", | |
| "text": "0_415, Adj _677 , SEM_681) True . CPU time = 0.050 sec _:-c7 (F,SC,_,A,SEM) . F = syusi SC = [cat(p, wo , [] , [] , [] , 0bj00_30)] A = [] SEM = [love,ken , 0bj 00_30] ; F = rentai SC = [] A = [cat(n, n, [] , [] , [] , inst (0bj00_38 , Type3_36))] SEM = inst (0bj 00_38 , [and , Type3_36 ,[love ,ken, 0bj 00_38] ]) no . CPU time = 0.017 sec This is an example ru n of JPSG parser in cu-Prolog. The fi rst line is a user's input. \"Ken ga ai suru\" has two readings: \"Ken loves (someone)\" and \"(someone) whom Ken loves.\" The parser draws a parse tree and returns information (constraint) on the st_ ructure of the top node. In this example, the ambiguity of the sentence is captured as the two solutions of the piece\u2022 of constraint c7 (F ,SC,_,A,SEM) . The fi rst solution corresponds to \"Ken loves (someone).\" and the second solution \"(someone) whom Ken loves.\" Parsing an ambiguous sentence.", | |
| "num": null | |
| }, | |
| "FIGREF4": { | |
| "uris": null, | |
| "type_str": "figure", | |
| "text": "( 1:3) p ---t a p -ppThe parsing of string aa \u2022 \u2022 \u2022 a under this grammar may be formulated in terms of the fo llowing constraint. 6( 1 4) :p(A 0 ,B) , A 0 =[alA 1 ], \u2022 \u2022\u2022, A n -l =[a] .p([alX] ,X) . p(X,Z) :p(X,Y) , p(Y,Z) .Note that the double occurrence of Y in the last clause does not count as a dependency, because the sec ond argument place of p is vacuous. Thus the only de pendency to eliminate now is that concerning AO \u2022 Here. we replace p(A 0 ,B) with p 0 (A 0 ), creating a new predi cate p 0 \u2022 ( 1 5 ) P o ( B) , A O = [ a I A 1 ] , \u2022 \u2022 \u2022 , A n -1 = [ a I A n ] . P o (A 1 ). po (Z) :-p(A 0 ,Y) , p(Y,Z) . p (A O , Y) in the last clause is folded and we get (16) po (Z) :-p 0 (Y) ,p(Y ,Z) .", | |
| "num": null | |
| }, | |
| "FIGREF5": { | |
| "uris": null, | |
| "type_str": "figure", | |
| "text": ":-po (B) , A 0 = [b lA 1 ], \u2022\u2022\u2022 po (Z) :-po (Y) , p(Y ,Z) .", | |
| "num": null | |
| }, | |
| "FIGREF6": { | |
| "uris": null, | |
| "type_str": "figure", | |
| "text": "Suppose for ins_ tance , that a is to penetrate through p ( \u2022 , \u2022 ) in '11 1 at the bottom stage in Figure 2. If we applied fo lding here. simply r\ufffdplacing this p ( \u2022, \u2022 ) with a q ( \u2022, \u2022 ) , the resulting configuration would lose the combination of <P3 and W 1-Penetration.", | |
| "num": null | |
| }, | |
| "FIGREF7": { | |
| "uris": null, | |
| "type_str": "figure", | |
| "text": "dependency encompassing argument places with greater information quantity should more readily trigger a penetration. b. The argument position with greater information quantity should be penetrated here.", | |
| "num": null | |
| }, | |
| "FIGREF8": { | |
| "uris": null, | |
| "type_str": "figure", | |
| "text": "28 ) : -Po (B) , A 0 =[a l A 1 ] , \u2022 : \u2022 , A n -l=[a l A n ] . Po (A 1 ). Po ( Z ) : -p (A O , Y) , p ( Y , Z) .The only relevant dependency here is the one concern ing the first argument of p(A 0 , Y) in the bottom clause. This literal is hence folded and replaced with p0 (Y) , the entire clause being transformed as follows.(29) po (Z) : -po (Y) , p (Y , Z) .", | |
| "num": null | |
| }, | |
| "FIGREF9": { | |
| "uris": null, | |
| "type_str": "figure", | |
| "text": ") : -P2 (Y) , p(Y,Z) . Shown below is what is finally obtained. (: 3, 5) :-q, A 0 =[a lA 1 ], \u2022 \u2022 \u2022, A n -l=[alA n ]. q : -po (B 0 ). q : -Po.i , B 0 =A i . (0 < i ::; n) Pi (Z) : -Pi.J , p/Z) . (0 ::; i < j < n) Pz (Z ) : -Pi (Y) , p(Y,Z ) . (0 ::;i<n) Pi.i+1 . (0 ::; i < n) Pi.k : -Pi.J , PJ.k \u2022 (0 ::; i < j < k < n) Figure : J : The meaning of Pi.k : -Pi.j , Pi.k .", | |
| "num": null | |
| }, | |
| "FIGREF10": { | |
| "uris": null, | |
| "type_str": "figure", | |
| "text": "38) s (X , Z ) (39) s(X,Z) np(X ,Y) , vp (Y,Z) .np (X ,Y) , adv (Y,U) , vp (Y,Z) .", | |
| "num": null | |
| }, | |
| "FIGREF11": { | |
| "uris": null, | |
| "type_str": "figure", | |
| "text": "VP --+ V NP VP --+ VP PP NP --+ NP NP V --+ see NP --+ a man PP --+ with a telescope ( 41) is a parsing program in terms of this grammar.", | |
| "num": null | |
| }, | |
| "FIGREF12": { | |
| "uris": null, | |
| "type_str": "figure", | |
| "text": "41) p(X,Z,C) : -p(X,Y,LC) ,p(Y,Z ,RC) , . c(LC ,RC,C) .", | |
| "num": null | |
| }, | |
| "FIGREF13": { | |
| "uris": null, | |
| "type_str": "figure", | |
| "text": "42) :-p(A 0 ,B,C) , A 0 =[seelA 1 ], A 1 =[a , man lA 2 ], [with ,a,tel escope lA 3 ] ,A 3 =[ ]. ( 42) is a question clause. This example shows that two meanings of \u2022'I see a man with a telescope\u2022' are derived from this program by the constraint transfor mation with the heuristic mentioned previously. The dependency to be processed is in terms of A 0 in ( 42) because LC and RC in ( 40) do not have depen dencies on ,ac\ufffdount qf vacuous argument places. Then. . apply dow\ufffdward penetration in terms of A O to ( 42). p0 (B,C) is equivalent to p(A 0 ,B,C) . (43) :-po (B,C) , ( 4-l ) p 0 (A 1 ; v) . (45) p 0 (B,C) :-p(A 0 ,Y,LC) ,p(Y ,B,RC) , , c(LC ,RC,C) . The first body literal of ( 4, 5) can be folded and we get (46) p 0 (B,Cat) :-p0 (Y,LC) ,p(Y ,B,RC) , c(LC ,RC,Cat) .", | |
| "num": null | |
| }, | |
| "FIGREF14": { | |
| "uris": null, | |
| "type_str": "figure", | |
| "text": ") : -po,1 . ( 48) Po,1 . (49 ) p0 (B ,Cat) :-po, 1 ,p(A 1 ,B,RC) ,c(v ,RC,Cat) .", | |
| "num": null | |
| }, | |
| "FIGREF15": { | |
| "uris": null, | |
| "type_str": "figure", | |
| "text": "Let c 0 (Cat) be c ( v, RC , Cat ) and you apply downward pen etration to (49), obtaining po(B,Cat ) :-Po,", | |
| "num": null | |
| }, | |
| "FIGREF16": { | |
| "uris": null, | |
| "type_str": "figure", | |
| "text": "54 ) p1 (Z) : -p( A 1 ,Y,np) ,p(Y,Z,pp) .The first body of ( . j 4) can be folded and we get( 55 ) p1 (Z) : -p1 (Y) ,p(Y,Z,pp ) . Upward penetration in ( . 5 2). p1 ,2 = p1 (A 2 ) ( -S 6) Po (A 2 , vp) : -po,1 , P1.2 . (. 57) P1 (Z) : -PI.2 ,p1 (A 2 ,Z,pp) . ( . 5 8) Pu\u2022 l:pward penetration m (-S6). p0,2 is equivalent to po (A 2 ,vp) . (-59) Po (B , Cat) : -po,2, p(A 2 , B ,RC) , c (vp ,RC, Cat)_\u2022 (60) Po.2 : -po .1 , P1.2 \u2022 Cnfold the category rnllstrai nt of (. S9 ). (61) po (B ,vp) :-p 0_2 , p (A 2 ,B,pp) . Here. the remaining clauses are (43). (-17 ). (-!8). (-SS). (60 ) . ( -!6). (. Sl). ( . S. 3). (. Si) and (61). Apply down ward penetration of A 2 in (61 ). p 2 (B) is equivalent to p(A 2 ,B,pp) . (62) p o (B ,vp) :-po.2 ,p 2 (B) . (6: 3) p 2 (A 3 ) . (64 ) p 2 (B) :-p(A 2 ,Y,LC) ,p(Y,B ,RC) , c(LC ,RC,pp) . Unfolding of the category constraint of ( 64: ) fails. Fold (. 5 7 ). ( 6 . 5 ) p1 (Z) :-p1.2 , p2 (Z) . lT pward penetration in (63). p 2 , 3 1s equivalent to P2 (A 3 ) . (66) po (A 3 ,vp) :-po\ufffd , P2\ufffd (67 ) P2.3 . (68) P1 (A 3 ) :-PI,2 , P2.3 \u2022 Upward penetration in (66). Po.3 = po (A 3 ,vp) .(69) p0 (B,Cat) :-p0,3 ,p(A 3 ,B,RC) ,c(vp ,RC,Cat) . ( 70 ) Po,3 . Unfolding of the category constraint in ( 69 ) fails. l' p w ard penetration in ( 70 ). p 1 ,3 = P i (A 3 ). ( 7 1) po (A 3 , vp) :-po,1 ,p1,3 \u2022 ( 72) Pi (Z) : -pl,3 ,p(A 3 ,z ,pp) . ( 7 3) p1,3 : -pu ,-P2.3 \u2022 ( 66) and (7 1 ) represent the two readings of \u2022\u2022see a man with a telescope.\u2022\u2022", | |
| "num": null | |
| }, | |
| "TABREF0": { | |
| "html": null, | |
| "content": "<table><tr><td>In [18] , we introduced a symbolic CLP language cu Prolog and showed how it applies to parsing based on JPSG (Japanese Phrase Structure Grammar ) [. j]. By treating grammatical principles and ambiguity con cerning polysemy or homonymy straightforwardly in terms of constraints. syntactic, semantic and other types of ambiguity are processed in an integrated man ner by Constraint Un ification ( CU). CU is the unifier employed in cu-Prolog, and is roughly regarded as the standard unification plus DP. cu-Prolog deals with var ious constraints on the structures of grammatical cat egories . without any special programming besides the encoding of the relevant constraints.</td></tr></table>", | |
| "type_str": "table", | |
| "text": "", | |
| "num": null | |
| }, | |
| "TABREF3": { | |
| "html": null, | |
| "content": "<table/>", | |
| "type_str": "table", | |
| "text": "In the current formulation. the computational com plexity for processing context-free languages is expo nential as to the sentence length. vV ith respect to the above example. suppose that predicate r 1 is s\ufffdch that for any assignment to variable Xi , there is a set of assignments to variables x 0 through x i -t under which ri (X i ) is equivalent to the following: P o may be regarded as r 0 . As it turns out. if a definition clause of ri is", | |
| "num": null | |
| }, | |
| "TABREF4": { | |
| "html": null, | |
| "content": "<table><tr><td>(20) P -a</td><td>Q ---ta</td></tr><tr><td>p -pp p -PQ</td><td>Q -pp Q -PQ</td></tr><tr><td colspan=\"2\">3.2 Penetration</td></tr><tr><td/><td>which coil \\\u2022eys</td></tr><tr><td colspan=\"2\">information across clause boundaries.</td></tr><tr><td colspan=\"2\">For instance. consider clause (21 ). where pwclicate p</td></tr><tr><td>is defined by ( 22).</td><td/></tr></table>", | |
| "type_str": "table", | |
| "text": "with j = 0, it follows from foduction on i that ri is created during the current parsing for 0 < i < n. A similar reasoning will prove that ex ponentially many corresponding predicates are created when the basic version of DP as described so far is ap plied to the following context-free grammar. because there are plural predicate symbols._", | |
| "num": null | |
| } | |
| } | |
| } | |
| } |