| { |
| "paper_id": "W98-0122", |
| "header": { |
| "generated_with": "S2ORC 1.0.0", |
| "date_generated": "2023-01-19T06:06:10.735622Z" |
| }, |
| "title": "An improved Earley parser with LTAG", |
| "authors": [ |
| { |
| "first": "Yannick", |
| "middle": [], |
| "last": "De Kercadio", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "LIMSI-CNRS", |
| "location": { |
| "postBox": "BP 133", |
| "postCode": "F-91403", |
| "settlement": "Orsay cedex", |
| "country": "FRANCE" |
| } |
| }, |
| "email": "kercadio@talana.linguist.jussieu.fr" |
| } |
| ], |
| "year": "", |
| "venue": null, |
| "identifiers": {}, |
| "abstract": "", |
| "pdf_parse": { |
| "paper_id": "W98-0122", |
| "_pdf_hash": "", |
| "abstract": [], |
| "body_text": [ |
| { |
| "text": "This paper presents an adaptation of the Earley algorithm (EARLEY, 1968) for parsing with lexicalized tree-adjoining grammars (LTAGs).", |
| "cite_spans": [ |
| { |
| "start": 58, |
| "end": 72, |
| "text": "(EARLEY, 1968)", |
| "ref_id": "BIBREF2" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Introduction", |
| "sec_num": "1" |
| }, |
| { |
| "text": "This algorithm constructs the derivation tree following a top-down strategy and verifies the valid prefix property. Many earlier algorithm do not have both of this properties (ScHABES, 1994) . The Earley-like algorithm described in (SCHABES and Josm, 1988) verifies the valid prefix property, but the algorithm presented here is thought to be easier to improve using some properties of LTAGs.", |
| "cite_spans": [ |
| { |
| "start": 175, |
| "end": 190, |
| "text": "(ScHABES, 1994)", |
| "ref_id": "BIBREF4" |
| }, |
| { |
| "start": 232, |
| "end": 256, |
| "text": "(SCHABES and Josm, 1988)", |
| "ref_id": null |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Introduction", |
| "sec_num": "1" |
| }, |
| { |
| "text": "A LTAG is a context-free grammar (CFG) on trees, the elementary operations of which ate the adjunction and the substitution. The Earley algorithm can be used for parsing with any CFG insofar as the elementary operation is the concatenation. Hence, the Earley algorithm cannot sir:nply be used for LTAGs, but the meaning of an edge in the derivation tree needs to be specified in t erms of words strings and concatenations. Substitution and terminal nodes can be handled using ordinary context-free rules. Such a rule represents a node in the derivation tree and An adjunction can be seen as two correlat('d substitutions: the derived string of t he part of the adjoined tree on the left. of t he foot nodr> j,.. inserted in some location while the ot lwr pa r1 of the string is inserted in some ot her loea t io11 farther in the string. The string located bei \\n\u2022i>11 the two substitution points is the d erivecl stri1t!!_ of the subtree under the adjoi11ed node. Tl11\u2022 correlation between these two su bst i tu tio11\" j,.. that either none or both of them shoulcl ocT11 r. thus a synchronization must be transmitted 11 p to the second location in order to preserYe t lt j,.. constraint.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Representation of a LTAG with a set of rules", |
| "sec_num": "2" |
| }, |
| { |
| "text": "The locations of these pairs of places follm\\\u2022,.. a stack order: there is an equal ntunber oJ' \"first places\" and \"second places'' between t\\\\\u2022n matching places. Therefore, a unique s~\u2022mb11I (# hereafter) can be used to represent an~\u2022 \u2022\u2022,..rr.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Representation of a LTAG with a set of rules", |
| "sec_num": "2" |
| }, |
| { |
| "text": "ond place\", while a \u00df.1Y notation can be llS<'d 11 1 represent a \"first place\" for an adjunction of\" tree with root X.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Representation of a LTAG with a set of rules", |
| "sec_num": "2" |
| }, |
| { |
| "text": "The figure 1 shows a few rules represe nti11!!, some elementary trees. A star denotes a foul node in an auxiliary tree. T he <l ra \\\\' ll link:< i111 \u2022 plements the correlation information bet\\\\'PPt1 the two substitution points representiiig an ad\u2022 junction. Because of the stack strnrturP or l lii.information, the links need not to be PXplki1 I~ st.ored . Also note tha.t t.hese trees ;irP fh11 ! 1111 VP). See (ABEILLE, 1991) . This is no t rna 11d :1\u2022 tory and the trees usually used for Cnglisli 1\u2022;111 be encoded the same way.", |
| "cite_spans": [ |
| { |
| "start": 413, |
| "end": 428, |
| "text": "(ABEILLE, 1991)", |
| "ref_id": null |
| } |
| ], |
| "ref_spans": [ |
| { |
| "start": 4, |
| "end": 12, |
| "text": "figure 1", |
| "ref_id": null |
| } |
| ], |
| "eq_spans": [], |
| "section": "Representation of a LTAG with a set of rules", |
| "sec_num": "2" |
| }, |
| { |
| "text": "As each node in the derivation trPP rPprPsP11lan elementary tree, and alj en~r~\u2022 ek\u2022111P11 I ;1 r_ \\ Rule for the transitive verb to love (anOVnl), without adjunctions:", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Representation of a LTAG with a set of rules", |
| "sec_num": "2" |
| }, |
| { |
| "text": "Rule for the transitive verb to love (anOVnl), with possible a.djunctions on Sand on F:", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "aS -+ aN love aN", |
| "sec_num": null |
| }, |
| { |
| "text": "aS -+ \u00dfS aN \u00dfV love # oiN #", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "aS -+ aN love aN", |
| "sec_num": null |
| }, |
| { |
| "text": "Rule for the determiner the (\u00dfDetN), with a possible adjunction on the root N: \u00dfN -+ \u00dfN the * # Figure 1 : Examples of rules tree can be represented by a rule which capture the. linear ward order of the derived string, this is a way to capture the linear ward order in the derivation tree. The usual derivation tree (as defined in (VIJAY-SHANKER, 1987)) can be obtained by linking the subtree of every Ufirst place\" to the left of the subtree of the matching \"second place\" and by storing the resulting structure under the \"second place\".", |
| "cite_spans": [], |
| "ref_spans": [ |
| { |
| "start": 96, |
| "end": 104, |
| "text": "Figure 1", |
| "ref_id": null |
| } |
| ], |
| "eq_spans": [], |
| "section": "aS -+ aN love aN", |
| "sec_num": null |
| }, |
| { |
| "text": "In this section, we show how the stacked relationships between the \"first places\" and \"second places\" can be represented in a structure which is suitable for the Earley algorithm. Following Earley, a partial parsing can be represented by an item, which consists in a rule, a.\u2022 position in the rule (all the symbols loca.ted on\u2022 its left have been recognized), and two lists of pairs of references to items. The first list keeps track of the requesters of the rule, that is to sa.y the items which are waiting for the rule to be recognized in order to be shifted. The second element of each pair is used as a relay storage during the recognition of the second part of an auxiliary tree. The second list implements the previously mentioned stack of \"first places\". The first elernent of each pair it contains is the data part of the stack item. lt is a reference to an item wa.iting on a foot symbol. The second element in each pair is used to implement the stack. lt is a reference to an item waiting for an adjunction.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Earley-like parsing driven by the derivation tree", |
| "sec_num": "3" |
| }, |
| { |
| "text": "A number of primitive operations will be ap-plied on this data structure. They are summed up in the table 2. When a primitive is applied on a given set, the second column indicates ho\\\\' many actions are to be taken. The rule and mark columns indicate which item is t.o be introduced. If no item with this rule and thi~ position mark is present in the set. it is introduced with the indicated lists for the requester~ list and the stack list. Otherwise, the in<licated lists are merged with the ones of the existing item in the set. This merging step ensures that the spa.tial complexity ha.s a. polynomial upper bound.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Earley-like parsing driven by the derivation tree", |
| "sec_num": "3" |
| }, |
| { |
| "text": "The algorithm consists in working on each set in turn, following the word order. The initial set is initialized using init. Then an evolution stage applies a predict or reduce primitive on every newly introduced item, the type of whkh is chosen from the symbol in the rule \\\\\u2022hich b right after the ma.rk. For instance. if it is an aX (a substitution is expected), then predict a(item, X) is used. If there is no such symbol. them a. reduce primitive is used, depending on the type (a or \u00df) of the left part of the rule.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Earley-like parsing driven by the derivation tree", |
| "sec_num": "3" |
| }, |
| { |
| "text": "\u2022 . This process is then run on each set in turn.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Earley-like parsing driven by the derivation tree", |
| "sec_num": "3" |
| }, |
| { |
| "text": "replacing inits with a shift 011 every item expecting (i.e. with the mark right on the left of) the word associated with the current set. .The sentence is accepted if there is an item in the last set with a ru!e deriving the axiom {S). with the mark at the end of the rnle. with nn empty requesters list. lt should be noted thnt this algorithm does not give an anal~\u2022sis of t hfl sentence. An additional structure is requirf'd in each item to keep the analysis information. Figure 2 : Primitives of the algorithm However, every edge in the derivation tree is detected through the fact that a reduce primitive is run. This additional structure should cope with the ambiguities and permit a. polynomial representa.tion of ambiguities from other level of. analysis (features unification, semantic analysis and so on). This is a. quite general matter: the number of solutions to the problem of parsing being (potentially) exponential, a. simple list of analyses would require a.n exponential time to be output. The usual assumption that the number of analyses is \"small\" is not acceptable in the context of pa.rsing oral utterances' (because of potential a.uto-repairing constructs}. 'therefore, the repreSenta.tion of the outputs .should grow polynomia.lly (and not exponentially) with the number of ambiguities.", |
| "cite_spans": [], |
| "ref_spans": [ |
| { |
| "start": 474, |
| "end": 482, |
| "text": "Figure 2", |
| "ref_id": null |
| } |
| ], |
| "eq_spans": [], |
| "section": "Earley-like parsing driven by the derivation tree", |
| "sec_num": "3" |
| }, |
| { |
| "text": "The toirdown strategy of this algorithm has a trivial, but very useful property: this algorithm do not require the uttera.nce to be cut into sentences in order to parse it. Instead, one can perform an init primitive in every set where a rule with the a.xiom as its left part a.nd an empty requesters list is found. lt has the effect of concurrently trying to parse a new sentence from this point. This property is very important when parsing oral utterances: there is no practical other way to find out where sentences begin and end. Moreover, the combina.tion of both the top-down strategy and the valid prefix property enables valuable performance improvements . Many of the LTAGs properties (SRIN!VAS. 1997) can be used to avoid the introduction of unreleva.nt elementary trees, thus allowing t.he use of a. richer gramma.r. The data structures construct a derivation tree. Therefore, a. rough sema.ntic analysis can be performed to check whether some newly discovered potential edge in the deriva.tion tm\u2022 makes sense or not. If not, it can be invalidated \u2022 as soon as it is discovered.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Benefits in using this strategy", |
| "sec_num": "4" |
| }, |
| { |
| "text": "When fea.tures are used, they can be checked following only the derivation tree (t.he derived tree is not needed) . This is due to the fa('t tha.t the nodes in the derivation tree a re mo rf' than simple atoms: they are the rules that havl' been used for parsing. Like with semantk an11lysis, the features unification can be clone 011 partial a.na.lysis, after every red uction. However, it is not clear whether this would result in \u2022 .a.n improvement or not: the cost of the 11nific11tion might overcome the benefits of invalidatin~ some partial analysis a.s soon as possible. Due to the lexicaliza.tion, terminals ( wor<I~) are put in the trees during lexicon access. \\ \\'hen a rule is invocated in a set S. it always contai11\" at least one terminal (lexicalization). All thl' symbols on the left of the first terminal h11ve tu be recognized before the set where this terntinal is to be found . This is a wa~\u2022 to filter t h~ candida.te rules for recognizing these s\u00bbml>ob.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Benefits in using this strategy", |
| "sec_num": "4" |
| }, |
| { |
| "text": "Former parsers atieady used the span of trees to eliminate trees that are too !arge to parse the sentence (XTAG, for instance), but this algorithm permits considering the span properties locally, at every prediction stage.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Benefits in using this strategy", |
| "sec_num": "4" |
| }, |
| { |
| "text": "Last but not least, the data structures used for this algorithm can be enriched in successive analysis stages. That is to say, when no analysis is found, it is possible to enrich the sets with new rules. This property is useful to construct a fault tolerant parser, accepting unknown words, using weighted syntadic rules (the weights indicating whether a given rule is linguistically perfect or somewhat deviant), and accounting for auto-repairing sequences in an oral utterance.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Benefits in using this strategy", |
| "sec_num": "4" |
| }, |
| { |
| "text": "Using these properties enables the design of an efficient oral-specific robust parser using a grammar of the written language (ABEILLE, 1991) . We plan to incorporate a syntactic LTAG-based component in a working real-time speech understanding system (GAUVAIN et al., 1997,) to improve its recognition performances.", |
| "cite_spans": [ |
| { |
| "start": 126, |
| "end": 141, |
| "text": "(ABEILLE, 1991)", |
| "ref_id": null |
| }, |
| { |
| "start": 251, |
| "end": 274, |
| "text": "(GAUVAIN et al., 1997,)", |
| "ref_id": null |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Prospects", |
| "sec_num": "5" |
| } |
| ], |
| "back_matter": [], |
| "bib_entries": { |
| "BIBREF0": { |
| "ref_id": "b0", |
| "title": "Une grammaire lexiCCJlisee d'Arbres Adjoints pour le frani", |
| "authors": [ |
| { |
| "first": "A", |
| "middle": [], |
| "last": "Abeill~", |
| "suffix": "" |
| } |
| ], |
| "year": 1991, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "ABEILL~, A. 1991. Une grammaire lexiCCJlisee d'Arbres Adjoints pour le frani;ais. Ph.D. thesisJ Universite Paris 7.", |
| "links": null |
| }, |
| "BIBREF1": { |
| "ref_id": "b1", |
| "title": "A lexicalized tree adjoining grammar for english", |
| "authors": [ |
| { |
| "first": "A", |
| "middle": [], |
| "last": "Abeille", |
| "suffix": "" |
| }, |
| { |
| "first": "K", |
| "middle": [], |
| "last": "Bishop", |
| "suffix": "" |
| }, |
| { |
| "first": ".", |
| "middle": [], |
| "last": "Cote", |
| "suffix": "" |
| }, |
| { |
| "first": "Y", |
| "middle": [], |
| "last": "Sch-Abes", |
| "suffix": "" |
| } |
| ], |
| "year": 1990, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "ABEILLE, A., K. BISHOP, s. COTE, and Y. ScH- ABES. 1990. A lexicalized tree adjoining gram- mar for english. Technical report, Department of Computer and Information Science, University of Pennsylvania, Philadelphia.", |
| "links": null |
| }, |
| "BIBREF2": { |
| "ref_id": "b2", |
| "title": "An efficient context-free pars\u2022 . ing algorithm", |
| "authors": [ |
| { |
| "first": "J", |
| "middle": [ |
| "C" |
| ], |
| "last": "Earley", |
| "suffix": "" |
| } |
| ], |
| "year": 1968, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "EARLEY, J. C. 1968. An efficient context-free pars\u2022 . ing algorithm. Ph.D. thesis, Carnegie-Mellon Uni- versity, Pittsburgh.", |
| "links": null |
| }, |
| "BIBREF3": { |
| "ref_id": "b3", |
| "title": "Spaken language component of the mask kiosk", |
| "authors": [ |
| { |
| "first": "J.-L", |
| "middle": [], |
| "last": "Gauyajn", |
| "suffix": "" |
| }, |
| { |
| "first": "S", |
| "middle": [], |
| "last": "Bennacef", |
| "suffix": "" |
| }, |
| { |
| "first": "L", |
| "middle": [], |
| "last": "Devillers", |
| "suffix": "" |
| }, |
| { |
| "first": "L", |
| "middle": [], |
| "last": "Lamel", |
| "suffix": "" |
| }, |
| { |
| "first": "S", |
| "middle": [ |
| "Rosset" |
| ], |
| "last": "", |
| "suffix": "" |
| } |
| ], |
| "year": 1997, |
| "venue": "Human Comfort and Security of InfoNT1ation Systems", |
| "volume": "", |
| "issue": "", |
| "pages": "93--103", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "GAUYAJN, J.-L., S. BENNACEF, L. DEVILLERS, L. LAMEL, and S. ROSSET. 1997. Spaken lan- guage component of the mask kiosk. In K. Vargh- ese and S. Pfleger, editors, Human Comfort and Security of InfoNT1ation Systems. Springer-Verlag, pages 93-103.", |
| "links": null |
| }, |
| "BIBREF4": { |
| "ref_id": "b4", |
| "title": "Left-to-right parsing of lexicalized tree-adjoining grammars", |
| "authors": [ |
| { |
| "first": "Y", |
| "middle": [], |
| "last": "Schabes", |
| "suffix": "" |
| } |
| ], |
| "year": 1994, |
| "venue": "Compulalional Linguistics", |
| "volume": "10", |
| "issue": "4", |
| "pages": "506--524", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "ScHABES, Y. 1994. Left-to-right parsing of lexi- calized tree-adjoining grammars. Compulalional Linguistics, 10(4):506-524.", |
| "links": null |
| }, |
| "BIBREF5": { |
| "ref_id": "b5", |
| "title": "An earley-type parsing algorithm of tree adjoining languages", |
| "authors": [ |
| { |
| "first": "Y", |
| "middle": [], |
| "last": "Schabes", |
| "suffix": "" |
| }, |
| { |
| "first": "A", |
| "middle": [ |
| "K" |
| ], |
| "last": "Joshj", |
| "suffix": "" |
| } |
| ], |
| "year": 1988, |
| "venue": "26th Meeting of the Associalion fr)I' C'omp11ffl-lional Linguistics", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "SCHABES, Y. and A. K. JosHJ. 1988. An earley-type parsing algorithm of tree adjoining languages. In 26th Meeting of the Associalion fr)I' C'omp11ffl- lional Linguistics, Buffalo.", |
| "links": null |
| }, |
| "BIBREF6": { |
| "ref_id": "b6", |
| "title": "Complexity of /o\u2022ical dfsc1\u2022i1'-tions and its releuance to partial pnrsing", |
| "authors": [ |
| { |
| "first": "B", |
| "middle": [], |
| "last": "Sr!njvas", |
| "suffix": "" |
| } |
| ], |
| "year": 1997, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "SR!NJVAS, B. 1997. Complexity of /o\u2022ical dfsc1\u2022i1'- tions and its releuance to partial pnrsing. Ph.D. thesis, Department of Computer and Inf\"\"\"lat.ion Science, University of Pennsylvania. Phila<lelphi11.", |
| "links": null |
| }, |
| "BIBREF7": { |
| "ref_id": "b7", |
| "title": "Departmenr of Computer and Information Science. L:ni\\\u2022ersity of Pennsylvania", |
| "authors": [ |
| { |
| "first": "K", |
| "middle": [], |
| "last": "Vijay-Shanker", |
| "suffix": "" |
| } |
| ], |
| "year": 1987, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "VIJAY-SHANKER, K. 1987. A study of l1'ff <trf- joining grammars. Ph.D. thesis. Departmenr of Computer and Information Science. L:ni\\\u2022ersity of Pennsylvania, Philadelphia.", |
| "links": null |
| }, |
| "BIBREF8": { |
| "ref_id": "b8", |
| "title": "Hl85. Som\" computational properties of tree adjoining grammars", |
| "authors": [ |
| { |
| "first": "K", |
| "middle": [], |
| "last": "Vijay-Shanker", |
| "suffix": "" |
| }, |
| { |
| "first": "A", |
| "middle": [ |
| "K" |
| ], |
| "last": "Josm", |
| "suffix": "" |
| } |
| ], |
| "year": null, |
| "venue": "23rd Meeting of tlle Associatio11 for Computational Linguistics", |
| "volume": "", |
| "issue": "", |
| "pages": "82--93", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "VIJAY-SHANKER, K. and A. K. Josm. Hl85. Som\" computational properties of tree adjoining gram- mars. In 23rd Meeting of tlle Associatio11 for Computational Linguistics, pages 82-93. Chicago.", |
| "links": null |
| } |
| }, |
| "ref_entries": { |
| "FIGREF0": { |
| "num": null, |
| "text": "l'd like to thank A. Abeille, M.-H. Candito, F. lssac and P. Paroubek for their valuable help and advices captures the linear word order of the derin\u2022.J string.", |
| "uris": null, |
| "type_str": "figure" |
| } |
| } |
| } |
| } |
