| { | |
| "paper_id": "W98-0127", | |
| "header": { | |
| "generated_with": "S2ORC 1.0.0", | |
| "date_generated": "2023-01-19T06:03:14.227989Z" | |
| }, | |
| "title": "Packing of Feature Structures for Optimizing the HPSG~style Gran1mar translated fro1n TAG", | |
| "authors": [ | |
| { | |
| "first": "Yusuke", | |
| "middle": [], | |
| "last": "Miyaof", | |
| "suffix": "", | |
| "affiliation": {}, | |
| "email": "" | |
| }, | |
| { | |
| "first": "Kentaro", | |
| "middle": [], | |
| "last": "Torisawaf", | |
| "suffix": "", | |
| "affiliation": {}, | |
| "email": "" | |
| }, | |
| { | |
| "first": "Yuka", | |
| "middle": [], | |
| "last": "Tateisif", | |
| "suffix": "", | |
| "affiliation": {}, | |
| "email": "" | |
| }, | |
| { | |
| "first": "Jun", | |
| "middle": [ | |
| "'" | |
| ], | |
| "last": "Ichi Tsujiift", | |
| "suffix": "", | |
| "affiliation": {}, | |
| "email": "" | |
| } | |
| ], | |
| "year": "", | |
| "venue": null, | |
| "identifiers": {}, | |
| "abstract": "", | |
| "pdf_parse": { | |
| "paper_id": "W98-0127", | |
| "_pdf_hash": "", | |
| "abstract": [], | |
| "body_text": [ | |
| { | |
| "text": "This paper describes a method for packing feature structures, which is used for reducing the number of constituents generated during parsing, and for improving the parsing speed. The method was developed for optimizing a parsing system for XHPSG (Tateisi et al\" 1998) translated from XTAG (The XTAG Research Group, 1995 ). The XHPSG system is a widecoverage parsing system for English based on HPSG framework (Pollard and Sag, 1994) . This system is also intended to be used for processing large amounts of texts, for the purposes such as information extraction. Current parsing speed of our system is not suffi.cient enough to achieve this goal.", | |
| "cite_spans": [ | |
| { | |
| "start": 246, | |
| "end": 267, | |
| "text": "(Tateisi et al\" 1998)", | |
| "ref_id": "BIBREF3" | |
| }, | |
| { | |
| "start": 289, | |
| "end": 319, | |
| "text": "(The XTAG Research Group, 1995", | |
| "ref_id": "BIBREF4" | |
| }, | |
| { | |
| "start": 409, | |
| "end": 432, | |
| "text": "(Pollard and Sag, 1994)", | |
| "ref_id": "BIBREF2" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "Our method improves the parsing speed by solving the problem which the XHPSG and the XTAG system have. That is, many lexical entries are assigned to a word, and many constituents are produced during parsing. The experimental results show that our method leads to a significant speed-up. The results also suggest the possihility of optimizing the XTAG system by introducing packing offeature structures and packing of tree structures, although these operations are not currently so apparent.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "This section describes the current status of the XHPSG system and the efficiency problem in the system. Both of the grammar and the parser in the XHPSG system are implemented with feature structure description language, LiLFeS (Makino et al., 1998) . The grammar consists of lexical entries for about 317 ,000 words, and 10 schemata, which follows schemata of the 'This work is partially founded by Japan Society for the Promotion of Science (JSPS-RFTF96P00502). HPSG framework in (Pollard and Sag, 199~) with slight modifications. The parser is a simplt-CKY-based parser.", | |
| "cite_spans": [ | |
| { | |
| "start": 227, | |
| "end": 248, | |
| "text": "(Makino et al., 1998)", | |
| "ref_id": "BIBREF1" | |
| }, | |
| { | |
| "start": 481, | |
| "end": 498, | |
| "text": "(Pollard and Sag,", | |
| "ref_id": null | |
| }, | |
| { | |
| "start": 499, | |
| "end": 504, | |
| "text": "199~)", | |
| "ref_id": null | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "The XHPSG System", | |
| "sec_num": "2" | |
| }, | |
| { | |
| "text": "Currently, the parsing speed of this system is not satisfactory, and we need further impro\\'e\u2022 ment of the parsing speed. One of the major reasons of ineffi.ciency is that the XHPSG system assigns many lexical entries to a sing]C' word. For example, a noun is assigned 11 lexica! entries, a verb is assigned 20-30 lexical entries, and some wor<ls are even assigned more than 100 entries.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "The XHPSG System", | |
| "sec_num": "2" | |
| }, | |
| { | |
| "text": "This characteristic is inherited from the XTAG grammar. The XTAG grammar assigns many elementary trees to a single word, and there is a one-to-one correspondence between a lexical entry in XHPSG and an elementary tree in the XTAG grammar. The XTAG system applies a POS tagger before parsing in or<ler to overcome this ineffi.ciency by reducing the num\u2022 her of lexical entries assigned to a word. However, this method sacrifices the soundness of the ", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "The XHPSG System", | |
| "sec_num": "2" | |
| }, | |
| { | |
| "text": "PASS J [ SUBJ ( \u2022\u2022\u2022\u2022 )] VAL COMPS () SPR () \"= cf11 r21 G1i", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "The XHPSG System", | |
| "sec_num": "2" | |
| }, | |
| { | |
| "text": ".O.~ ~m~t;T,\u00b06oofe4a 1 iooleo\u2022), ( unodc...pfGrft plu, pla,)} Figure 3 : A packed feature structure for the lexical entries shown in Figure 2 .", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 62, | |
| "end": 70, | |
| "text": "Figure 3", | |
| "ref_id": null | |
| }, | |
| { | |
| "start": 133, | |
| "end": 141, | |
| "text": "Figure 2", | |
| "ref_id": null | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "The XHPSG System", | |
| "sec_num": "2" | |
| }, | |
| { | |
| "text": "parsing process. In the case that the tagger fails to assign the correct POS to a word, correct syntactic structures may not be created even when the grammar potentiilly covers such structures. To solve the same problem, we propose a new method described in the next section. The method can gain a similar effect, but does not sacrifice the soundness of parsing.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "The XHPSG System", | |
| "sec_num": "2" | |
| }, | |
| { | |
| "text": "The left hand side of Figure 1 illustrates the data flow of the original parser of the XHPSG system. There are two major operations, unification and factoring. When we apply a schema to daughters, a unification operation is performed, and a mothcr is created. ~A.. set of mothers are reduced to a smaller set of feature struct ures by facto ring operation 1 , and these con- The right hand side of Figure 1 illustrates the parser with the packing module. The unification and the factoring operation in the original parser wa.s replaced by unification of packed feature structures and dynamic packing. These operations are more efficient than the cor\u2022 responding one, because multiple appUcation~ of schemata are reduced to one unification or packed feature structures, and multiple opera \u2022 tions of factoring are reduced to one dynamic packing. In addition, dynamic packing reduces the constituents further than the factoring op\u2022 eration.", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 22, | |
| "end": 30, | |
| "text": "Figure 1", | |
| "ref_id": "FIGREF1" | |
| }, | |
| { | |
| "start": 398, | |
| "end": 406, | |
| "text": "Figure 1", | |
| "ref_id": "FIGREF1" | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Packing of Feature Structures", | |
| "sec_num": "3" | |
| }, | |
| { | |
| "text": "createJ>FS(F) C:==J_, v:=(), 5:=() for each f E features(F) if f E DisjFentures then v :={follo11(C, f))fJv 5 := {follow(F, f))tJ5 ' else F' :=follow(F ,f) (C',v', {5'}) =createJ>FS(F') C :=Cu[! C'J v :== vfJv', 5 := 6tJ5' end..if end_for return (C,v,{5})", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Packing of Feature Structures", | |
| "sec_num": "3" | |
| }, | |
| { | |
| "text": "With a simple example, now we see how fea\u2022 ture structures are packed into one. Figure 2 shows two of the lexical entries that the XH PSG system assigns to an English verb \"walked'\". These lexical en tri es correspond to distinct ele\u2022 mentary trees of XTAG. They are different in only a few features, whil~ each feature structure has over 100 features. That is, most of them have equivalent values, so that it is redundant to have each of them as two independent featurC' structures.", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 80, | |
| "end": 88, | |
| "text": "Figure 2", | |
| "ref_id": null | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Packing of Feature Structures", | |
| "sec_num": "3" | |
| }, | |
| { | |
| "text": "For these feature structures, a packed feature structure is described as in Figure 3 . C specifies the common part of the original two feature explosion of the time complexity of the parsing of CFG. In the case of HPSG, the similar effect can be accorn\u2022 plished by the factoring operation, which iden tifies lhe constituents with equivalent feature structures in this ca.se. We have observed that parsing time with syntartic grammars can be reduced significant!y, though this operation does not lead to a reduction of compulational Lime complexity Lo polynomial. structures. v expresses the nodes 2 in the feature structure, to which disjunctive structures are incorporated. The nodes are expressed as tags for structure sharings such as ITJ. ,6. expresses a set of different values, that come to the position specified by the nodes in v. Hence, the original feature structures are obtained by unifying one of the elements of ,6. to the nodes in v. A packed feature structure holds exactly the equivalent information of the original feature structures with a smaller data size.", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 76, | |
| "end": 84, | |
| "text": "Figure 3", | |
| "ref_id": null | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Packing of Feature Structures", | |
| "sec_num": "3" | |
| }, | |
| { | |
| "text": "This section describes three algorithms, (l)conversion of a feature structure to a packed feature structure, (2)packing of packed feature struc-2 Though feature structures are expressed in a conventional matrix-like notation, they can be seen as directed graph with a root whose nodes and arcs are labeled. Features are labels for arcs and the labels for nodes are called types.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Algorithms", | |
| "sec_num": "4" | |
| }, | |
| { | |
| "text": "tures and (3)unification of packed feature structures.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Algorithms", | |
| "sec_num": "4" | |
| }, | |
| { | |
| "text": "The last two algorithm requires packed feature structures as their inputs and the first algorithm is used for convert non-disjunctive feature structures to such inputs to the two algorithms. Figure 4 shows the first algorithm for converting a feature structure to a new packed feature structure. vVe assume that a packed feature structure is given as a triple (C, v, .6.) as described in Section 3. The input to this algorithm is a (non-disjunctive) feature structure and a set of features, to which the disjunction is ure , P FS denotes a given set of packed feature structures, and P FS' denotes a newly created set of packed feature structures. The function paths(F, n) returns a set of all the paths to the node n in F. The algorithm for packing lexical entries is straightforwardly obtained from this algorithm and the previous algorithm. Figure 6 shows the algorithm for unifying two packed feature structures. The overall algorithm is similar to the one in (Kasper, 1987) , although data structures for disjunctive feature structures are different. Intui tively, we first unify common parts (C1 and C2), and next check consistency of each combination of disjuncts in .6.1 and .6.2. The operator U denotes the unificaHon of non-disjunctive feature structures 3 . The unification is regarded as an destructive procedure in the figure. lt has a side effect to the input feature structures. For instance, suppose that feature structures stored in the variables F and F' have the nooes stored in the variable n and n' as their substructure and that for some path rr fol/ow(F, rr) = n, J ol low( F', rr) = n' an d n :/:. n'. After performing the . unifi~ation FUF',.the values of F,F',n and n' are automatically updated and, as a result of the update, F = F' and n = n' hold.", | |
| "cite_spans": [ | |
| { | |
| "start": 360, | |
| "end": 371, | |
| "text": "(C, v, .6.)", | |
| "ref_id": null | |
| }, | |
| { | |
| "start": 964, | |
| "end": 978, | |
| "text": "(Kasper, 1987)", | |
| "ref_id": "BIBREF0" | |
| } | |
| ], | |
| "ref_spans": [ | |
| { | |
| "start": 191, | |
| "end": 199, | |
| "text": "Figure 4", | |
| "ref_id": "FIGREF3" | |
| }, | |
| { | |
| "start": 519, | |
| "end": 522, | |
| "text": "ure", | |
| "ref_id": null | |
| }, | |
| { | |
| "start": 844, | |
| "end": 852, | |
| "text": "Figure 6", | |
| "ref_id": null | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Algorithms", | |
| "sec_num": "4" | |
| }, | |
| { | |
| "text": "In the algorithm in the figure, this type of si<le- The rnechanisrns for the side-effect and its canceling are similar to the execution rnechansims of Prolog, including backtracking. They are also irnplernented in LiLFeS. The copy is a procedure to create a distinct feature structure equivalent to the input feature structure and the newly created feature structure is free frorn the side-effect of the unification against the original input feature structure.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Algorithms", | |
| "sec_num": "4" | |
| }, | |
| { | |
| "text": "This section shows the experimental results of the current implernentation of our packing method. Experiments are performed by specifying features originated in XTAG and a few other features as in Table 1 . The packing module .is irnplemented with LiLFeS, and is incorporated into the XH-PSG systern.", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 197, | |
| "end": 204, | |
| "text": "Table 1", | |
| "ref_id": "TABREF2" | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Experiments", | |
| "sec_num": "5" | |
| }, | |
| { | |
| "text": "We compared the parsing times of (l)Test set A (337 sentences, 8.37 words/sentence) 4 and (2)Test set B (16 sentences, 11.88 words/sentence ) 5 , between the (l)New System (with the packing rnodule) and the (2)0ld System (without the packing rnodule). The parsers of both systems are simple CKY-based parsers. As Table 2 shows, the parsing speed improves by 1.79 times in Testset A, and 2.46 times in Testset B, which consists of sentences costing much time to parse. In Test set A, the nurnber of lexical entries is reduced by 35.3%, and that of constituents in the CI\\Y table by 46. 7% on average.", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 313, | |
| "end": 320, | |
| "text": "Table 2", | |
| "ref_id": "TABREF3" | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Experiments", | |
| "sec_num": "5" | |
| }, | |
| { | |
| "text": "We proposed a method for packing feature structures by introducing disjunctions into feature structures. This method reduces the number of lexical entries in HPSG grarnrnars and constituents created during parsing. As a result, we achieved 1. 7 4 tim es irnprovement in parsing time for the test corpus bundled in the XTAG systern. We expect to gain the sirnilar effect with the XTAG system by applying our packing method, though it is currently not so apparent.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Conclusion and Future Work", | |
| "sec_num": "6" | |
| }, | |
| { | |
| "text": "For realizing a practical parsing system, we are currently integrating our packing method with other two optimization techniques: (l)irnplementation with a native compiler version of LiLFeS (Makino et al., 1998) , and (2)compilation of HPSG to CFG (Torisawa and Tsujii, 1996) . As a result of the latter optimization, current XHPSG system can parse sentences in the ATIS corpus in 1.12 seconds on average without any POS taggers. Further speed-up is expected by integrating our rnethod to this system.", | |
| "cite_spans": [ | |
| { | |
| "start": 190, | |
| "end": 211, | |
| "text": "(Makino et al., 1998)", | |
| "ref_id": "BIBREF1" | |
| }, | |
| { | |
| "start": 248, | |
| "end": 275, | |
| "text": "(Torisawa and Tsujii, 1996)", | |
| "ref_id": "BIBREF5" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Conclusion and Future Work", | |
| "sec_num": "6" | |
| }, | |
| { | |
| "text": "A factoring operation in a CI\\Y parser for CFG reduces the number of constituents by identifying constituenls described by the identical non-terminals. The operation plays a crucial role for avoiding an exponential", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "Unification of tuples is a tuple of the results of the unification of corresponding elements of the tuples.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "Test set Ais bundled in the XTAG systern for checking the grammar.5 Test set B is a subset of Test set A. The subset consists of 16 sentences, each of which costs more than 10 second!' to parse.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| } | |
| ], | |
| "back_matter": [], | |
| "bib_entries": { | |
| "BIBREF0": { | |
| "ref_id": "b0", | |
| "title": "A unification method for disjunctive feature descriptions", | |
| "authors": [ | |
| { | |
| "first": "Robert", | |
| "middle": [ | |
| "T" | |
| ], | |
| "last": "Kasper", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1987, | |
| "venue": "Proc. 26th ACL", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "235--242", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Robert T. Kasper. 1987. A unification method for disjunctive feature descriptions. In Proc. 26th ACL, pages 235-242.", | |
| "links": null | |
| }, | |
| "BIBREF1": { | |
| "ref_id": "b1", | |
| "title": "LiLFeS -towards a practical HPSG parser", | |
| "authors": [ | |
| { | |
| "first": "Takaki", | |
| "middle": [], | |
| "last": "Makino", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Minoru", | |
| "middle": [], | |
| "last": "Yoshida", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Kentaro", | |
| "middle": [], | |
| "last": "Torisawa", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Jun'ichi", | |
| "middle": [], | |
| "last": "Tsujii", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1998, | |
| "venue": "Proc. of COLING-ACL '98", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Takaki Makino, Minoru Yoshida, Kentaro Torisawa, and Jun'ichi Tsujii. 1998. LiLFeS -towards a practical HPSG parser. In Proc. of COLING- ACL '98. To appear.", | |
| "links": null | |
| }, | |
| "BIBREF2": { | |
| "ref_id": "b2", | |
| "title": "Head-Driven Phrase Structure Grammar", | |
| "authors": [ | |
| { | |
| "first": "C", | |
| "middle": [], | |
| "last": "Pollard", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": ".", | |
| "middle": [ | |
| "A" | |
| ], | |
| "last": "Sag", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1994, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "C. Pollard and 1. A. Sag. 1994. Head-Driven Phrase Structure Grammar. University of Chicago Press.", | |
| "links": null | |
| }, | |
| "BIBREF3": { | |
| "ref_id": "b3", | |
| "title": "Translating the XTAG English grammar to HPSG", | |
| "authors": [ | |
| { | |
| "first": "Yuka", | |
| "middle": [], | |
| "last": "Tateisi", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Kentaro", | |
| "middle": [], | |
| "last": "Torisawa", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Yusuke", | |
| "middle": [], | |
| "last": "Miyao", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Jun'ichi", | |
| "middle": [], | |
| "last": "Tsujii", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1998, | |
| "venue": "Proc. TAG+ Workshop", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Yuka Tateisi, Kentaro Torisawa, Yusuke Miyao, and Jun'ichi Tsujii. 1998. Translating the XTAG En- glish grammar to HPSG. In Proc. TAG+ Work- shop.", | |
| "links": null | |
| }, | |
| "BIBREF4": { | |
| "ref_id": "b4", | |
| "title": "A lexicalized tree adjoining grammar for Eng!ish. Tedmirnl report", | |
| "authors": [ | |
| { | |
| "first": "Xtag Research", | |
| "middle": [], | |
| "last": "The", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "", | |
| "middle": [], | |
| "last": "Group", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1995, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "The XTAG Research Group. 1995. A lexicalized tree adjoining grammar for Eng!ish. Tedmirnl re- port, IRCS Report 95-03, University of Pennsyl- vania, March.", | |
| "links": null | |
| }, | |
| "BIBREF5": { | |
| "ref_id": "b5", | |
| "title": "Computing phrasal-signs in HPSG prior to parsing", | |
| "authors": [ | |
| { | |
| "first": "Kentaro", | |
| "middle": [], | |
| "last": "Torisawa", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "Jun'ichi", | |
| "middle": [], | |
| "last": "Tsujii", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1996, | |
| "venue": "Proc. 16th COLING", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "949--955", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Kentaro Torisawa and Jun'ichi Tsujii. 1996. Com- puting phrasal-signs in HPSG prior to parsing. In Proc. 16th COLING, pages 949-955.", | |
| "links": null | |
| } | |
| }, | |
| "ref_entries": { | |
| "FIGREF1": { | |
| "type_str": "figure", | |
| "num": null, | |
| "uris": null, | |
| "text": "Data flow in the parsers for the X 11-PSG system." | |
| }, | |
| "FIGREF3": { | |
| "type_str": "figure", | |
| "num": null, | |
| "uris": null, | |
| "text": "Algorithm for creating new packed feature structure from a feature structure. '\u00a9' denotes the concatenation operation of sequences. stituents are put into CKY table." | |
| }, | |
| "FIGREF4": { | |
| "type_str": "figure", | |
| "num": null, | |
| "uris": null, | |
| "text": "pack_f ea ture ..s tructures (P :F S) P:FS' := r/J for each P = (C, v, D.) E P:FS if P' = (C',v',D.') E P:FS' such that C' is equivalent to C and, for each i(O < i < /,:) paths(C, ni) = -paths{C 1 , nl) where v = (no , \u2022 \u2022 \u2022, 11~) and v' = (nb, \u2022 \u2022 \u2022, n~) then D.\" := D. u D.' P:FS' := (P:FS'\\{P'})U {(C,v,D.\")} else P:FS' := P:FS' u {P} end_if end.for return P:FS'" | |
| }, | |
| "FIGREF5": { | |
| "type_str": "figure", | |
| "num": null, | |
| "uris": null, | |
| "text": "Algorithm for packing a set of packed feature structures. uni f y _packed_feature ..s tructures (P 1 , P 2 ) Pi= (Ci.v1,D.1) P2 = (C2, V2, ll.1) il := <P if success C :=Ci U C2 then v := v1 0v2 for each 01 E ll.1 and 62 E D.2 o :;:; copy((v1uoi)fl(v2u62)) .. U1 D. := ll. u {o} Cancel the side-effect of U occuring during computation of U 1 \u2022 end_for end_if return ( C, v, \u00df) Algorithm for unifying two packed feature structures." | |
| }, | |
| "FIGREF6": { | |
| "type_str": "figure", | |
| "num": null, | |
| "uris": null, | |
| "text": "introduced. In the figure, F is a feature structure and DisjFeatures is a set of features. The function follo\\l(F, j) returns the node in F reached by the feature f from a root of F. What the algorithm does is to split F into two parts, the first part is C and the other part is a set of nodes and a set of substructures represented by v and ,6. respectively." | |
| }, | |
| "FIGREF7": { | |
| "type_str": "figure", | |
| "num": null, | |
| "uris": null, | |
| "text": "Figure 5shows the algorithm for packing already packed feature structures. In the fig-" | |
| }, | |
| "TABREF0": { | |
| "content": "<table><tr><td>1 _ \u2022t:SYNSEM[LOC[CAT [HEAD r ~i;;~~r.~ ] l PHON ( \" w\u2022lkcd\" ) VAL COMPS () [ slBJ ( \u2022\u2022\u2022\u2022 )] SPR ()</td></tr><tr><td>Figure 2: Two of the lexlcal entries for an En-</td></tr><tr><td>glish verb \"walked\". Underlined values are dif-</td></tr><tr><td>ferent. Most of the features are omitted for sim-</td></tr><tr><td>plicity.</td></tr><tr><td>wortl</td></tr><tr><td>PHON ( \" walked \" )</td></tr><tr><td>c, SYNSEM[LOCICAT [llEAD [f.~~N~RI] ]</td></tr></table>", | |
| "type_str": "table", | |
| "text": "", | |
| "num": null, | |
| "html": null | |
| }, | |
| "TABREF1": { | |
| "content": "<table><tr><td>Other features</td></tr><tr><td>HEADPHON, MARKING, CONT, TRF</td></tr></table>", | |
| "type_str": "table", | |
| "text": "Features incorporated from XTAG PRO, CASE, PRON. REFL, VMODE, l\\IAJNV, EXTRACT, TRANS, PASS, PERF, PROG, ASSIGN_CASE, JNV", | |
| "num": null, | |
| "html": null | |
| }, | |
| "TABREF2": { | |
| "content": "<table><tr><td colspan=\"2\">arsing time m avg. (sec.</td></tr><tr><td>est set</td><td>est set</td></tr><tr><td>2.31</td><td>14.45</td></tr><tr><td>1.29</td><td>5.88</td></tr><tr><td>1.79</td><td>2.46</td></tr><tr><td colspan=\"2\">The experiments are performed on Alpha Station 500</td></tr><tr><td colspan=\"2\">(500MHz CPU, 256MB Memory), and the times are</td></tr><tr><td>measured in User Time.</td><td/></tr></table>", | |
| "type_str": "table", | |
| "text": "Specified features for the experiments.", | |
| "num": null, | |
| "html": null | |
| }, | |
| "TABREF3": { | |
| "content": "<table><tr><td>effects is assumed to occur for the values stored</td></tr><tr><td>in the variables such as C1,C2,v1,v2,8 1 and 82.</td></tr></table>", | |
| "type_str": "table", | |
| "text": "Results of the experiments.", | |
| "num": null, | |
| "html": null | |
| } | |
| } | |
| } | |
| } |