| { | |
| "paper_id": "Y98-1026", | |
| "header": { | |
| "generated_with": "S2ORC 1.0.0", | |
| "date_generated": "2023-01-19T13:37:24.903732Z" | |
| }, | |
| "title": "A Computational Method for Resolving Ambiguities in Coordinate Structures", | |
| "authors": [ | |
| { | |
| "first": "Haodong", | |
| "middle": [], | |
| "last": "Wu", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "University of Electro-Communications", | |
| "location": { | |
| "addrLine": "1-5-1 Chofugaoka", | |
| "postCode": "182", | |
| "settlement": "Chofu", | |
| "region": "Tokyo", | |
| "country": "JAPAN" | |
| } | |
| }, | |
| "email": "" | |
| }, | |
| { | |
| "first": "Teiji", | |
| "middle": [], | |
| "last": "Furugori", | |
| "suffix": "", | |
| "affiliation": { | |
| "laboratory": "", | |
| "institution": "University of Electro-Communications", | |
| "location": { | |
| "addrLine": "1-5-1 Chofugaoka", | |
| "postCode": "182", | |
| "settlement": "Chofu", | |
| "region": "Tokyo", | |
| "country": "JAPAN" | |
| } | |
| }, | |
| "email": "iwujurugorij@phaeton.cs.uec.ac.jp" | |
| } | |
| ], | |
| "year": "", | |
| "venue": null, | |
| "identifiers": {}, | |
| "abstract": "This paper describes a method for determining syntactic structure in coordinate constructions. It is based on the information taken from semantic similarities, selectional restrictions, and some other linguistic cues. We discuss the role the information plays in resolving ambiguities that appear in coordinate constructions, describe the means of acquiring the necessary information automatically from two on-line corpora and a lexical database, and devise two algorithms for disambiguating coordinate constructions. An experiment that follows shows effectiveness of our method and its applicability to resolving ambiguities in some other syntactic structures.", | |
| "pdf_parse": { | |
| "paper_id": "Y98-1026", | |
| "_pdf_hash": "", | |
| "abstract": [ | |
| { | |
| "text": "This paper describes a method for determining syntactic structure in coordinate constructions. It is based on the information taken from semantic similarities, selectional restrictions, and some other linguistic cues. We discuss the role the information plays in resolving ambiguities that appear in coordinate constructions, describe the means of acquiring the necessary information automatically from two on-line corpora and a lexical database, and devise two algorithms for disambiguating coordinate constructions. An experiment that follows shows effectiveness of our method and its applicability to resolving ambiguities in some other syntactic structures.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Abstract", | |
| "sec_num": null | |
| } | |
| ], | |
| "body_text": [ | |
| { | |
| "text": "Syntactic ambiguity appears, among others, in coordinate constructions. It is an annoying problem in analyzing structure and meaning of a sentence. A parser, for instance, is to detect the scope of a coordinate structure and identify its inner modification relations. However, the current parsers (e.g., the Link parser) often fail to handle the problem and/or produce a large number of parses.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "There are a few computational studies that have tried to resolve ambiguities in coordinate constructions (e.g., Paritong, 1992; Cooper, 1991; Bayer, 1996) . For example, Kurohashi and Nagao (1994) , in analyzing long Japanese sentences, proposed a syntactic analysis method for detecting conjunctive structures by using lexical similarity and structural parallelism. Mela and Fouquere (1996) used a direct process to determine the scope of a coordinate structure based on the concept of functor, argument and subcategorization. Unfortunately, neither of them has sufficiently dealt with .the syntactic structure of a coordination especially when a coordinator (such as and, or and comma) has two or possibly more preceding and succeeding constituents.", | |
| "cite_spans": [ | |
| { | |
| "start": 112, | |
| "end": 127, | |
| "text": "Paritong, 1992;", | |
| "ref_id": "BIBREF12" | |
| }, | |
| { | |
| "start": 128, | |
| "end": 141, | |
| "text": "Cooper, 1991;", | |
| "ref_id": "BIBREF3" | |
| }, | |
| { | |
| "start": 142, | |
| "end": 154, | |
| "text": "Bayer, 1996)", | |
| "ref_id": "BIBREF1" | |
| }, | |
| { | |
| "start": 170, | |
| "end": 196, | |
| "text": "Kurohashi and Nagao (1994)", | |
| "ref_id": "BIBREF9" | |
| }, | |
| { | |
| "start": 367, | |
| "end": 391, | |
| "text": "Mela and Fouquere (1996)", | |
| "ref_id": "BIBREF10" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "We in this paper propose a method for determining the structure of a coordinate construction using information on similarities, selectional restrictions, and oilier linguistic cues. In Section 2 we identify the problem and describe the ideas behind our method in Section 3. We give disambiguation algorithms, show an disambiguation experiment, and evaluate its results in Section 4. In Section 5 we suggest an applicability of our method to resolving other syntactic ambiguities.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Introduction", | |
| "sec_num": "1" | |
| }, | |
| { | |
| "text": "Resolving ambiguities in a coordinate construction is to determine the way of conjoining constituents (words, phrases, or clauses) and/or to determine the scope of coordination, i.e., immediacy relations among the constituents involved. For instance, in sweet and sour pork, the right immediacy relation is ((sweet and sour) pork) rather than (sweet and (sour pork)).", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Modification Relation in Coordination", | |
| "sec_num": "2" | |
| }, | |
| { | |
| "text": "Consider other examples.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Modification Relation in Coordination", | |
| "sec_num": "2" | |
| }, | |
| { | |
| "text": "(1) Tom is a ((stock and estate) keeper).", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Modification Relation in Coordination", | |
| "sec_num": "2" | |
| }, | |
| { | |
| "text": "(2) John is a (student and (chess player)).", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Modification Relation in Coordination", | |
| "sec_num": "2" | |
| }, | |
| { | |
| "text": "(3) Old men and women were left at the village.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Modification Relation in Coordination", | |
| "sec_num": "2" | |
| }, | |
| { | |
| "text": "In each of these sentences, a noun or an adjective that appears in the left hand side of coordinator(s) has two (or more) modificant candidates: it may or may not modify the head noun in the right hand side.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Modification Relation in Coordination", | |
| "sec_num": "2" | |
| }, | |
| { | |
| "text": "In 1, stock and estate should be conjoined to modify keeper. In (2), there is not a modification relation between student and player and its interpretation should be (student and (chess player)).", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Modification Relation in Coordination", | |
| "sec_num": "2" | |
| }, | |
| { | |
| "text": "We are able to make unique interpretation for (1) and (2), but we may have two interpretations for (3).", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Modification Relation in Coordination", | |
| "sec_num": "2" | |
| }, | |
| { | |
| "text": "(3a) Women are left at the village and old men were left at the village. (3b) Old men were left at the village and old women were left at the village.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Modification Relation in Coordination", | |
| "sec_num": "2" | |
| }, | |
| { | |
| "text": "Later in the paper, we try to resolve the ambiguities by determining the modification relation in the coordinate constructions (CCs) such as in (1) to (3).", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Modification Relation in Coordination", | |
| "sec_num": "2" | |
| }, | |
| { | |
| "text": "We have found through linguistic observations that a variety of information supplies important cues for disambiguation. Some of them are computable and effectively used in a computer model of disambiguation. The ones we thought most important include: similarities in syntactic forms and/or meanings, selectional restrictions, and orthographic forms.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Identifying Modification Relation in Coordination", | |
| "sec_num": "3" | |
| }, | |
| { | |
| "text": "Similarities in Syntactic Forms and Meanings We see that similarities on forms and meanings are crucial to determine the structure of a coordinate construction.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Linguistic Observations", | |
| "sec_num": "3.1" | |
| }, | |
| { | |
| "text": "If the modifier is not an adjective, for instance, it is likely that two constituents before and after the coordinator are conjoined when they belong to the same subcategory and match in number:", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Linguistic Observations", | |
| "sec_num": "3.1" | |
| }, | |
| { | |
| "text": "(4) ((business and management) sections) (5) (businesses and (culture activities))", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Linguistic Observations", | |
| "sec_num": "3.1" | |
| }, | |
| { | |
| "text": "In the following examples, (6) ((research and development) section) (7) (researcher and (system engineer)) it is obvious that the research and development have more in common in meaning than research and section, and that researcher and engineer are semantically more similar than researcher and system. Likewise, we see:", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Linguistic Observations", | |
| "sec_num": "3.1" | |
| }, | |
| { | |
| "text": "(8) (lovely (cats and dogs)) *((lovely cats ) and dogs)", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Linguistic Observations", | |
| "sec_num": "3.1" | |
| }, | |
| { | |
| "text": "Selectional Restrictions Consider the sentence:", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Linguistic Observations", | |
| "sec_num": "3.1" | |
| }, | |
| { | |
| "text": "(9) Peter likes ((green vegetables) and (music)).", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Linguistic Observations", | |
| "sec_num": "3.1" | |
| }, | |
| { | |
| "text": "we know in (9) that green as a color can be used to modify concrete entities like vegetables, but not abstract one like music. This means that selectional restriction (SR), a semantic restriction imposed on lexical items when forming a sentence, is an important factor to determining the structure of coordinate constructions. In this paper, we discuss SR in the context of adj+nl+and+n2 and its extension (e.g., adj n 1 , nk)2.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Linguistic Observations", | |
| "sec_num": "3.1" | |
| }, | |
| { | |
| "text": "Other Linguistic Cues Orthographic forms often play an important role in disambiguating the structures of coordinate constructions. It is likely that all nouns can be conjoined when they are in capital forms. An example:", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Linguistic Observations", | |
| "sec_num": "3.1" | |
| }, | |
| { | |
| "text": "(10) ((Research and Development) Section)", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Linguistic Observations", | |
| "sec_num": "3.1" | |
| }, | |
| { | |
| "text": "When the conjoined nouns in a coordinate structure are preceded by a determiner, the usual interpretation is that the determiner applies to the conjoins: (11) Old men and women were left to organize the community.", | |
| "cite_spans": [ | |
| { | |
| "start": 154, | |
| "end": 158, | |
| "text": "(11)", | |
| "ref_id": null | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Linguistic Observations", | |
| "sec_num": "3.1" | |
| }, | |
| { | |
| "text": "The structure (old (men and women)) is more likely than ((old men) and women) because there is a tendency that the determiner is not repeated in the noninitial conjoins, e.g., the (man and women) (Quirk et al. 1985) .", | |
| "cite_spans": [ | |
| { | |
| "start": 196, | |
| "end": 215, | |
| "text": "(Quirk et al. 1985)", | |
| "ref_id": null | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Linguistic Observations", | |
| "sec_num": "3.1" | |
| }, | |
| { | |
| "text": "Semantic similarity has been measured in a number of ways (e.g., Resnik 1993; Kozima and Furugori, 1993; Dagan et al., 1994) . Many researchers opt to investigate methods for deriving measures of semantic similarity among words based on distributional behavior observed in corpora or machine-readable dictionaries. We however employ a hybrid method for combining the quantitative method with an existing broad-coverage source of lexical knowledge.", | |
| "cite_spans": [ | |
| { | |
| "start": 65, | |
| "end": 77, | |
| "text": "Resnik 1993;", | |
| "ref_id": "BIBREF13" | |
| }, | |
| { | |
| "start": 78, | |
| "end": 104, | |
| "text": "Kozima and Furugori, 1993;", | |
| "ref_id": null | |
| }, | |
| { | |
| "start": 105, | |
| "end": 124, | |
| "text": "Dagan et al., 1994)", | |
| "ref_id": null | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Computational Measurements of Similarities", | |
| "sec_num": "3.2" | |
| }, | |
| { | |
| "text": "To capture the similarity between two words in context, we consider two kinds of semantic relations useful and effective: taxonomic relation and co-occurrence relation.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Computational Measurements of Similarities", | |
| "sec_num": "3.2" | |
| }, | |
| { | |
| "text": "Computationally, one way of measuring semantic similarity is to use the taxonomic relations in the WordNet (Miller, 1990) , a widely used lexical database including four kinds of semantic relationships in a word sense network: synonymy, hyponymy, meronymy, and antonymy. We found that the antonymy relationship appears often in CCs (e.g., brother and sister, man and woman, and boy and girl) and shows a strong tie between two nouns in the coordinate structure like (14).", | |
| "cite_spans": [ | |
| { | |
| "start": 107, | |
| "end": 121, | |
| "text": "(Miller, 1990)", | |
| "ref_id": "BIBREF11" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Computational Measurements of Similarities", | |
| "sec_num": "3.2" | |
| }, | |
| { | |
| "text": "Another way of measuring semantic similarity is to use the mutual information(MI) (Church and Hanks, 1990) . It may be taken from the co-occurrence relations using two corpora: the EDR English Corpus and Brown Corpus'. We define the similarity between two words w 1 and w2 when appearing with an adjacent word w. In an example, (13) new books and hamburgers where w i , w2 , and w are book, hamburger and new, respectively. We can measure two similarities, the left side similarity and right side similarity. w comes to the left of w1 and w2 in the former like young in (14) and it comes to the right side of w1 and w2 in' the latter like investigation in (15). The left side similarity and the right side similarity are defined as:", | |
| "cite_spans": [ | |
| { | |
| "start": 82, | |
| "end": 106, | |
| "text": "(Church and Hanks, 1990)", | |
| "ref_id": "BIBREF2" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "(12) (cheerful (boys and girls))", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "simi, (W1 W27 W) max(I(w, wi),/(w,w2)) ming(wi , w), I(w2 , w)) SiMR (W1 ' W2 ' W) max(I(w i ,w), Aw2,w))", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "(12) (cheerful (boys and girls))", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "Here, for two words x and y, I(x,y) is MI between x and y, where x is on the left side of y, and I(y,x) is MI between y and x, where x is on the right side of y.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "(12) (cheerful (boys and girls))", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "If w appears in either side of w1 , or in either side of w 2 , we can define two-sided similarity as:", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "(12) (cheerful (boys and girls))", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "SiML (W1, W2, W) SiMR(w i, w2, w) sim(wi , W27 W) 2", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "(12) (cheerful (boys and girls))", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "Here, w may be a verb, a noun, an adjective, or a phrase like agree with.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "(12) (cheerful (boys and girls))", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "3EDR English Corpus, compiled by Japan Electronic Dictionary Research Institute, Ltd., contains ", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "(12) (cheerful (boys and girls))", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "Often the word pairs from w, w 1 and w2 are unobserved in the corpus being used. This problem is known as the sparse data problem. There are a number of methods for dealing with this problem (e.g., Resnik, 1993; Dagan et al., 1995) . Our solution in this paper is to use a synonym set (S) in the WordNet to substituting for the sparse word w. The above formulae in this case becomes:", | |
| "cite_spans": [ | |
| { | |
| "start": 198, | |
| "end": 211, | |
| "text": "Resnik, 1993;", | |
| "ref_id": "BIBREF13" | |
| }, | |
| { | |
| "start": 212, | |
| "end": 231, | |
| "text": "Dagan et al., 1995)", | |
| "ref_id": "BIBREF4" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "(12) (cheerful (boys and girls))", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "simL(wi, w2 , w ) = sim R(w i, w2, w) =", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "(12) (cheerful (boys and girls))", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "EcES min (I (c,w 1 ) , -4;11)2)) EcES max (I(c, w1 ) ", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 9, | |
| "end": 20, | |
| "text": "(I (c,w 1 )", | |
| "ref_id": null | |
| }, | |
| { | |
| "start": 42, | |
| "end": 52, | |
| "text": "(I(c, w1 )", | |
| "ref_id": null | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "(12) (cheerful (boys and girls))", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "EQUATION", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [ | |
| { | |
| "start": 0, | |
| "end": 8, | |
| "text": "EQUATION", | |
| "ref_id": "EQREF", | |
| "raw_str": ", w2)) EcES min(gwi ' c), i(w2 ' c)) EcEs max(I(wl ' c), 1(w2 ' c))", | |
| "eq_num": "(4)" | |
| } | |
| ], | |
| "section": "(12) (cheerful (boys and girls))", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "here c stands for a synonym set related to the word w. wi and w2 may be replaced by their synonym sets S1 and S2 in the WordNet. In this case, the similarity between them is estimated by:", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "(12) (cheerful (boys and girls))", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "c1Es1,c2Es2 E cES ming (c, c 1 ), I (c, C2)) simL(w i, w2, w) Max max (I (c, c1 ), I (c, c2)) ci es' ,c2Es2 EcES ming I ( c27 c)) simR(wi w2, w) = Max ci,ac2 E cES max(' c), (c2, c))", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "(12) (cheerful (boys and girls))", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "The values produced by (6) and 7have an intuitive interpretation. Each of them denotes a maximum similarity between synonyms of w 1 and synonyms of w2 with synonyms of w.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "(12) (cheerful (boys and girls))", | |
| "sec_num": null | |
| }, | |
| { | |
| "text": "Selectional restrictions(SRs) are often defined on an empirical basis using AI(artificial intelligence) techniques. Recently serveral research efforts have turned to corpus-based methods to define and acquire them. Along the line, we search a new approach for finding SRs using on-line corpora and a lexical database.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Selectional Restrictions as Constraints", | |
| "sec_num": "3.3" | |
| }, | |
| { | |
| "text": "We view SR as negative information, a constraint between two words. The information on Sits may be computed from corpora: for a particular adjective and a particular noun, we try to find 'similar' words and then check if they co-occur in the corpora. If no co-occurrences are observed in the corpora, we assume that there is a SR between the adjective and the noun.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Selectional Restrictions as Constraints", | |
| "sec_num": "3.3" | |
| }, | |
| { | |
| "text": "A problem we encounter here is how to find 'similar' words. It has been proven by an experiment that similar words generated by corpus-based clustering methods does not work well (Grefenstette 1993) . We choose to acquire similar words from a taxonomy (the WordNet).", | |
| "cite_spans": [ | |
| { | |
| "start": 179, | |
| "end": 198, | |
| "text": "(Grefenstette 1993)", | |
| "ref_id": "BIBREF5" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Selectional Restrictions as Constraints", | |
| "sec_num": "3.3" | |
| }, | |
| { | |
| "text": "It seems to be reasonable for a noun to use its direct hypernym (father node in IS-A hierarchy in the WordNet) and hyponyms of the hypernym (the siblings of the noun) as similar words. But it does not work for an adjective. A compromising measure for an adjective is to use synonyms rather than hypernyms as similar words, e.g., {pure,unmixed,undiluted} for pure.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Selectional Restrictions as Constraints", | |
| "sec_num": "3.3" | |
| }, | |
| { | |
| "text": "Thus, if n2 or its similar words co-occur adj or its synonyms (in the WordNet) in the corpora (the EDR English corpus and the Brown corpus), we may conclude that the CC has the structure (adj (n1 and n2)). Otherwise, it would be ((adj n1) and n2) as there is a SR between adj and n2. Take (16) for an example, (16) ((Fresh air) and sunshine) bring me health and feelings of joy.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Selectional Restrictions as Constraints", | |
| "sec_num": "3.3" | |
| }, | |
| { | |
| "text": "We are sure that fresh air and sunshine has the structure ((fresh air) and sunshine) as no cooccurrence between the synonyms of fresh and the similar words of sunshine is found.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Selectional Restrictions as Constraints", | |
| "sec_num": "3.3" | |
| }, | |
| { | |
| "text": "We have devised disambiguation algorithms based on what we have described in section 3. Algorithm 1 below is for CCs in the form adj+n1+and+n2 and algorithm 2 is for CCs in the form nl+and+n2+n3. Algorithm 3 describes the process of decomposing a normal CC into CCs in the forms adj+nl+and+n2 and n1+and+n2+n3, and determining its inner syntactic structure.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Disambiguation for Structure of Coordinate Constructions", | |
| "sec_num": "4" | |
| }, | |
| { | |
| "text": "Algorithm 1: Disambiguation of CCs in the form adj+nl+and+n2 6) (71. Check adj+n2 in the corpora. If it is observed, produce (adj (n1 and n2)). 2. Otherwise, if there is a synonym of adj and a synonym of n2 and if they co-occur in the corpora, produce (adj (n1 and n2)). 3. Otherwise, create a class which includes n2, its parent (hypernym), and its siblings (the hyponyms of the hypernym) in IS-A hierarchies of the WordNet; If no member in this class co-occurs with adj and any of its synonyms, produce ((adj n1) and n2). 4. Otherwise, produce (adj (n1 and n2)) as a statistics-based default.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Disambiguation for Structure of Coordinate Constructions", | |
| "sec_num": "4" | |
| }, | |
| { | |
| "text": "Algorithm 2: Disambiguation of CCs in the form n1-1-and-l-n2-1-n3 1. If all of the three nouns are capitalized, produce ((nl and n2) n3). 2. Otherwise, (a) if ni and n2 match in number and n1 and n3 do not, produce ((nl and n2) n3). (b) if n1 and n3 match in number and n1 and n2 do not, produce (n1 and (n2 n3)). 3. Otherwise, (a) if n1 is the antonym of n2, produce ((nl and n2) n3); (b) if n1 is the antonym of n3, produce (n1 and (n2 n3)). 4. Otherwise, (a) if simR (S1,S2,S3)>simL (S2,S3,S1)4 , produce ((n1 and n2) n3). (b) if simR (S1,S2,S3)<sim L (S2,S3,S1), produce (n1 and (n2 n3)). 5. Otherwise, produce ((nl and n2) n3).", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Disambiguation for Structure of Coordinate Constructions", | |
| "sec_num": "4" | |
| }, | |
| { | |
| "text": "Experimental Results To evaluate the performance of the disambiguation algorithm, we randomly selected two sets of 300 coordinate structures of the form adj+nl+and+n2 and nl+and+n2+n3 from on-line CNN news using the method proposed by Mela and Fouquere (1996) and ran the algorithms on a computer.", | |
| "cite_spans": [ | |
| { | |
| "start": 235, | |
| "end": 259, | |
| "text": "Mela and Fouquere (1996)", | |
| "ref_id": "BIBREF10" | |
| } | |
| ], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Disambiguation for Structure of Coordinate Constructions", | |
| "sec_num": "4" | |
| }, | |
| { | |
| "text": "The result for disambiguating CCs of adj+nl+and+n2 is shown in Table 1 and that for nl+and+n2+n3 in Table 2 . The algorithms have adopted a back-off form to integrate different cues in the disambiguation process and the reliable cues with certainty are used first to achieve a better overall performance. As we can see from the results, the cues (orthographic forms, syntactic constraints, antonymy relation, observed patterns) with high success rates have comparatively low recall rates (from 3.7% to 40.3%); other cues, such as selectional restriction and semantic similarity, on the other hand, have comparatively high recall rates.", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 63, | |
| "end": 70, | |
| "text": "Table 1", | |
| "ref_id": "TABREF1" | |
| }, | |
| { | |
| "start": 100, | |
| "end": 107, | |
| "text": "Table 2", | |
| "ref_id": "TABREF2" | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Disambiguation for Structure of Coordinate Constructions", | |
| "sec_num": "4" | |
| }, | |
| { | |
| "text": "Evaluation Table 3 shows the results of the performance achieved by our method and those by others for the structure of nl+and+n2+n3. All these methods use the same data. Here, (1) shows the result obtained from attaching the modifier to the nearest head (Kimball, 1973) , i.e., (nl 4 S1 is a set which consists of ni and its synonyms extracted from the WordNet. Similarly, S2 is a set including n2 and its synonyms, and S3 is a set including n3 and its synonyms. and (n2 n3)); (2) shows the result of the method proposed by Resnik (1993) in which class-based similarity and a measure of noun-noun modification estimated from corpora are used in resolving ambiguous coordinations; (3) shows the performance of our method; and (4) is the performance of human judgment by three native speakers who were just presented the words of nl, coordinator, n2 and n3 without surrounding contexts. The lower bound and the upper bound on the performance of our method seem to be 65.3% scored by the simple heuristics of closest attachment (1) and 91.3% by human beings (4). We can clearly see here that the performance of our method (3) is better than those of (1) and 2, and is close to that of human beings. Table 4 shows the results achieved by our method and those by others for the structure of adj+nl+and+n2. Here, each row of (1), (3), and (4) is analogous to that in Table 3. (2) shows the result from estimating the strength of association between adj and n2 using the maximum mutual information over their classes (Resnik, 1993; Alves, 1996) . The performance by the closest attachment (1) is so poor that it would be unusable in any real applications. The method (2) using maximum MI performed better. But again ours is much better.", | |
| "cite_spans": [ | |
| { | |
| "start": 255, | |
| "end": 270, | |
| "text": "(Kimball, 1973)", | |
| "ref_id": "BIBREF7" | |
| }, | |
| { | |
| "start": 525, | |
| "end": 538, | |
| "text": "Resnik (1993)", | |
| "ref_id": "BIBREF13" | |
| }, | |
| { | |
| "start": 1511, | |
| "end": 1525, | |
| "text": "(Resnik, 1993;", | |
| "ref_id": "BIBREF13" | |
| }, | |
| { | |
| "start": 1526, | |
| "end": 1538, | |
| "text": "Alves, 1996)", | |
| "ref_id": "BIBREF0" | |
| } | |
| ], | |
| "ref_spans": [ | |
| { | |
| "start": 11, | |
| "end": 18, | |
| "text": "Table 3", | |
| "ref_id": "TABREF3" | |
| }, | |
| { | |
| "start": 1197, | |
| "end": 1204, | |
| "text": "Table 4", | |
| "ref_id": "TABREF4" | |
| }, | |
| { | |
| "start": 1362, | |
| "end": 1370, | |
| "text": "Table 3.", | |
| "ref_id": "TABREF3" | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Disambiguation for Structure of Coordinate Constructions", | |
| "sec_num": "4" | |
| }, | |
| { | |
| "text": "This is not to say our method is prefect, however. Selectional restrictions do not work well in idiomatic or fixed expressions (e.g., green peace) or when the adjective has multiple senses. Take (17), for instance, (17) I bought some soft balls and drinks in a drugstore.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Disambiguation for Structure of Coordinate Constructions", | |
| "sec_num": "4" | |
| }, | |
| { | |
| "text": "The algorithm 1 produces the parse (soft (balls and drinks)), but the correct one should be ((soft balls) and drinks).", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Disambiguation for Structure of Coordinate Constructions", | |
| "sec_num": "4" | |
| }, | |
| { | |
| "text": "The judgment made by semantic \u2022 similarity succeeded in about 80% of the cases (Table 2) , but often failed when the co-occurrences are low and especially when the words involved are polysemous. We see that we need to use a larger corpus to overcome this problem.", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 79, | |
| "end": 88, | |
| "text": "(Table 2)", | |
| "ref_id": "TABREF2" | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Disambiguation for Structure of Coordinate Constructions", | |
| "sec_num": "4" | |
| }, | |
| { | |
| "text": "Applications to Complex Coordinations and Nomial Compounds The method presented in this paper can be directly used for resolving complicated cases of coordinate structures (e.g., freshman training and personal management system). The coordinate structure of adj+ n.lefti+...+n_lefti +and+n_right i +...+n_rightk can be reduced to CCs of adj +n1 +and+n2, adj+n2 + and+ n3,---7 adj+nk _ 1 + and+nk . So we can apply algorithm 1 to disambiguate these constructions and integrate them to acquire the overall structure of the CC.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Discussion", | |
| "sec_num": "5" | |
| }, | |
| { | |
| "text": "The coordinate structure of n_lefti + ...+n_lefti + and + n_righti + ...+n_right k , on the other hand, can be quite complex. Theoretically, 1 or k may be quite a big number. We found that in the most of the cases (>99.3%), 1 is no greater than 2 and k is no greater than 3 in real texts, however. For a complex CC in form of nl+n2+and+n3+n4+n5, for instance, we can use semantic similarity defined in formulae (6) and (7) in Section 3 to find the word in the right hand side of the coordinator that is most similar to n2. Suppose n2 is similar to n4, we then check whether n1 co-occurs with n3 in pattern n1 n3 in the corpora, if the answer is yes, then produce (n1 (n2 and (n3 n4)) n5), otherwise produce (((n1 n2) and (n3 n4)) n5). Using this method, the analysis for CC freshman training and personal management system is (((freshmen training) and (personal menagement)) system); while for food handling and storage procedure the result turns to be ((food (handling and storage) ) procedure).", | |
| "cite_spans": [], | |
| "ref_spans": [ | |
| { | |
| "start": 953, | |
| "end": 982, | |
| "text": "((food (handling and storage)", | |
| "ref_id": null | |
| } | |
| ], | |
| "eq_spans": [], | |
| "section": "Discussion", | |
| "sec_num": "5" | |
| }, | |
| { | |
| "text": "The method can also be applied to analyzing structures of nominal compounds. Take the phrase novice song bird feeder kit, for instance. Selectional restriction may play an important role in judging which constituent the adjective novice refers to in the candidates song, bird, feeder, kit and their combinations. Co-occurrence relation, on the other hand, is crucial to determining the structure of a nominal compound like song bird feeder. We may collect statistics to see if song bird is observed more often than song feeder.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Discussion", | |
| "sec_num": "5" | |
| }, | |
| { | |
| "text": "Conclusion Resolving ambiguities in coordinate structure has a great importance for its applications to text understanding and machine translation. It is crucial to information retrieval, too (e.g., internet information retrieval). Given \"training system\" as a conjoined retrieval condition, for example, the phrase freshman training and personal management system in the target text can be retrieved using the method presented in this paper, but it can hardly be found with the retrieval techniques available so far.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Discussion", | |
| "sec_num": "5" | |
| }, | |
| { | |
| "text": "The disambiguation method proposed is scalable since it does not depend on any handcrafted rules. It is strong at the data sparseness problem also as we use co-occurrences between semantic classes, rather than words, extracted from a lexical database.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Discussion", | |
| "sec_num": "5" | |
| }, | |
| { | |
| "text": "The disambiguation experiment has proven that our method for disambiguating syntactic structures is valid, effective, and useful in practical applications. The performance of our method is significantly better than those of other work. We think that the performance can be improved further by using a larger corpus that contributes to the precision for estimating semantic similarities and/or selectional restrictions.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "Discussion", | |
| "sec_num": "5" | |
| }, | |
| { | |
| "text": "In this case, the coordination of stock and estate keeper is considered to be the reduced form of stock keeper and estate keeper.2 Hereafter, and in adj+nl+and+n2 and in nl\u00b1and-f-n2+n3 represents a coordinator such as and, or, comma, and the like.", | |
| "cite_spans": [], | |
| "ref_spans": [], | |
| "eq_spans": [], | |
| "section": "", | |
| "sec_num": null | |
| } | |
| ], | |
| "back_matter": [], | |
| "bib_entries": { | |
| "BIBREF0": { | |
| "ref_id": "b0", | |
| "title": "The Selection of the Most Probable Dependency Structure in Japanese Using Mutual Information", | |
| "authors": [ | |
| { | |
| "first": "E", | |
| "middle": [], | |
| "last": "Alves", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1996, | |
| "venue": "Proceedings of the 34th Annual Meeting of ACL", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "372--374", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Alves, E. (1996). \"The Selection of the Most Probable Dependency Structure in Japanese Using Mutual Information.\" Proceedings of the 34th Annual Meeting of ACL, 372-374.", | |
| "links": null | |
| }, | |
| "BIBREF1": { | |
| "ref_id": "b1", | |
| "title": "The Coordination of Unlike Categories", | |
| "authors": [ | |
| { | |
| "first": "S", | |
| "middle": [], | |
| "last": "Bayer", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1996, | |
| "venue": "Language", | |
| "volume": "72", | |
| "issue": "3", | |
| "pages": "579--616", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Bayer, S. (1996). \"The Coordination of Unlike Categories.\" Language, 72(3):579-616.", | |
| "links": null | |
| }, | |
| "BIBREF2": { | |
| "ref_id": "b2", | |
| "title": "Word Association Norms, Mutual Information, and Lexicography", | |
| "authors": [ | |
| { | |
| "first": "K", | |
| "middle": [ | |
| "W" | |
| ], | |
| "last": "Church", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "P", | |
| "middle": [], | |
| "last": "Hanks", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1990, | |
| "venue": "Computational Linguistics", | |
| "volume": "16", | |
| "issue": "1", | |
| "pages": "22--29", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Church, K. W. and Hanks, P. (1990). \"Word Association Norms, Mutual Information, and Lexi- cography.\" Computational Linguistics, 16(1):22-29.", | |
| "links": null | |
| }, | |
| "BIBREF3": { | |
| "ref_id": "b3", | |
| "title": "Coordination in Unification-based Grammars", | |
| "authors": [ | |
| { | |
| "first": "R", | |
| "middle": [ | |
| "P" | |
| ], | |
| "last": "Cooper", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1991, | |
| "venue": "Proceedings of the 29th Annual Meeting of ACL", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "167--172", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Cooper, R. P. (1991). \"Coordination in Unification-based Grammars.\" Proceedings of the 29th Annual Meeting of ACL, 167-172.", | |
| "links": null | |
| }, | |
| "BIBREF4": { | |
| "ref_id": "b4", | |
| "title": "Contextual Word Similarity and Estimation from Sparse Data", | |
| "authors": [ | |
| { | |
| "first": "I", | |
| "middle": [], | |
| "last": "Dagan", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "S", | |
| "middle": [], | |
| "last": "Marcus", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "S", | |
| "middle": [], | |
| "last": "Markovitch", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1995, | |
| "venue": "Computer Speech and Language", | |
| "volume": "9", | |
| "issue": "", | |
| "pages": "123--152", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Dagan, I.; Marcus, S.; and Markovitch S. (1995). \"Contextual Word Similarity and Estimation from Sparse Data.\" Computer Speech and Language, 9:123-152.", | |
| "links": null | |
| }, | |
| "BIBREF5": { | |
| "ref_id": "b5", | |
| "title": "Evaluation Techniques for Automatic Semantic Extraction: Comparing Syntactic and Window-based Approaches", | |
| "authors": [ | |
| { | |
| "first": "G", | |
| "middle": [], | |
| "last": "Grefenstette", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1993, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Grefenstette, G. (1993). \"Evaluation Techniques for Automatic Semantic Extraction: Comparing Syntactic and Window-based Approaches.\" Technical Report, Department of Computer Science, University of Pittsburgh.", | |
| "links": null | |
| }, | |
| "BIBREF6": { | |
| "ref_id": "b6", | |
| "title": "Structural Ambiguity and Lexical Relations", | |
| "authors": [ | |
| { | |
| "first": "D", | |
| "middle": [], | |
| "last": "Hindle", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "M", | |
| "middle": [], | |
| "last": "Rooth", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1993, | |
| "venue": "Computational Linguistics", | |
| "volume": "19", | |
| "issue": "1", | |
| "pages": "103--120", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Hindle, D. and Rooth, M. (1993). \"Structural Ambiguity and Lexical Relations.\" Computational Linguistics, 19(1):103-120.", | |
| "links": null | |
| }, | |
| "BIBREF7": { | |
| "ref_id": "b7", | |
| "title": "Seven Principles of Surface Structure Parsing in Natural Language", | |
| "authors": [ | |
| { | |
| "first": "S", | |
| "middle": [], | |
| "last": "Kimball", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1973, | |
| "venue": "Cognition", | |
| "volume": "2", | |
| "issue": "", | |
| "pages": "15--47", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Kimball, S. (1973). \"Seven Principles of Surface Structure Parsing in Natural Language.\" Cogni- tion, 2:15-47.", | |
| "links": null | |
| }, | |
| "BIBREF8": { | |
| "ref_id": "b8", | |
| "title": "Segmenting Narrative Text into Coherent Scenes", | |
| "authors": [ | |
| { | |
| "first": "H", | |
| "middle": [], | |
| "last": "Kozima", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "T", | |
| "middle": [], | |
| "last": "Furugori", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1994, | |
| "venue": "Literary and Linguistic Computing", | |
| "volume": "9", | |
| "issue": "", | |
| "pages": "13--19", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Kozima, H. and Furugori, T. (1994). \"Segmenting Narrative Text into Coherent Scenes.\" Literary and Linguistic Computing, 9(1):13-19.", | |
| "links": null | |
| }, | |
| "BIBREF9": { | |
| "ref_id": "b9", | |
| "title": "A Syntactic Analysis Method of Long Japanese Sentences Based on the Detection of Conjunctive Structures", | |
| "authors": [ | |
| { | |
| "first": "S", | |
| "middle": [], | |
| "last": "Kurohashi", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "M", | |
| "middle": [], | |
| "last": "Nagao", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1994, | |
| "venue": "Computational Linguistics", | |
| "volume": "20", | |
| "issue": "4", | |
| "pages": "507--534", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Kurohashi, S. and Nagao, M. (1994): \"A Syntactic Analysis Method of Long Japanese Sentences Based on the Detection of Conjunctive Structures.\" Computational Linguistics, 20(4):507-534.", | |
| "links": null | |
| }, | |
| "BIBREF10": { | |
| "ref_id": "b10", | |
| "title": "Coordination as a Direct Process", | |
| "authors": [ | |
| { | |
| "first": "A", | |
| "middle": [], | |
| "last": "Mela", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "C", | |
| "middle": [], | |
| "last": "Fouquere", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1996, | |
| "venue": "Proceedings of the 34th ACL Meeting", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "124--130", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Mela, A. and Fouquere, C. (1996). \"Coordination as a Direct Process.\" Proceedings of the 34th ACL Meeting, 124-130.", | |
| "links": null | |
| }, | |
| "BIBREF11": { | |
| "ref_id": "b11", | |
| "title": "WordNet: An On-Line Lexical Database", | |
| "authors": [ | |
| { | |
| "first": "G", | |
| "middle": [], | |
| "last": "Miller", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1990, | |
| "venue": "International Journal of Lexicograph", | |
| "volume": "3", | |
| "issue": "4", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Miller, G. (1990). \"WordNet: An On-Line Lexical Database.\" International Journal of Lexico- graph, 3(4) (special issue).", | |
| "links": null | |
| }, | |
| "BIBREF12": { | |
| "ref_id": "b12", | |
| "title": "Constituent Coordination in HPSG", | |
| "authors": [ | |
| { | |
| "first": "M", | |
| "middle": [], | |
| "last": "Paritong", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1992, | |
| "venue": "KONVENS", | |
| "volume": "92", | |
| "issue": "", | |
| "pages": "228--237", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Paritong, M. (1992). \"Constituent Coordination in HPSG.\" KONVENS 92, 228-237, Springer Verlag.", | |
| "links": null | |
| }, | |
| "BIBREF13": { | |
| "ref_id": "b13", | |
| "title": "Selection and Information: A Class-Based Approach to Lexical Relationships", | |
| "authors": [ | |
| { | |
| "first": "P", | |
| "middle": [ | |
| "S" | |
| ], | |
| "last": "Resnik", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1993, | |
| "venue": "Doctoral Dissertation", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Resnik, P. S. (1993). \"Selection and Information: A Class-Based Approach to Lexical Relation- ships.\" Doctoral Dissertation, University of Pennsylvania, Philadelphia, P.A.", | |
| "links": null | |
| }, | |
| "BIBREF14": { | |
| "ref_id": "b14", | |
| "title": "Parsing English with a Link Grammar", | |
| "authors": [ | |
| { | |
| "first": "D", | |
| "middle": [], | |
| "last": "Sleater", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "D", | |
| "middle": [], | |
| "last": "Temperly", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1991, | |
| "venue": "", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Sleater, D. and Temperly, D. (1991). \"Parsing English with a Link Grammar.\" Carnegie Mellon University Computer Science technical report CMU-CS-91-196.", | |
| "links": null | |
| }, | |
| "BIBREF15": { | |
| "ref_id": "b15", | |
| "title": "A Hybrid Disambiguation Model for Prepositional Phrase Attachment", | |
| "authors": [ | |
| { | |
| "first": "H", | |
| "middle": [], | |
| "last": "Wu", | |
| "suffix": "" | |
| }, | |
| { | |
| "first": "T", | |
| "middle": [], | |
| "last": "Furugori", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1996, | |
| "venue": "Literary and Linguistic Computing", | |
| "volume": "11", | |
| "issue": "4", | |
| "pages": "187--192", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Wu, H. and Furugori, T. (1996). \"A Hybrid Disambiguation Model for Prepositional Phrase Attachment.\" Literary and Linguistic Computing, 11(4), 187-192.", | |
| "links": null | |
| }, | |
| "BIBREF16": { | |
| "ref_id": "b16", | |
| "title": "A Hybrid Model for Resolving Syntactic Ambiguities in Natural Language Processing", | |
| "authors": [ | |
| { | |
| "first": "H", | |
| "middle": [], | |
| "last": "Wu", | |
| "suffix": "" | |
| } | |
| ], | |
| "year": 1997, | |
| "venue": "Doctoral Dissertation, the University of Electro-Communications", | |
| "volume": "", | |
| "issue": "", | |
| "pages": "", | |
| "other_ids": {}, | |
| "num": null, | |
| "urls": [], | |
| "raw_text": "Wu, H. (1997). \"A Hybrid Model for Resolving Syntactic Ambiguities in Natural Language Pro- cessing.\" Doctoral Dissertation, the University of Electro-Communications, Tokyo, Japan.", | |
| "links": null | |
| } | |
| }, | |
| "ref_entries": { | |
| "FIGREF0": { | |
| "uris": null, | |
| "num": null, | |
| "type_str": "figure", | |
| "text": "(14) young nations and superpowers (15) finance and market investigation" | |
| }, | |
| "TABREF0": { | |
| "type_str": "table", | |
| "text": "160,000 sentences with annotated morphological, syntactic and semantic information. The Brown Corpus was compiled in the early 1960s at Brown University, USA, under the direction of W. Nelson Francis and Henry Kucera. It contains 500 text samples representing 15 categories of American English texts printed in 1961.", | |
| "num": null, | |
| "html": null, | |
| "content": "<table><tr><td>min(I(w, w1),I(w, w2))</td></tr></table>" | |
| }, | |
| "TABREF1": { | |
| "type_str": "table", | |
| "text": "Experimental Results on Testing adj+nl+and+n2", | |
| "num": null, | |
| "html": null, | |
| "content": "<table><tr><td>Step</td><td colspan=\"2\">Recall (%) I Number</td><td>Accuracy</td></tr><tr><td>(1) directly observed pattern</td><td>38.0</td><td>114</td><td>100.0%</td></tr><tr><td>(2) indirectly observed pattern</td><td>40.3</td><td>75</td><td>86.7%</td></tr><tr><td>(3) selectional restriction</td><td>46.8</td><td>52</td><td>86.5%</td></tr><tr><td>(4) default</td><td>100.0.</td><td>59</td><td>66.7%</td></tr><tr><td>Total</td><td>100.0 I</td><td>300</td><td>87.7%</td></tr></table>" | |
| }, | |
| "TABREF2": { | |
| "type_str": "table", | |
| "text": "", | |
| "num": null, | |
| "html": null, | |
| "content": "<table><tr><td colspan=\"3\">: Experimental Results on Testing nl+and+n2+n3</td><td/></tr><tr><td>Step</td><td>Recall (%)</td><td>Number</td><td>Accuracy</td></tr><tr><td>(1) orthographic form</td><td>3.7</td><td>11</td><td>100.0%</td></tr><tr><td>(2) antonym</td><td>8.3</td><td>24</td><td>95.2%</td></tr><tr><td>(3) similarity in form</td><td>38.8</td><td>104</td><td>91.4%</td></tr><tr><td>(4) semantic similarity</td><td>100</td><td>159</td><td>79.9%</td></tr><tr><td>Total</td><td>100.0</td><td>300 I</td><td>85.3%</td></tr></table>" | |
| }, | |
| "TABREF3": { | |
| "type_str": "table", | |
| "text": "Comparison with Other Work for Determining nl+and+n2+n3", | |
| "num": null, | |
| "html": null, | |
| "content": "<table><tr><td>Method</td><td>Success Rate</td></tr><tr><td>(1) Closest attachment</td><td>65:3%</td></tr><tr><td>(2) Resnik's method</td><td>80.7%</td></tr><tr><td>(3) Our method</td><td>85.3%</td></tr><tr><td>(4) Average human</td><td>91.7%</td></tr></table>" | |
| }, | |
| "TABREF4": { | |
| "type_str": "table", | |
| "text": "Comparison with Other Work for Determining adj+nl+and+n2", | |
| "num": null, | |
| "html": null, | |
| "content": "<table><tr><td>Method</td><td>Success Rate</td></tr><tr><td>(1) Closest attachment</td><td>64.7%</td></tr><tr><td>(2) Maximum MI</td><td>82.3%</td></tr><tr><td>(3) Our method</td><td>87.7%</td></tr><tr><td>. (4) Average human</td><td>93.3%</td></tr></table>" | |
| } | |
| } | |
| } | |
| } |