acl acl2013 acl2013-280 knowledge-graph by maker-knowledge-mining

280 acl-2013-Plurality, Negation, and Quantification:Towards Comprehensive Quantifier Scope Disambiguation


Source: pdf

Author: Mehdi Manshadi ; Daniel Gildea ; James Allen

Abstract: Recent work on statistical quantifier scope disambiguation (QSD) has improved upon earlier work by scoping an arbitrary number and type of noun phrases. No corpusbased method, however, has yet addressed QSD when incorporating the implicit universal of plurals and/or operators such as negation. In this paper we report early, though promising, results for automatic QSD when handling both phenomena. We also present a general model for learning to build partial orders from a set of pairwise preferences. We give an n log n algorithm for finding a guaranteed approximation of the optimal solution, which works very well in practice. Finally, we significantly improve the performance of the pre- vious model using a rich set of automatically generated features.

Reference: text


Summary: the most important sentenses genereted by tfidf model

sentIndex sentText sentNum sentScore

1 edu Abstract Recent work on statistical quantifier scope disambiguation (QSD) has improved upon earlier work by scoping an arbitrary number and type of noun phrases. [sent-3, score-0.739]

2 No corpusbased method, however, has yet addressed QSD when incorporating the implicit universal of plurals and/or operators such as negation. [sent-4, score-0.44]

3 We also present a general model for learning to build partial orders from a set of pairwise preferences. [sent-6, score-0.15]

4 1 Introduction The sentence there is one faculty member in every graduate committee is ambiguous with respect to quantifier scoping, since there are at least two possible readings: If one has wide scope, there is a unique faculty member on every committee. [sent-9, score-0.231]

5 Over the past decade there has been some work on statistical quantifier scope disambiguation (QSD) (Higgins and Sadock, 2003; Galen and MacCartney, 2004; Manshadi and Allen, 2011a). [sent-11, score-0.345]

6 Third, QSD has often been con- sidered only in the context of explicit quantification such as each and every versus some and a/an. [sent-16, score-0.147]

7 For example, Higgins and Sadock (2003) find fewer than 1000 sentences with two or more explicit quantifiers in the Wall Street journal section of Penn Treebank. [sent-18, score-0.145]

8 Furthermore, for more than 60% ofthose sentences, the order ofthe quantifiers does not matter, either as a result of the logical equivalence (as in two existentials), or because they do not have any scope interaction. [sent-19, score-0.293]

9 A deep understanding of this sentence, requires deciding whether each word in the set, referred to by Three words, starts with a potentially distinct capital letter (as in Apple, Orange, Banana) or there is a unique capital letter which each word starts with (as in Apple, Adam, Athens). [sent-28, score-0.393]

10 In general, every plural NP potentially introduces an implicit universal, ranging 1For example, Liang et al. [sent-30, score-0.339]

11 (2011) in their state-of-the-art statistical semantic parser within the domain of natural language queries to databases, explicitly devise quantifier scoping in the semantic model. [sent-31, score-0.515]

12 While explicit universals may not occur very often in natural language, the usage of plurals is very common. [sent-36, score-0.272]

13 , Morante and Blanco, 2012) has investigated automatically detecting the scope and focus of negation. [sent-42, score-0.198]

14 However, the scope of negation with respect to quantifiers is a different phenomenon. [sent-43, score-0.46]

15 Transforming this sentence into a meaning representation language, for almost any practical purposes, requires deciding whether the NP a capital letter lies in the scope of the negation or outside of it. [sent-47, score-0.509]

16 The former describes the preferred reading where The word starts with a lowercase letter as in apple, orange, banana, but the latter gives the unlikely reading, according to which there exists a particular capital letter, say A, that The word starts with, as in apple, Orange, Banana. [sent-48, score-0.218]

17 By not involving negation in quantifier scoping, a semantic parser may produce an unintended interpretation. [sent-49, score-0.314]

18 Their system disambiguates the scope of exactly two explicitly quantified NPs in a sentence, ignoring indefinite a/an, definites and bare NPs. [sent-52, score-0.238]

19 Manshadi and Allen (201 1a), hence MA11, go beyond those limitations and scope an arbitrary number of NPs in a sentence with no restriction on the type ofquantification. [sent-53, score-0.198]

20 However, although their corpus annotates the scope of negations and the implicit universal of plurals, their QSD system does not handle those. [sent-54, score-0.596]

21 As a step towards comprehensive automatic QSD, in this paper we present our work on automatic scoping of the implicit universal of plurals and negations. [sent-55, score-0.805]

22 The new revision, called QuanText, carries a more detailed, fine-grained scope annotation (Manshadi et al. [sent-58, score-0.198]

23 The performance of 2Although plurals carry different types of quantification (Herbelot and Copestake, 2010), almost always there exists an implicit universal. [sent-60, score-0.342]

24 The importance of scoping this universal, however, may vary based on the type of quantification. [sent-61, score-0.368]

25 2 Task definition In QuanText, scope-bearing elements (or, as we call them, scopal terms) of each sentence have been identified using labeled chunks, as in (3). [sent-72, score-0.253]

26 Outscoping relations are used to specify the relative scope of scopal terms. [sent-76, score-0.39]

27 The relation i > j means that chunk ioutscopes (or has wide scope over) chunk j. [sent-77, score-0.51]

28 Equivalently, chunk j is said to have narrow scope with respect to i. [sent-78, score-0.354]

29 Each sentence is annotated with its most preferred scoping (according to the annotators’ judgement), represented as a partial order: 4. [sent-79, score-0.431]

30 This happens ifboth orders are equivalent (as in two existentials) or when the two chunks have no scope interaction. [sent-81, score-0.366]

31 For example, G1 in Figure 1 represents the scoping in (4). [sent-83, score-0.368]

32 In order to take the transitivity of outscoping relations into account, we use the transitive closure (TC) of DAGs. [sent-86, score-0.27]

33 A disadvantage of this metric is that it gives the same weight to outscoping and incomparability relations. [sent-91, score-0.251]

34 In practice, if two scopal terms with equivalent ordering (and hence, no outscoping relation) are incorrectly labeled with an outscoping, the logical form still remains valid. [sent-92, score-0.393]

35 But if an outscoping relation is mislabeled, it will change the interpretation ofthe sentence. [sent-93, score-0.201]

36 Therefore, in MA1 1, we suggest defining a precision/recall based on the number of outscoping relations recovered correctly: 4 P+=|E+p|E∩p+ E|g+|, R+=|Ep+|E∩g+ E|+g| 3 (u, v) (2) ∨ ∈ G+ ⇐⇒ ((u, v) ∈ G ∃∈w 1G . [sent-94, score-0.201]

37 1 Our framework Learning to do QSD Since we defined QSD as a partial ordering, automatic QSD would become the problem of learning to build partial orders. [sent-107, score-0.126]

38 Learning to build partial orders has not attracted much attention in the learning community, although as seen shortly, the techniques developed for ranking can be adopted for learning to build partial orders. [sent-114, score-0.171]

39 Although there could be many DAGs representing a partial order P, only one of those is a transitive DAG. [sent-116, score-0.132]

40 Every non-transitive DAG will be converted into its transitive counterpart by taking its transitive closure (as shown in Figure 1). [sent-118, score-0.138]

41 Therefore, we have reduced the problem of predicting a partial order to pairwise comparison, analogous to ranking by pairwise comparison or RPC (Hullermeier et al. [sent-136, score-0.147]

42 The difference though is that in RPC, the comparison is a (soft) binary classification, while for partial orders we have the case of incomparability (the label ? [sent-138, score-0.158]

43 T−h)e, proof −is) beyond the scope of this paper, but the idea is similar to that of Cohen et al. [sent-152, score-0.198]

44 2 Dealing with plurality and negation Consider the following sentence with the plural NP chunk the lines. [sent-202, score-0.5]

45 SI : (1c > 1d > 2; 1d > 3) 10 In QuanText, plural chunks are indexed with a number followed by the lowercase letter “p”. [sent-206, score-0.308]

46 As seen in (6), the scoping looks different from before in that the terms 1d and 1c are not the label of any chunk. [sent-207, score-0.368]

47 These two terms refer to the two quantified terms introduced by the plural chunk 1p: 1c (for collection) represents the set (or in better words collection) ofentities, defined by the plural, and 1d (for distribution) refers to the implicit universal, introduced by the plural. [sent-208, score-0.493]

48 In other words, for a plural chunk ip, id represents the universally quantified entity over the collection ic. [sent-209, score-0.351]

49 The outscoping relation 1d > 2 in (6) states that every line in the collection, denoted by 1c, starts with its own punctuation character. [sent-210, score-0.28]

50 In QuanText, negation chunks are labeled with an uppercase “N” followed by a number. [sent-214, score-0.29]

51 txt”], which starts with [3/ a capital letter], but does [N1/ not] end with [4/ a capital letter]. [sent-218, score-0.193]

52 SI : (2 > 1> 3; 1> N1 > 4) As seen here, a negation simply introduces a chunk, which participates in outscoping relations like an NP chunk. [sent-220, score-0.368]

53 Figure 4(b) represents the scoping in (8) as a DAG. [sent-221, score-0.368]

54 From these examples, as long as we create two nodes in the DAG corresponding to each plural chunk, and one node corresponding to each negation, there is no need to modify the underlying model (defined in the previous section). [sent-222, score-0.233]

55 However, when u (or v) is a negation (Ni) or an implicit universal (id) node, the probabilities puλ,v (λ ∈ {+, ? [sent-223, score-0.471]

56 3 −, Feature selection Previous work has shown that the lexical item of quantifiers and syntactic clues (often extracted from phrase structure trees) are good at predicting quantifier scoping. [sent-229, score-0.242]

57 Another point to mention here is that the features that are predictive of the relative scope of quantifiers are not necessarily as helpful when determining the scope ofnegation and vice versa. [sent-235, score-0.491]

58 Therefore we do not use exactly the same set of features when 68 one of the scopal terms in the pair11 is a negation, although most of the features are quite similar. [sent-236, score-0.192]

59 1 NP chunks We first describe the set of features we have adopted when both scopal terms in a pair are NPchunks. [sent-239, score-0.315]

60 a Implicit universal of a plural Remember that every plural chunk i introduces two nodes in the DAG, ic and id. [sent-246, score-0.619]

61 Both nodes are introduced by the same chunk i, therefore they use the same set of features. [sent-247, score-0.213]

62 The only exception is a single additional binary feature for plural NP • chunks, which determines whether the given node refers to the implicit universal of the plural (i. [sent-248, score-0.599]

63 In the following, by chunk we mean the deepest phrase-level node in T dominating all the words in the chunk. [sent-259, score-0.213]

64 The choice of the word noun, however, may be sloppy, as the head of an NP chunk may not be a noun. [sent-269, score-0.156]

65 Therefore, for every pair of nodes with one negation, the other node must re- fer to an NP chunk. [sent-277, score-0.156]

66 • Lexical item of the determiner for the NP chunk •• LDoexeisc tahle i eNmP cohf tuhnek dceotnetrmainin ae cr ofonsr tthanet? [sent-279, score-0.186]

67 Of all the NP chunks, 320 (more than 18%) are plural, each introducing an implicit universal, that is, an additional node in the DAG. [sent-301, score-0.235]

68 Therefore, a sentence with n scopal elements creates C(n, 2) = n(n 1)/2 samples for classifcirceaatiteosn. [sent-303, score-0.253]

69 C W(nh,e2n) a =ll nth(en e −lem 1)e/n2ts s are tlaekse fno rin ctloa account,12 the total number of samples in the corpus will be: − 12Here by all elements we mean explicit chunks and the implicit universals. [sent-304, score-0.412]

70 In particular, some nouns introduce two entities: a type and a 69 XC(ni,2) ≈ 4500 (9) Xi Where ni is the number of scopal terms introduced by sentence i. [sent-306, score-0.192]

71 Out of the 4500 samples, around 1800 involve at least one implicit universal (i. [sent-307, score-0.304]

72 We evaluate the performance ofthe system for implicit universals and negation both separately and in the context of full scope disambiguation. [sent-310, score-0.656]

73 14 In order to run the baseline system on implicit universals, we take the feature vector of a plural NP and add a feature to indicate that this feature vector represents the implicit universal of the corresponding chunk. [sent-317, score-0.601]

74 Similarly, for negation we add a feature to show that the chunk represents a negation. [sent-318, score-0.323]

75 13Since the percentage of sentences with negation is small, we made sure that those sentences are distributed uniformly between three sets. [sent-328, score-0.167]

76 However, similar to plurals, NP conjunctions (disjunctions) introduce an extra scopal element: a universal (existential). [sent-333, score-0.318]

77 457 (a) Scoping explicit NP chunks BnO aevusgreamrtlainoldne (salMyn(sARdt1eiPm C)p-S(liVcnMtcl-u1nd3ivn)egrsal)0σ. [sent-341, score-0.173]

78 A457 (b) Scoping all elements (including id and Ni) Figure 5: Performance on QuanText data tem on the task that it was actually defined to perform (that is scoping only explicit NP chunks). [sent-347, score-0.515]

79 A sentence counts as correct iff every pair of scopal elements has been labeled correctly. [sent-350, score-0.332]

80 In computing AR, a sentence counts as correct iff all the outscoping relations have been recovered correctly in other words, iff R = 100%, regardless – of the value of P. [sent-354, score-0.275]

81 AR may be more practically meaningful, because if in the correct scoping of the sentence there is no outscoping between two elements, inserting one does not affect the interpretation ofthe sentence. [sent-355, score-0.569]

82 Figure 5(b) gives the performance of the overall model when all the elements including the implicit universals and the negations are taken into account. [sent-357, score-0.446]

83 042 higher than F-score for the first indicates that scoping implicit universals and/or negations must be easier than scoping explicit NP chunks. [sent-359, score-1.171]

84 In order to find how much one or both ofthe two elements contribute to this gain, we have run two more experiments, scoping only the pairs with at least one implicit universal and pairs with one negation, respectively. [sent-360, score-0.733]

85 As seen, the contribution in boosting the overall performance comes from the implicit universals while negations, in fact, lower the performance. [sent-362, score-0.291]

86 The performance for pairs with implicit universal is higher because universals, in general, 70 wIBOmiuatshrpealmitcnoieldt e(ualMsn(AtiRov1Pen1Cer)-sSiaVdlM)s-o1n3l)y(pairs0P. [sent-363, score-0.304]

87 F65+2394 (a) Pairs with at least one implicit universal (b) Pairs with at least one negation Figure 6: Implicit universals and negations are easier to scope, even for the human annotators. [sent-369, score-0.678]

88 Scoping a negation relative to an NP chunk, with which it has a long distance dependency, often depends on the scope of the elements in between. [sent-373, score-0.426]

89 Third, scoping negation usually requires a deep semantic analysis. [sent-374, score-0.566]

90 In order to see how well our approximation algorithm is working, similar to the approach of Chambers and Jurafsky (2008), we tried an ILP solver18 for DAGs with at most 8 nodes to find the optimum solution, but we found the difference insignificant. [sent-375, score-0.137]

91 19 5 Related work Since automatic QSD is in general challenging, traditionally quantifier scoping is left underspecified in deep linguistic processing systems (Alshawi and Crouch, 1992; Bos, 1996; Copestake et al. [sent-377, score-0.546]

92 In or17Trivially, we have taken the relation outscoping ic > id for granted and not counted it towards higher performance. [sent-382, score-0.237]

93 der to evade scope disambiguation, yet be able to perform entailment, Koller and Thater (2010) propose an algorithm to calculate the weakest readings20 from a scope-underspecified representation. [sent-389, score-0.224]

94 The first three only scope two scopal terms in a sentence, where the scopal term is an NP with an explicit quantification. [sent-392, score-0.632]

95 MA1 1 is the first to scope any number of NPs in a sentence with no restric- tion on the type of quantification. [sent-393, score-0.198]

96 Besides ignoring negation and implicit universals, their system has some other limitations too. [sent-394, score-0.345]

97 6 Summary and future work We develop the first statistical QSD model addressing the interaction of quantifiers with negation and the implicit universal of plurals, defining a baseline for this task on QuanText data (Manshadi et al. [sent-399, score-0.566]

98 This work can be improved in many directions, among which are scoping more elements such as other scopal operators and implicit entities, deploying more complex learning models, and de- veloping models which require less supervision. [sent-402, score-0.826]

99 Regular tree grammars as a formalism for scope underspecification. [sent-490, score-0.198]

100 Quantifier scope disambiguation using extracted pragmatic knowledge: preliminary results. [sent-523, score-0.198]


similar papers computed by tfidf model

tfidf for this paper:

wordName wordTfidf (topN-words)

[('qsd', 0.435), ('scoping', 0.368), ('manshadi', 0.217), ('outscoping', 0.201), ('scope', 0.198), ('scopal', 0.192), ('implicit', 0.178), ('negation', 0.167), ('dag', 0.164), ('chunk', 0.156), ('quantifier', 0.147), ('quantext', 0.134), ('np', 0.131), ('universal', 0.126), ('chunks', 0.123), ('plural', 0.119), ('universals', 0.113), ('plurals', 0.109), ('quantifiers', 0.095), ('negations', 0.094), ('dags', 0.089), ('hullermeier', 0.084), ('allen', 0.079), ('capital', 0.078), ('transitive', 0.069), ('higgins', 0.068), ('ddooeess', 0.067), ('furnkranz', 0.067), ('rpc', 0.067), ('sadock', 0.067), ('tdag', 0.067), ('letter', 0.066), ('partial', 0.063), ('elements', 0.061), ('mehdi', 0.059), ('plurality', 0.058), ('nodes', 0.057), ('node', 0.057), ('xv', 0.056), ('td', 0.055), ('quantification', 0.055), ('incomparability', 0.05), ('explicit', 0.05), ('nps', 0.049), ('approximation', 0.046), ('orders', 0.045), ('pairwise', 0.042), ('every', 0.042), ('apple', 0.041), ('iem', 0.041), ('srinivasan', 0.041), ('quantified', 0.04), ('koller', 0.038), ('starts', 0.037), ('galen', 0.037), ('iff', 0.037), ('marneffe', 0.036), ('id', 0.036), ('entailed', 0.035), ('optimum', 0.034), ('tthhee', 0.034), ('copestake', 0.034), ('agr', 0.033), ('dwoheast', 0.033), ('eexxiiccaall', 0.033), ('existentials', 0.033), ('eyke', 0.033), ('hamm', 0.033), ('idfo', 0.033), ('oseos', 0.033), ('otehes', 0.033), ('soes', 0.033), ('tteemm', 0.033), ('tympeed', 0.033), ('typed', 0.032), ('deep', 0.031), ('ar', 0.031), ('opt', 0.031), ('orange', 0.03), ('banana', 0.03), ('cohf', 0.03), ('ternary', 0.03), ('herbelot', 0.03), ('pu', 0.029), ('solution', 0.029), ('operators', 0.027), ('readings', 0.027), ('untyped', 0.027), ('graph', 0.026), ('noun', 0.026), ('iinn', 0.026), ('alshawi', 0.026), ('weakest', 0.026), ('pv', 0.026), ('cohen', 0.025), ('hhee', 0.024), ('topological', 0.024), ('comprehensive', 0.024), ('maccartney', 0.024), ('dependency', 0.024)]

similar papers list:

simIndex simValue paperId paperTitle

same-paper 1 0.99999964 280 acl-2013-Plurality, Negation, and Quantification:Towards Comprehensive Quantifier Scope Disambiguation

Author: Mehdi Manshadi ; Daniel Gildea ; James Allen

Abstract: Recent work on statistical quantifier scope disambiguation (QSD) has improved upon earlier work by scoping an arbitrary number and type of noun phrases. No corpusbased method, however, has yet addressed QSD when incorporating the implicit universal of plurals and/or operators such as negation. In this paper we report early, though promising, results for automatic QSD when handling both phenomena. We also present a general model for learning to build partial orders from a set of pairwise preferences. We give an n log n algorithm for finding a guaranteed approximation of the optimal solution, which works very well in practice. Finally, we significantly improve the performance of the pre- vious model using a rich set of automatically generated features.

2 0.11280052 149 acl-2013-Exploring Word Order Universals: a Probabilistic Graphical Model Approach

Author: Xia Lu

Abstract: In this paper we propose a probabilistic graphical model as an innovative framework for studying typological universals. We view language as a system and linguistic features as its components whose relationships are encoded in a Directed Acyclic Graph (DAG). Taking discovery of the word order universals as a knowledge discovery task we learn the graphical representation of a word order sub-system which reveals a finer structure such as direct and indirect dependencies among word order features. Then probabilistic inference enables us to see the strength of such relationships: given the observed value of one feature (or combination of features), the probabilities of values of other features can be calculated. Our model is not restricted to using only two values of a feature. Using imputation technique and EM algorithm it can handle missing val- ues well. Model averaging technique solves the problem of limited data. In addition the incremental and divide-and-conquer method addresses the areal and genetic effects simultaneously instead of separately as in Daumé III and Campbell (2007). 1

3 0.097024657 229 acl-2013-Leveraging Synthetic Discourse Data via Multi-task Learning for Implicit Discourse Relation Recognition

Author: Man Lan ; Yu Xu ; Zhengyu Niu

Abstract: To overcome the shortage of labeled data for implicit discourse relation recognition, previous works attempted to automatically generate training data by removing explicit discourse connectives from sentences and then built models on these synthetic implicit examples. However, a previous study (Sporleder and Lascarides, 2008) showed that models trained on these synthetic data do not generalize very well to natural (i.e. genuine) implicit discourse data. In this work we revisit this issue and present a multi-task learning based system which can effectively use synthetic data for implicit discourse relation recognition. Results on PDTB data show that under the multi-task learning framework our models with the use of the prediction of explicit discourse connectives as auxiliary learning tasks, can achieve an averaged F1 improvement of 5.86% over baseline models.

4 0.09153676 41 acl-2013-Aggregated Word Pair Features for Implicit Discourse Relation Disambiguation

Author: Or Biran ; Kathleen McKeown

Abstract: We present a reformulation of the word pair features typically used for the task of disambiguating implicit relations in the Penn Discourse Treebank. Our word pair features achieve significantly higher performance than the previous formulation when evaluated without additional features. In addition, we present results for a full system using additional features which achieves close to state of the art performance without resorting to gold syntactic parses or to context outside the relation.

5 0.089747973 189 acl-2013-ImpAr: A Deterministic Algorithm for Implicit Semantic Role Labelling

Author: Egoitz Laparra ; German Rigau

Abstract: This paper presents a novel deterministic algorithm for implicit Semantic Role Labeling. The system exploits a very simple but relevant discursive property, the argument coherence over different instances of a predicate. The algorithm solves the implicit arguments sequentially, exploiting not only explicit but also the implicit arguments previously solved. In addition, we empirically demonstrate that the algorithm obtains very competitive and robust performances with respect to supervised approaches that require large amounts of costly training data.

6 0.073954724 372 acl-2013-Using CCG categories to improve Hindi dependency parsing

7 0.062610619 131 acl-2013-Dual Training and Dual Prediction for Polarity Classification

8 0.06201414 368 acl-2013-Universal Dependency Annotation for Multilingual Parsing

9 0.056848712 291 acl-2013-Question Answering Using Enhanced Lexical Semantic Models

10 0.055630133 227 acl-2013-Learning to lemmatise Polish noun phrases

11 0.054681376 274 acl-2013-Parsing Graphs with Hyperedge Replacement Grammars

12 0.052271407 144 acl-2013-Explicit and Implicit Syntactic Features for Text Classification

13 0.05188597 125 acl-2013-Distortion Model Considering Rich Context for Statistical Machine Translation

14 0.049734585 247 acl-2013-Modeling of term-distance and term-occurrence information for improving n-gram language model performance

15 0.049285654 44 acl-2013-An Empirical Examination of Challenges in Chinese Parsing

16 0.048572063 205 acl-2013-Joint Apposition Extraction with Syntactic and Semantic Constraints

17 0.04838388 80 acl-2013-Chinese Parsing Exploiting Characters

18 0.048153732 28 acl-2013-A Unified Morpho-Syntactic Scheme of Stanford Dependencies

19 0.046479248 166 acl-2013-Generalized Reordering Rules for Improved SMT

20 0.043856341 176 acl-2013-Grounded Unsupervised Semantic Parsing


similar papers computed by lsi model

lsi for this paper:

topicId topicWeight

[(0, 0.142), (1, -0.006), (2, -0.046), (3, -0.036), (4, -0.059), (5, 0.035), (6, 0.01), (7, 0.007), (8, 0.02), (9, 0.023), (10, 0.021), (11, 0.001), (12, -0.026), (13, -0.0), (14, -0.054), (15, -0.019), (16, 0.037), (17, 0.01), (18, -0.066), (19, -0.033), (20, 0.0), (21, 0.048), (22, 0.002), (23, -0.035), (24, -0.002), (25, 0.017), (26, -0.012), (27, 0.013), (28, -0.005), (29, -0.031), (30, -0.005), (31, -0.039), (32, 0.031), (33, 0.035), (34, 0.047), (35, -0.028), (36, -0.039), (37, -0.083), (38, -0.015), (39, 0.033), (40, 0.016), (41, -0.056), (42, 0.016), (43, -0.024), (44, -0.045), (45, -0.042), (46, 0.024), (47, -0.013), (48, 0.002), (49, -0.078)]

similar papers list:

simIndex simValue paperId paperTitle

same-paper 1 0.89560366 280 acl-2013-Plurality, Negation, and Quantification:Towards Comprehensive Quantifier Scope Disambiguation

Author: Mehdi Manshadi ; Daniel Gildea ; James Allen

Abstract: Recent work on statistical quantifier scope disambiguation (QSD) has improved upon earlier work by scoping an arbitrary number and type of noun phrases. No corpusbased method, however, has yet addressed QSD when incorporating the implicit universal of plurals and/or operators such as negation. In this paper we report early, though promising, results for automatic QSD when handling both phenomena. We also present a general model for learning to build partial orders from a set of pairwise preferences. We give an n log n algorithm for finding a guaranteed approximation of the optimal solution, which works very well in practice. Finally, we significantly improve the performance of the pre- vious model using a rich set of automatically generated features.

2 0.65205467 41 acl-2013-Aggregated Word Pair Features for Implicit Discourse Relation Disambiguation

Author: Or Biran ; Kathleen McKeown

Abstract: We present a reformulation of the word pair features typically used for the task of disambiguating implicit relations in the Penn Discourse Treebank. Our word pair features achieve significantly higher performance than the previous formulation when evaluated without additional features. In addition, we present results for a full system using additional features which achieves close to state of the art performance without resorting to gold syntactic parses or to context outside the relation.

3 0.64114153 367 acl-2013-Universal Conceptual Cognitive Annotation (UCCA)

Author: Omri Abend ; Ari Rappoport

Abstract: Syntactic structures, by their nature, reflect first and foremost the formal constructions used for expressing meanings. This renders them sensitive to formal variation both within and across languages, and limits their value to semantic applications. We present UCCA, a novel multi-layered framework for semantic representation that aims to accommodate the semantic distinctions expressed through linguistic utterances. We demonstrate UCCA’s portability across domains and languages, and its relative insensitivity to meaning-preserving syntactic variation. We also show that UCCA can be effectively and quickly learned by annotators with no linguistic background, and describe the compilation of a UCCAannotated corpus.

4 0.63844371 149 acl-2013-Exploring Word Order Universals: a Probabilistic Graphical Model Approach

Author: Xia Lu

Abstract: In this paper we propose a probabilistic graphical model as an innovative framework for studying typological universals. We view language as a system and linguistic features as its components whose relationships are encoded in a Directed Acyclic Graph (DAG). Taking discovery of the word order universals as a knowledge discovery task we learn the graphical representation of a word order sub-system which reveals a finer structure such as direct and indirect dependencies among word order features. Then probabilistic inference enables us to see the strength of such relationships: given the observed value of one feature (or combination of features), the probabilities of values of other features can be calculated. Our model is not restricted to using only two values of a feature. Using imputation technique and EM algorithm it can handle missing val- ues well. Model averaging technique solves the problem of limited data. In addition the incremental and divide-and-conquer method addresses the areal and genetic effects simultaneously instead of separately as in Daumé III and Campbell (2007). 1

5 0.59879941 28 acl-2013-A Unified Morpho-Syntactic Scheme of Stanford Dependencies

Author: Reut Tsarfaty

Abstract: Stanford Dependencies (SD) provide a functional characterization of the grammatical relations in syntactic parse-trees. The SD representation is useful for parser evaluation, for downstream applications, and, ultimately, for natural language understanding, however, the design of SD focuses on structurally-marked relations and under-represents morphosyntactic realization patterns observed in Morphologically Rich Languages (MRLs). We present a novel extension of SD, called Unified-SD (U-SD), which unifies the annotation of structurally- and morphologically-marked relations via an inheritance hierarchy. We create a new resource composed of U-SDannotated constituency and dependency treebanks for the MRL Modern Hebrew, and present two systems that can automatically predict U-SD annotations, for gold segmented input as well as raw texts, with high baseline accuracy.

6 0.59665251 85 acl-2013-Combining Intra- and Multi-sentential Rhetorical Parsing for Document-level Discourse Analysis

7 0.59059459 229 acl-2013-Leveraging Synthetic Discourse Data via Multi-task Learning for Implicit Discourse Relation Recognition

8 0.58593661 189 acl-2013-ImpAr: A Deterministic Algorithm for Implicit Semantic Role Labelling

9 0.57952642 161 acl-2013-Fluid Construction Grammar for Historical and Evolutionary Linguistics

10 0.5667721 364 acl-2013-Typesetting for Improved Readability using Lexical and Syntactic Information

11 0.53420609 94 acl-2013-Coordination Structures in Dependency Treebanks

12 0.5312295 331 acl-2013-Stop-probability estimates computed on a large corpus improve Unsupervised Dependency Parsing

13 0.52118814 225 acl-2013-Learning to Order Natural Language Texts

14 0.51886111 205 acl-2013-Joint Apposition Extraction with Syntactic and Semantic Constraints

15 0.51624346 311 acl-2013-Semantic Neighborhoods as Hypergraphs

16 0.51351035 242 acl-2013-Mining Equivalent Relations from Linked Data

17 0.51249522 260 acl-2013-Nonconvex Global Optimization for Latent-Variable Models

18 0.50643367 270 acl-2013-ParGramBank: The ParGram Parallel Treebank

19 0.50625694 17 acl-2013-A Random Walk Approach to Selectional Preferences Based on Preference Ranking and Propagation

20 0.50293702 61 acl-2013-Automatic Interpretation of the English Possessive


similar papers computed by lda model

lda for this paper:

topicId topicWeight

[(0, 0.034), (6, 0.035), (11, 0.072), (15, 0.019), (24, 0.043), (26, 0.065), (28, 0.019), (35, 0.052), (42, 0.074), (44, 0.232), (48, 0.045), (70, 0.052), (84, 0.014), (88, 0.046), (90, 0.018), (95, 0.06)]

similar papers list:

simIndex simValue paperId paperTitle

same-paper 1 0.80111957 280 acl-2013-Plurality, Negation, and Quantification:Towards Comprehensive Quantifier Scope Disambiguation

Author: Mehdi Manshadi ; Daniel Gildea ; James Allen

Abstract: Recent work on statistical quantifier scope disambiguation (QSD) has improved upon earlier work by scoping an arbitrary number and type of noun phrases. No corpusbased method, however, has yet addressed QSD when incorporating the implicit universal of plurals and/or operators such as negation. In this paper we report early, though promising, results for automatic QSD when handling both phenomena. We also present a general model for learning to build partial orders from a set of pairwise preferences. We give an n log n algorithm for finding a guaranteed approximation of the optimal solution, which works very well in practice. Finally, we significantly improve the performance of the pre- vious model using a rich set of automatically generated features.

2 0.74418253 373 acl-2013-Using Conceptual Class Attributes to Characterize Social Media Users

Author: Shane Bergsma ; Benjamin Van Durme

Abstract: We describe a novel approach for automatically predicting the hidden demographic properties of social media users. Building on prior work in common-sense knowledge acquisition from third-person text, we first learn the distinguishing attributes of certain classes of people. For example, we learn that people in the Female class tend to have maiden names and engagement rings. We then show that this knowledge can be used in the analysis of first-person communication; knowledge of distinguishing attributes allows us to both classify users and to bootstrap new training examples. Our novel approach enables substantial improvements on the widelystudied task of user gender prediction, ob- taining a 20% relative error reduction over the current state-of-the-art.

3 0.65228021 134 acl-2013-Embedding Semantic Similarity in Tree Kernels for Domain Adaptation of Relation Extraction

Author: Barbara Plank ; Alessandro Moschitti

Abstract: Relation Extraction (RE) is the task of extracting semantic relationships between entities in text. Recent studies on relation extraction are mostly supervised. The clear drawback of supervised methods is the need of training data: labeled data is expensive to obtain, and there is often a mismatch between the training data and the data the system will be applied to. This is the problem of domain adaptation. In this paper, we propose to combine (i) term generalization approaches such as word clustering and latent semantic analysis (LSA) and (ii) structured kernels to improve the adaptability of relation extractors to new text genres/domains. The empirical evaluation on ACE 2005 domains shows that a suitable combination of syntax and lexical generalization is very promising for domain adaptation.

4 0.58719134 80 acl-2013-Chinese Parsing Exploiting Characters

Author: Meishan Zhang ; Yue Zhang ; Wanxiang Che ; Ting Liu

Abstract: Characters play an important role in the Chinese language, yet computational processing of Chinese has been dominated by word-based approaches, with leaves in syntax trees being words. We investigate Chinese parsing from the character-level, extending the notion of phrase-structure trees by annotating internal structures of words. We demonstrate the importance of character-level information to Chinese processing by building a joint segmentation, part-of-speech (POS) tagging and phrase-structure parsing system that integrates character-structure features. Our joint system significantly outperforms a state-of-the-art word-based baseline on the standard CTB5 test, and gives the best published results for Chinese parsing.

5 0.58706498 132 acl-2013-Easy-First POS Tagging and Dependency Parsing with Beam Search

Author: Ji Ma ; Jingbo Zhu ; Tong Xiao ; Nan Yang

Abstract: In this paper, we combine easy-first dependency parsing and POS tagging algorithms with beam search and structured perceptron. We propose a simple variant of “early-update” to ensure valid update in the training process. The proposed solution can also be applied to combine beam search and structured perceptron with other systems that exhibit spurious ambiguity. On CTB, we achieve 94.01% tagging accuracy and 86.33% unlabeled attachment score with a relatively small beam width. On PTB, we also achieve state-of-the-art performance. 1

6 0.58536947 155 acl-2013-Fast and Accurate Shift-Reduce Constituent Parsing

7 0.58180237 318 acl-2013-Sentiment Relevance

8 0.58166373 343 acl-2013-The Effect of Higher-Order Dependency Features in Discriminative Phrase-Structure Parsing

9 0.57834452 225 acl-2013-Learning to Order Natural Language Texts

10 0.57763451 70 acl-2013-Bilingually-Guided Monolingual Dependency Grammar Induction

11 0.57553917 123 acl-2013-Discriminative Learning with Natural Annotations: Word Segmentation as a Case Study

12 0.57395446 358 acl-2013-Transition-based Dependency Parsing with Selectional Branching

13 0.57371634 82 acl-2013-Co-regularizing character-based and word-based models for semi-supervised Chinese word segmentation

14 0.5729593 276 acl-2013-Part-of-Speech Induction in Dependency Trees for Statistical Machine Translation

15 0.57265931 16 acl-2013-A Novel Translation Framework Based on Rhetorical Structure Theory

16 0.57033622 7 acl-2013-A Lattice-based Framework for Joint Chinese Word Segmentation, POS Tagging and Parsing

17 0.56998307 164 acl-2013-FudanNLP: A Toolkit for Chinese Natural Language Processing

18 0.56962502 56 acl-2013-Argument Inference from Relevant Event Mentions in Chinese Argument Extraction

19 0.56943232 98 acl-2013-Cross-lingual Transfer of Semantic Role Labeling Models

20 0.56943119 333 acl-2013-Summarization Through Submodularity and Dispersion