acl acl2011 acl2011-3 knowledge-graph by maker-knowledge-mining
Source: pdf
Author: Ivan Titov ; Alexandre Klementiev
Abstract: We propose a non-parametric Bayesian model for unsupervised semantic parsing. Following Poon and Domingos (2009), we consider a semantic parsing setting where the goal is to (1) decompose the syntactic dependency tree of a sentence into fragments, (2) assign each of these fragments to a cluster of semantically equivalent syntactic structures, and (3) predict predicate-argument relations between the fragments. We use hierarchical PitmanYor processes to model statistical dependencies between meaning representations of predicates and those of their arguments, as well as the clusters of their syntactic realizations. We develop a modification of the MetropolisHastings split-merge sampler, resulting in an efficient inference algorithm for the model. The method is experimentally evaluated by us- ing the induced semantic representation for the question answering task in the biomedical domain.
Reference: text
sentIndex sentText sentNum sentScore
1 de Abstract We propose a non-parametric Bayesian model for unsupervised semantic parsing. [sent-3, score-0.355]
2 We use hierarchical PitmanYor processes to model statistical dependencies between meaning representations of predicates and those of their arguments, as well as the clusters of their syntactic realizations. [sent-5, score-0.546]
3 The method is experimentally evaluated by us- ing the induced semantic representation for the question answering task in the biomedical domain. [sent-7, score-0.478]
4 1 Introduction Statistical approaches to semantic parsing have recently received considerable attention. [sent-8, score-0.296]
5 These approaches cluster semantically equivalent verbalizations of relations, often relying on syntactic fragments as features for relation extraction and clustering (Lin and Pantel, 2001 ; Banko et al. [sent-15, score-0.443]
6 The success of these methods suggests that semantic parsing can also be tackled as clustering of syntactic realizations of predicate-argument relations. [sent-17, score-0.626]
7 , 2009; Lang and Lapata, 2010), the recent work of (Poon and Domingos, 2009) takes it one step further by not only predicting predicate-argument structure of a sentence but also assigning sentence fragments to clusters of semantically similar expressions. [sent-19, score-0.26]
8 For example, for a pair of sentences on Figure 1, in addition to inducing predicate-argument structure, they aim to assign expressions “Steelers” and “the Pittsburgh team” to the same semantic class Steelers, and group expressions “defeated” and “secured the victory over”. [sent-20, score-0.407]
9 Such semantic representation can be useful for entailment or question answering tasks, as an entailment model can abstract away from specifics of syntactic and lexical realization relying instead on the induced semantic representation. [sent-21, score-0.895]
10 For example, the two sentences in Figure 1have identical semantic representation, and therefore can be hypothesized to be equivalent. [sent-22, score-0.232]
11 From the statistical modeling point of view, joint learning of predicate-argument structure and discovery of semantic clusters of expressions can also be beneficial, because it results in a more compact model of selectional preference, less prone to the data-sparsity problem (Zapirain et al. [sent-26, score-0.459]
12 Hierarchical Pitman-Yor processes (or their special case, hierarchical Dirichlet processes) have previously been used in NLP, for example, in the context of syntactic parsing (Liang et al. [sent-34, score-0.416]
13 In our case, the state space size equals 1446 the total number of distinct semantic clusters, and, thus, is expected to be exceedingly large even for moderate datasets: for example, the MLN model induces 18,543 distinct clusters from 18,471 sentences of the GENIA corpus (Poon and Domingos, 2009). [sent-42, score-0.383]
14 Instead, we use a split-merge MH sampling algorithm, which is a standard and efficient inference tool for non-hierarchical PY processes (Jain and Neal, 2000; Dahl, 2003) but has not previously been used in hierarchical setting. [sent-44, score-0.331]
15 We evaluate our model both qualitatively, examining the revealed clustering of syntactic structures, and quantitatively, on a question answering task. [sent-47, score-0.324]
16 Section 2 begins with a definition of the semantic parsing task. [sent-51, score-0.296]
17 2 Semantic Parsing In this section, we briefly define the unsupervised semantic parsing task and underlying aspects and assumptions relevant to our model. [sent-56, score-0.378]
18 Unlike (Poon and Domingos, 2009), we do not use the lambda calculus formalism to define our task but rather treat it as an instance of frame-semantic parsing, or a specific type of semantic role labeling (Gildea and Jurafsky, 2002). [sent-57, score-0.377]
19 The reason for this is two-fold: first, the frame semantics view is more standard in computational linguistics, sufficient to describe induced semantic representation and convenient to relate our method to the previous work. [sent-58, score-0.575]
20 Each frame is characterized by a set of semantic roles (frame el- ements) corresponding to the arguments of the predicate. [sent-62, score-0.58]
21 It is evoked by a frame evoking element (a predicate). [sent-63, score-0.234]
22 The same frame can be evoked by different but semantically similar predicates: for example, both verbs “buy” and “purchase ” evoke frame Commerce buy in FrameNet (Fillmore et al. [sent-64, score-0.461]
23 The aim of the semantic role labeling task is to identify all of the frames evoked in a sentence and label their semantic role fillers. [sent-66, score-0.722]
24 We extend this task and treat semantic parsing as recursive prediction of predicate-argument structure and clustering of argument fillers. [sent-67, score-0.587]
25 In this work, we define a joint model for the labeling and argument identification stages. [sent-72, score-0.32]
26 Similarly to core semantic roles in FrameNet, semantic roles are treated as frame-specific in our model, as our model does not try to discover any correspondences between roles in different frames. [sent-73, score-0.667]
27 As you can see from the above description, frames (which groups predicates with similar meaning such as the WinPri ze frame in our example) and clusters of argument fillers (Ravens and Steelers) are treated in our definition in a similar way. [sent-74, score-0.758]
28 For convenience, we will refer to both types of clusters as semantic classes. [sent-75, score-0.342]
29 Second, we assume that the semantic arguments are local in the dependency tree; that is, one lexical item can be a semantic argument of another one only if they are connected by an arc in the dependency tree. [sent-80, score-0.931]
30 This is a slight simplification of the semantic role labeling problem but one often made. [sent-81, score-0.336]
31 Thus, the argument identification and labeling stages consist of labeling each syntactic arc with a semantic role label. [sent-82, score-0.746]
32 In comparison, the MLN model does not explicitly assume contiguity of lexical items and does not make this directionality assumption but their clustering algorithm uses initialization and clusterization moves such that the resulting model also obeys both of these constraints. [sent-83, score-0.27]
33 Third, as in (Poon and Domingos, 2009), we do not model polysemy as we assume 1Semantic classes correspond to lambda-form clusters in (Poon and Domingos, 2009) terminology. [sent-84, score-0.263]
34 that each syntactic fragment corresponds to a single semantic class. [sent-85, score-0.409]
35 As in some of the recent work on learning semantic representations (Eisenstein et al. [sent-88, score-0.232]
36 Though we include generation of the syntactic structure in our model, we would not expect that this syntactic component would result in an accurate syntactic model, even if trained in a supervised way, as the chosen independence assumptions are oversimplistic. [sent-92, score-0.393]
37 The reason for this is that MLNs are undirected models and when learned to maximize likelihood of syntactically annotated sentences, they would require marginalization over semantic representation but also over the entire space of syntactic structures and lexical units. [sent-97, score-0.408]
38 Given the complexity ofthe semantic parsing task and the need to tackle large datasets, even approximate methods are likely to be infeasible. [sent-98, score-0.296]
39 The class of such restricted MLNs is equivalent to the class of directed graphical models over the same set of random variables corresponding to fragments of syntactic and semantic structure. [sent-101, score-0.658]
40 We use PY processes to model distributions of semantic classes appearing as an argument of other semantic classes. [sent-106, score-1.073]
41 We also use them to model distributions of syntactic realizations for each semantic class and distributions of syntactic dependency arcs for argument types. [sent-107, score-1.252]
42 PY processes are expected to be more appropriate for many NLP problems, as they model power-law type distributions common for natural language (Teh, 2006). [sent-126, score-0.307]
43 5 A Model for Semantic Parsing Our model of semantics associates with each semantic class a set of distributions which govern the generation of corresponding syntactic realizations2 and the selection of semantic classes for its arguments. [sent-137, score-0.99]
44 Each sentence is generated starting from the root of its dependency tree, recursively drawing a semantic class, its syntactic realization, arguments and semantic classes for the arguments. [sent-138, score-0.88]
45 Below we describe the model by first defining the set ofthe model parameters and then explaining the generation of in2Syntactic realizations are syntactic tree fragments, and therefore they correspond both to syntactic and lexical variations. [sent-139, score-0.483]
46 We associate with each semantic class c, c = 1, 2, . [sent-142, score-0.326]
47 , a distribution of its syntactic realizations φc. [sent-145, score-0.323]
48 For example, for the frame WinPri ze illustrated in Figure 1 this distribution would concentrate at syntactic fragments corresponding to lexical items “defeated”, “secured the victory” and “won ”. [sent-146, score-0.55]
49 The distribution is drawn from DP(w(C), H(C)), where H(C) is a base measure over syntactic subtrees. [sent-147, score-0.254]
50 , mT) where mt is the number of times an argument of type t appears for a given semantic frame occur- q(C), rence3. [sent-161, score-0.688]
51 For the frame WinPrize these parameters would enforce that there exists exactly one Winner and exactly one Opponent for each occurrence of WinPri ze. [sent-162, score-0.246]
52 The parameter ψc,t defines the probability of having at least one argument of type t. [sent-163, score-0.272]
53 If 0 is drawn from ψc,t then mt = 0, otherwise the number of additional arguments of type t (mt − 1) is drawn from the geometric distribution Geom(ψ+c,t). [sent-164, score-0.344]
54 rTawhins generative story is flexible enough to accommodate both argument types which appear at most once per semantic class occurrence (e. [sent-165, score-0.717]
55 , agents), and argument types which frequently appear multiple times per semantic class occurrence (e. [sent-167, score-0.619]
56 , T, define the dis3For simplicity, we assume that each semantic class has T associated argument types, note that this is not a restrictive assumption as some of the argument types can remain unused, and T can be selected to be sufficiently large to accommodate all important arguments. [sent-173, score-0.788]
57 tributions over syntactic paths for the argument type t. [sent-175, score-0.403]
58 In our example, for argument type Opponent, this distribution would associate most of the probability mass with relations pp over, dobj and pp against. [sent-176, score-0.367]
59 These distributions are drawn from , In this paper we only consider paths consisting of a single relation, therefore the base probability distribution H(A) is just normalized frequencies of dependency relations in the treebank. [sent-177, score-0.323]
60 The crucial part of the model are the selectionpreference parameters θc,t, the distributions of semantic classes c0 for each argument type t of class c. [sent-178, score-0.846]
61 For arguments Winner and Opponent of the frame WinPri ze these distributions would assign most of the probability mass to semantic classes denoting teams or players. [sent-179, score-0.808]
62 Distributions θc,t are drawn from a hierarchical PY process: first, top-level proportions of classes γ are drawn from GEM(α0, β0), and then the individual distributions θc,t over c0 are chosen from PY (α, β, γ) . [sent-180, score-0.438]
63 1450 sponding to the root of the dependency tree from the root-specific distribution of semantic classes θroot. [sent-182, score-0.46]
64 For a class c, we generate the syntactic realization s and for each of the T types, decide how many arguments of that type to generate (see GenSemClass in Figure 2). [sent-184, score-0.437]
65 Then we generate each of the arguments (see GenArgument) by first generating a syntactic arc ac,t, choosing a class as its filler c0c,t and, finally, recursing. [sent-185, score-0.335]
66 6 Inference In our model, latent states, modeled with hierarchical PY processes, correspond to distinct semantic classes and, therefore, their number is expected to be very large for any reasonable model of semantics. [sent-186, score-0.476]
67 One of the standard and most efficient samplers for non-hierarchical PY processes are split-merge MH samplers (Jain and Neal, 2000; Dahl, 2003). [sent-188, score-0.27]
68 1 Split and Merge Moves On each move, split-merge samplers decide either to merge two states into one (in our case, merge two semantic classes), or split one state into two. [sent-191, score-0.394]
69 Note that for any reasonable model of semantics only a small subset of the entire set of semantic classes can be used as an argument for some fixed semantic class due to selectional preferences exhibited by predicates. [sent-193, score-1.071]
70 For instance, only teams or players can fill arguments of the frame WinPri ze in our running example. [sent-194, score-0.369]
71 2 Compose and Decompose Moves The compose-decompose operations modify syntactic fragments assigned to semantic classes, composing two neighboring dependency sub-trees or decomposing a dependency sub-tree. [sent-199, score-0.596]
72 c Tlashsi sw fhraicghcan later be merged with other classes, such as the ones containing syntactic realizations “defeat” or “win”. [sent-205, score-0.27]
73 Conversely, if both randomly-selected syntactic fragments are already composed in the corresponding class, we attempt to split them. [sent-206, score-0.238]
74 3 Role-Syntax Alignment Move Merge, compose and decompose moves require re- computation of mapping between argument types (semantic roles) and syntactic fragments. [sent-208, score-0.602]
75 Though it is a part of the above 3 moves, this alignment move is also used on its own to induce semantic arguments for classes (frames) with a single syntactic realization. [sent-211, score-0.669]
76 The Gibbs scan procedure is also used during the split move to select one of the newly introduced classes for each considered syntactic fragment. [sent-212, score-0.327]
77 4 Informed Proposals Since the number of classes is very large, selecting examples at random would result in a relatively low proportion of moves getting accepted, and, consequently, in a slow-mixing Markov chain. [sent-214, score-0.24]
78 Instead of selecting both class occurrences uniformly, we select the first occurrence from a uniform distribution and then use a simple but effective proposal distribution for selecting the second class occurrence. [sent-215, score-0.356]
79 Let us denote the class corresponding to the first 1451 occurrence as c1 and its syntactic realization as s1 with a head word w1. [sent-216, score-0.348]
80 We then select a random occurrence of w2; if it is a part of syntactic realization of c1 then a decompose move is attempted. [sent-220, score-0.396]
81 As the moves are dependent only on syntactic representations, all the proposal distributions can be computed once at the initialization stage. [sent-225, score-0.354]
82 4 7 Empirical Evaluation We induced a semantic representation over a collection of texts and evaluated it by answering questions about the knowledge contained in the corpus. [sent-226, score-0.381]
83 Let us examine some of the induced semantic classes (Table 1) before turning to the question answering task. [sent-237, score-0.497]
84 ClassVariations realizations have a clear semantic connection. [sent-242, score-0.371]
85 Verbs in class 9 are used in the context of providing support for a finding or an action, and many of them are listed as evoking elements for the Evidence frame in FrameNet. [sent-245, score-0.278]
86 Argument types of the induced classes also show a tendency to correspond to semantic roles. [sent-246, score-0.405]
87 For example, an argument type of class 2 is modeled as a distribution over two argument parts, prep of and prep from. [sent-247, score-0.65]
88 The first set of baselines looks for answers by attempting to match a verb and its argument in the question with the input text. [sent-253, score-0.34]
89 When the extracted rela- tion and one of the arguments match those in a given 1452 question, the second argument is returned as an answer. [sent-256, score-0.341]
90 The EXACT versions of the methods return answers when they match the question argument exactly, and the SUB versions produce answers containing the question argument as a substring. [sent-259, score-0.68]
91 recursively decompose it into lexical items and assign them to semantic classes induced at training. [sent-263, score-0.463]
92 Using this semantic representation, we look for the type of an argument missing in the question, which, if found, is reported as an answer. [sent-264, score-0.504]
93 It is clear that overly coarse clusters of argument fillers or clustering of semantically related but not equivalent relations can hurt precision for this evaluation method. [sent-265, score-0.553]
94 Though all these clusters have clear semantic interpretation (white blood cells, predicates corresponding to changes and cykotines associated with cancer progression, respectively), they appear to be too coarse for the QA method we use in our experiments. [sent-275, score-0.385]
95 8 Related Work There is a growing body of work on statistical learning for different versions of the semantic parsing problem (e. [sent-277, score-0.296]
96 Unsupervised semantic role labeling with a generative model has also been considered (Grenager and Manning, 2006), however, they do not attempt to discover frames and deal only with isolated pred1453 icates. [sent-286, score-0.476]
97 9 Conclusions and Future Work In this work, we introduced a non-parametric Bayesian model for the semantic parsing problem based on the hierarchical Pitman-Yor process. [sent-293, score-0.428]
98 The model defines a generative story for recursive generation of lexical items, syntactic and semantic structures. [sent-294, score-0.502]
99 The second general direction is the use of the unsupervised methods we propose to expand the coverage of existing semantic resources, which typically require substantial human effort to produce. [sent-298, score-0.314]
100 Evaluating framenet-style semantic parsing: the role of coverage gaps in framenet. [sent-435, score-0.288]
wordName wordTfidf (topN-words)
[('semantic', 0.232), ('argument', 0.231), ('domingos', 0.231), ('poon', 0.231), ('mln', 0.225), ('py', 0.194), ('frame', 0.184), ('realizations', 0.139), ('syntactic', 0.131), ('processes', 0.13), ('moves', 0.128), ('ravens', 0.115), ('winpri', 0.115), ('classes', 0.112), ('arguments', 0.11), ('clusters', 0.11), ('fragments', 0.107), ('distributions', 0.095), ('class', 0.094), ('mlns', 0.092), ('hierarchical', 0.091), ('bayesian', 0.085), ('move', 0.084), ('unsupervised', 0.082), ('opponent', 0.081), ('steelers', 0.081), ('victory', 0.081), ('selectional', 0.076), ('ze', 0.075), ('goldwasser', 0.07), ('samplers', 0.07), ('drawn', 0.07), ('dahl', 0.069), ('defeated', 0.069), ('winp', 0.069), ('dirichlet', 0.068), ('fillers', 0.067), ('parsing', 0.064), ('mooney', 0.063), ('dependency', 0.063), ('occurrence', 0.062), ('zettlemoyer', 0.062), ('beal', 0.061), ('genia', 0.061), ('induced', 0.061), ('realization', 0.061), ('answers', 0.06), ('sampling', 0.06), ('clustering', 0.06), ('decompose', 0.058), ('mh', 0.058), ('markov', 0.057), ('gem', 0.056), ('cluster', 0.056), ('role', 0.056), ('clarke', 0.056), ('compose', 0.054), ('roles', 0.054), ('distribution', 0.053), ('semantics', 0.053), ('yates', 0.052), ('generative', 0.051), ('gibbs', 0.05), ('evoked', 0.05), ('inference', 0.05), ('liang', 0.05), ('question', 0.049), ('biomedical', 0.048), ('winner', 0.048), ('labeling', 0.048), ('frames', 0.048), ('story', 0.047), ('geom', 0.046), ('hyperpriors', 0.046), ('lers', 0.046), ('markovian', 0.046), ('mmci', 0.046), ('riz', 0.046), ('secure', 0.046), ('secured', 0.046), ('zapirain', 0.046), ('relation', 0.046), ('fragment', 0.046), ('merge', 0.046), ('lang', 0.045), ('teh', 0.045), ('representation', 0.045), ('uniformly', 0.044), ('answering', 0.043), ('predicates', 0.043), ('fillmore', 0.043), ('jain', 0.043), ('logical', 0.043), ('semantically', 0.043), ('relations', 0.042), ('type', 0.041), ('model', 0.041), ('framenet', 0.04), ('grenager', 0.04), ('infinite', 0.04)]
simIndex simValue paperId paperTitle
same-paper 1 0.9999997 3 acl-2011-A Bayesian Model for Unsupervised Semantic Parsing
Author: Ivan Titov ; Alexandre Klementiev
Abstract: We propose a non-parametric Bayesian model for unsupervised semantic parsing. Following Poon and Domingos (2009), we consider a semantic parsing setting where the goal is to (1) decompose the syntactic dependency tree of a sentence into fragments, (2) assign each of these fragments to a cluster of semantically equivalent syntactic structures, and (3) predict predicate-argument relations between the fragments. We use hierarchical PitmanYor processes to model statistical dependencies between meaning representations of predicates and those of their arguments, as well as the clusters of their syntactic realizations. We develop a modification of the MetropolisHastings split-merge sampler, resulting in an efficient inference algorithm for the model. The method is experimentally evaluated by us- ing the induced semantic representation for the question answering task in the biomedical domain.
2 0.30308983 324 acl-2011-Unsupervised Semantic Role Induction via Split-Merge Clustering
Author: Joel Lang ; Mirella Lapata
Abstract: In this paper we describe an unsupervised method for semantic role induction which holds promise for relieving the data acquisition bottleneck associated with supervised role labelers. We present an algorithm that iteratively splits and merges clusters representing semantic roles, thereby leading from an initial clustering to a final clustering of better quality. The method is simple, surprisingly effective, and allows to integrate linguistic knowledge transparently. By combining role induction with a rule-based component for argument identification we obtain an unsupervised end-to-end semantic role labeling system. Evaluation on the CoNLL 2008 benchmark dataset demonstrates that our method outperforms competitive unsupervised approaches by a wide margin.
3 0.20720874 170 acl-2011-In-domain Relation Discovery with Meta-constraints via Posterior Regularization
Author: Harr Chen ; Edward Benson ; Tahira Naseem ; Regina Barzilay
Abstract: We present a novel approach to discovering relations and their instantiations from a collection of documents in a single domain. Our approach learns relation types by exploiting meta-constraints that characterize the general qualities of a good relation in any domain. These constraints state that instances of a single relation should exhibit regularities at multiple levels of linguistic structure, including lexicography, syntax, and document-level context. We capture these regularities via the structure of our probabilistic model as well as a set of declaratively-specified constraints enforced during posterior inference. Across two domains our approach successfully recovers hidden relation structure, comparable to or outperforming previous state-of-the-art approaches. Furthermore, we find that a small , set of constraints is applicable across the domains, and that using domain-specific constraints can further improve performance. 1
4 0.19597301 274 acl-2011-Semi-Supervised Frame-Semantic Parsing for Unknown Predicates
Author: Dipanjan Das ; Noah A. Smith
Abstract: We describe a new approach to disambiguating semantic frames evoked by lexical predicates previously unseen in a lexicon or annotated data. Our approach makes use of large amounts of unlabeled data in a graph-based semi-supervised learning framework. We construct a large graph where vertices correspond to potential predicates and use label propagation to learn possible semantic frames for new ones. The label-propagated graph is used within a frame-semantic parser and, for unknown predicates, results in over 15% absolute improvement in frame identification accuracy and over 13% absolute improvement in full frame-semantic parsing F1 score on a blind test set, over a state-of-the-art supervised baseline.
5 0.19393864 79 acl-2011-Confidence Driven Unsupervised Semantic Parsing
Author: Dan Goldwasser ; Roi Reichart ; James Clarke ; Dan Roth
Abstract: Current approaches for semantic parsing take a supervised approach requiring a considerable amount of training data which is expensive and difficult to obtain. This supervision bottleneck is one of the major difficulties in scaling up semantic parsing. We argue that a semantic parser can be trained effectively without annotated data, and introduce an unsupervised learning algorithm. The algorithm takes a self training approach driven by confidence estimation. Evaluated over Geoquery, a standard dataset for this task, our system achieved 66% accuracy, compared to 80% of its fully supervised counterpart, demonstrating the promise of unsupervised approaches for this task.
6 0.17895155 269 acl-2011-Scaling up Automatic Cross-Lingual Semantic Role Annotation
8 0.14585529 167 acl-2011-Improving Dependency Parsing with Semantic Classes
9 0.13308933 293 acl-2011-Template-Based Information Extraction without the Templates
10 0.13109347 15 acl-2011-A Hierarchical Pitman-Yor Process HMM for Unsupervised Part of Speech Induction
11 0.13012123 57 acl-2011-Bayesian Word Alignment for Statistical Machine Translation
12 0.12250817 277 acl-2011-Semi-supervised Relation Extraction with Large-scale Word Clustering
13 0.12141785 39 acl-2011-An Ensemble Model that Combines Syntactic and Semantic Clustering for Discriminative Dependency Parsing
14 0.11987621 137 acl-2011-Fine-Grained Class Label Markup of Search Queries
15 0.11689694 200 acl-2011-Learning Dependency-Based Compositional Semantics
16 0.11476521 127 acl-2011-Exploiting Web-Derived Selectional Preference to Improve Statistical Dependency Parsing
17 0.1112935 29 acl-2011-A Word-Class Approach to Labeling PSCFG Rules for Machine Translation
18 0.10708024 103 acl-2011-Domain Adaptation by Constraining Inter-Domain Variability of Latent Feature Representation
19 0.10550055 188 acl-2011-Judging Grammaticality with Tree Substitution Grammar Derivations
20 0.1047722 174 acl-2011-Insights from Network Structure for Text Mining
topicId topicWeight
[(0, 0.291), (1, 0.016), (2, -0.162), (3, -0.078), (4, 0.084), (5, 0.012), (6, 0.018), (7, 0.088), (8, -0.031), (9, -0.037), (10, 0.113), (11, -0.062), (12, 0.095), (13, 0.129), (14, -0.105), (15, -0.136), (16, -0.189), (17, -0.121), (18, -0.013), (19, -0.004), (20, -0.011), (21, 0.145), (22, -0.135), (23, -0.118), (24, 0.112), (25, -0.038), (26, -0.087), (27, -0.126), (28, -0.038), (29, -0.016), (30, -0.066), (31, -0.062), (32, -0.058), (33, -0.023), (34, 0.047), (35, -0.032), (36, 0.1), (37, -0.072), (38, -0.025), (39, -0.037), (40, -0.031), (41, 0.026), (42, -0.066), (43, 0.096), (44, 0.004), (45, 0.001), (46, -0.07), (47, -0.06), (48, -0.106), (49, -0.008)]
simIndex simValue paperId paperTitle
same-paper 1 0.97244954 3 acl-2011-A Bayesian Model for Unsupervised Semantic Parsing
Author: Ivan Titov ; Alexandre Klementiev
Abstract: We propose a non-parametric Bayesian model for unsupervised semantic parsing. Following Poon and Domingos (2009), we consider a semantic parsing setting where the goal is to (1) decompose the syntactic dependency tree of a sentence into fragments, (2) assign each of these fragments to a cluster of semantically equivalent syntactic structures, and (3) predict predicate-argument relations between the fragments. We use hierarchical PitmanYor processes to model statistical dependencies between meaning representations of predicates and those of their arguments, as well as the clusters of their syntactic realizations. We develop a modification of the MetropolisHastings split-merge sampler, resulting in an efficient inference algorithm for the model. The method is experimentally evaluated by us- ing the induced semantic representation for the question answering task in the biomedical domain.
2 0.89158928 324 acl-2011-Unsupervised Semantic Role Induction via Split-Merge Clustering
Author: Joel Lang ; Mirella Lapata
Abstract: In this paper we describe an unsupervised method for semantic role induction which holds promise for relieving the data acquisition bottleneck associated with supervised role labelers. We present an algorithm that iteratively splits and merges clusters representing semantic roles, thereby leading from an initial clustering to a final clustering of better quality. The method is simple, surprisingly effective, and allows to integrate linguistic knowledge transparently. By combining role induction with a rule-based component for argument identification we obtain an unsupervised end-to-end semantic role labeling system. Evaluation on the CoNLL 2008 benchmark dataset demonstrates that our method outperforms competitive unsupervised approaches by a wide margin.
3 0.83280295 68 acl-2011-Classifying arguments by scheme
Author: Vanessa Wei Feng ; Graeme Hirst
Abstract: Argumentation schemes are structures or templates for various kinds of arguments. Given the text of an argument with premises and conclusion identified, we classify it as an instance ofone offive common schemes, using features specific to each scheme. We achieve accuracies of 63–91% in one-against-others classification and 80–94% in pairwise classification (baseline = 50% in both cases).
4 0.76278532 269 acl-2011-Scaling up Automatic Cross-Lingual Semantic Role Annotation
Author: Lonneke van der Plas ; Paola Merlo ; James Henderson
Abstract: Broad-coverage semantic annotations for training statistical learners are only available for a handful of languages. Previous approaches to cross-lingual transfer of semantic annotations have addressed this problem with encouraging results on a small scale. In this paper, we scale up previous efforts by using an automatic approach to semantic annotation that does not rely on a semantic ontology for the target language. Moreover, we improve the quality of the transferred semantic annotations by using a joint syntacticsemantic parser that learns the correlations between syntax and semantics of the target language and smooths out the errors from automatic transfer. We reach a labelled F-measure for predicates and arguments of only 4% and 9% points, respectively, lower than the upper bound from manual annotations.
5 0.76043248 274 acl-2011-Semi-Supervised Frame-Semantic Parsing for Unknown Predicates
Author: Dipanjan Das ; Noah A. Smith
Abstract: We describe a new approach to disambiguating semantic frames evoked by lexical predicates previously unseen in a lexicon or annotated data. Our approach makes use of large amounts of unlabeled data in a graph-based semi-supervised learning framework. We construct a large graph where vertices correspond to potential predicates and use label propagation to learn possible semantic frames for new ones. The label-propagated graph is used within a frame-semantic parser and, for unknown predicates, results in over 15% absolute improvement in frame identification accuracy and over 13% absolute improvement in full frame-semantic parsing F1 score on a blind test set, over a state-of-the-art supervised baseline.
6 0.70004094 200 acl-2011-Learning Dependency-Based Compositional Semantics
7 0.65270454 79 acl-2011-Confidence Driven Unsupervised Semantic Parsing
8 0.65156317 322 acl-2011-Unsupervised Learning of Semantic Relation Composition
9 0.61600554 293 acl-2011-Template-Based Information Extraction without the Templates
10 0.60989982 320 acl-2011-Unsupervised Discovery of Domain-Specific Knowledge from Text
11 0.59135801 174 acl-2011-Insights from Network Structure for Text Mining
12 0.53066623 167 acl-2011-Improving Dependency Parsing with Semantic Classes
13 0.51199001 170 acl-2011-In-domain Relation Discovery with Meta-constraints via Posterior Regularization
14 0.50751489 15 acl-2011-A Hierarchical Pitman-Yor Process HMM for Unsupervised Part of Speech Induction
15 0.50689483 84 acl-2011-Contrasting Opposing Views of News Articles on Contentious Issues
17 0.49497285 319 acl-2011-Unsupervised Decomposition of a Document into Authorial Components
18 0.46422487 224 acl-2011-Models and Training for Unsupervised Preposition Sense Disambiguation
19 0.46060362 284 acl-2011-Simple Unsupervised Grammar Induction from Raw Text with Cascaded Finite State Models
20 0.45868626 17 acl-2011-A Large Scale Distributed Syntactic, Semantic and Lexical Language Model for Machine Translation
topicId topicWeight
[(5, 0.039), (8, 0.144), (17, 0.07), (26, 0.023), (31, 0.019), (37, 0.121), (39, 0.056), (41, 0.084), (55, 0.043), (59, 0.084), (72, 0.024), (91, 0.051), (96, 0.156)]
simIndex simValue paperId paperTitle
1 0.94545263 296 acl-2011-Terminal-Aware Synchronous Binarization
Author: Licheng Fang ; Tagyoung Chung ; Daniel Gildea
Abstract: We present an SCFG binarization algorithm that combines the strengths of early terminal matching on the source language side and early language model integration on the target language side. We also examine how different strategies of target-side terminal attachment during binarization can significantly affect translation quality.
2 0.89732504 128 acl-2011-Exploring Entity Relations for Named Entity Disambiguation
Author: Danuta Ploch
Abstract: Named entity disambiguation is the task of linking an entity mention in a text to the correct real-world referent predefined in a knowledge base, and is a crucial subtask in many areas like information retrieval or topic detection and tracking. Named entity disambiguation is challenging because entity mentions can be ambiguous and an entity can be referenced by different surface forms. We present an approach that exploits Wikipedia relations between entities co-occurring with the ambiguous form to derive a range of novel features for classifying candidate referents. We find that our features improve disambiguation results significantly over a strong popularity baseline, and are especially suitable for recognizing entities not contained in the knowledge base. Our system achieves state-of-the-art results on the TAC-KBP 2009 dataset.
same-paper 3 0.88948911 3 acl-2011-A Bayesian Model for Unsupervised Semantic Parsing
Author: Ivan Titov ; Alexandre Klementiev
Abstract: We propose a non-parametric Bayesian model for unsupervised semantic parsing. Following Poon and Domingos (2009), we consider a semantic parsing setting where the goal is to (1) decompose the syntactic dependency tree of a sentence into fragments, (2) assign each of these fragments to a cluster of semantically equivalent syntactic structures, and (3) predict predicate-argument relations between the fragments. We use hierarchical PitmanYor processes to model statistical dependencies between meaning representations of predicates and those of their arguments, as well as the clusters of their syntactic realizations. We develop a modification of the MetropolisHastings split-merge sampler, resulting in an efficient inference algorithm for the model. The method is experimentally evaluated by us- ing the induced semantic representation for the question answering task in the biomedical domain.
4 0.85041416 324 acl-2011-Unsupervised Semantic Role Induction via Split-Merge Clustering
Author: Joel Lang ; Mirella Lapata
Abstract: In this paper we describe an unsupervised method for semantic role induction which holds promise for relieving the data acquisition bottleneck associated with supervised role labelers. We present an algorithm that iteratively splits and merges clusters representing semantic roles, thereby leading from an initial clustering to a final clustering of better quality. The method is simple, surprisingly effective, and allows to integrate linguistic knowledge transparently. By combining role induction with a rule-based component for argument identification we obtain an unsupervised end-to-end semantic role labeling system. Evaluation on the CoNLL 2008 benchmark dataset demonstrates that our method outperforms competitive unsupervised approaches by a wide margin.
5 0.84437311 126 acl-2011-Exploiting Syntactico-Semantic Structures for Relation Extraction
Author: Yee Seng Chan ; Dan Roth
Abstract: In this paper, we observe that there exists a second dimension to the relation extraction (RE) problem that is orthogonal to the relation type dimension. We show that most of these second dimensional structures are relatively constrained and not difficult to identify. We propose a novel algorithmic approach to RE that starts by first identifying these structures and then, within these, identifying the semantic type of the relation. In the real RE problem where relation arguments need to be identified, exploiting these structures also allows reducing pipelined propagated errors. We show that this RE framework provides significant improvement in RE performance.
6 0.84078789 170 acl-2011-In-domain Relation Discovery with Meta-constraints via Posterior Regularization
7 0.83731639 65 acl-2011-Can Document Selection Help Semi-supervised Learning? A Case Study On Event Extraction
8 0.83387589 190 acl-2011-Knowledge-Based Weak Supervision for Information Extraction of Overlapping Relations
9 0.83275175 164 acl-2011-Improving Arabic Dependency Parsing with Form-based and Functional Morphological Features
10 0.83153915 277 acl-2011-Semi-supervised Relation Extraction with Large-scale Word Clustering
11 0.82883596 58 acl-2011-Beam-Width Prediction for Efficient Context-Free Parsing
12 0.82823551 311 acl-2011-Translationese and Its Dialects
13 0.82444268 269 acl-2011-Scaling up Automatic Cross-Lingual Semantic Role Annotation
14 0.8222816 244 acl-2011-Peeling Back the Layers: Detecting Event Role Fillers in Secondary Contexts
15 0.82109869 274 acl-2011-Semi-Supervised Frame-Semantic Parsing for Unknown Predicates
16 0.82070577 289 acl-2011-Subjectivity and Sentiment Analysis of Modern Standard Arabic
17 0.81997097 202 acl-2011-Learning Hierarchical Translation Structure with Linguistic Annotations
18 0.8187499 119 acl-2011-Evaluating the Impact of Coder Errors on Active Learning
19 0.81851351 121 acl-2011-Event Discovery in Social Media Feeds
20 0.81844807 209 acl-2011-Lexically-Triggered Hidden Markov Models for Clinical Document Coding