acl acl2011 acl2011-322 knowledge-graph by maker-knowledge-mining
Source: pdf
Author: Eduardo Blanco ; Dan Moldovan
Abstract: This paper presents an unsupervised method for deriving inference axioms by composing semantic relations. The method is independent of any particular relation inventory. It relies on describing semantic relations using primitives and manipulating these primitives according to an algebra. The method was tested using a set of eight semantic relations yielding 78 inference axioms which were evaluated over PropBank.
Reference: text
sentIndex sentText sentNum sentScore
1 edu as Abstract This paper presents an unsupervised method for deriving inference axioms by composing semantic relations. [sent-3, score-0.935]
2 It relies on describing semantic relations using primitives and manipulating these primitives according to an algebra. [sent-5, score-0.963]
3 The method was tested using a set of eight semantic relations yielding 78 inference axioms which were evaluated over PropBank. [sent-6, score-1.004]
4 Previous research has mostly focused on relations between particular kind of arguments, e. [sent-9, score-0.199]
5 Semantic role labelers exclusively detect the relations indicated with solid arrows, which correspond to the sentence syntactic dependencies. [sent-14, score-0.199]
6 On top of those roles, there are at least three more relations (discontinuous arrows) that encode semantics other than the verbargument relations. [sent-15, score-0.267]
7 In this paper, we venture beyond semantic relation extraction from text and investigate techniques to compose them. [sent-16, score-0.263]
8 Going a step further, we consider nonobvious inferences involving AGENT, PURPOSE and other semantic relations. [sent-31, score-0.155]
9 First, an extended definition for semantic relations is proposed, including (1) semantic restrictions for their domains and ranges, and (2) semantic primitives. [sent-33, score-0.703]
10 Second, an algorithm for obtaining inference axioms is described. [sent-34, score-0.647]
11 Axioms take as their premises chains of two relations and output a new relation linking the ends of the chain. [sent-35, score-0.444]
12 The conclusion of an axiom is identified using an algebra for composing semantic primitives. [sent-40, score-0.638]
13 The extended definition, set of primitives, algebra to compose primitives and CSR algorithm are independent of any particular set of relations. [sent-42, score-0.575]
14 In this paper, we extend that work using a different set of primitives and relations. [sent-44, score-0.325]
15 Seventy eight inference axioms are obtained and an empirical evaluation shows that inferred relations have high accuracies. [sent-45, score-0.89]
16 2 Semantic Relations Semantic relations are underlying relations between concepts. [sent-46, score-0.398]
17 Following (Helbig, 2005), we propose an extended definition for semantic relations, including semantic restrictions for its arguments. [sent-50, score-0.359]
18 For example, AGENT(x, y) holds between an animate concrete object x and a situation y. [sent-51, score-0.191]
19 Moreover, we propose to characterize relations by semantic primitives. [sent-52, score-0.313]
20 Besides having a better understanding of each relation, this extended definition allows us to identify possible and not possible combinations of relations, as well as to automatically determine the conclusion of composing a possible combination. [sent-56, score-0.263]
21 , semantic restrictions for x and y); and (b) PR (i. [sent-59, score-0.156]
22 The inverse relation R−1 can be obtained by switching domain and range, and defining PR−1 as depicted in Table 1. [sent-62, score-0.164]
23 1 Semantic Primitives Semantic primitives capture deep characteristics of relations. [sent-64, score-0.325]
24 They are independently determinable for each relation and specify a property between an element of the domain and an element of the range of the relation being described (Huhns and Stephens, 1989). [sent-65, score-0.281]
25 Coupled with domain and range restrictions, primitives allow us to automatically manipulate and reason over relations. [sent-72, score-0.402]
26 −+ − +−R0 2+ + +−R01− −0R0 0 2−+ 0−R+01− +0+R0 02+ + −+R01− −0R0 0 2 + − +0R+−01×− 0R 0 0 02+×+0+−R01×− −+−R0 2+ ×+ Table 2: Algebra for composing semantic primitives. [sent-77, score-0.288]
27 The set of primitives used in this paper (Table 1) is heavily based on previous work in Knowledge Bases (Huhns and Stephens, 1989), but we considered some new primitives. [sent-78, score-0.325]
28 The new primitives are justified by the fact that we aim at composing relations capturing the semantics from natural language. [sent-79, score-0.766]
29 1 An Algebra for Composing Semantic Primitives The key to automatically obtain inference axioms is −, × the ability to know the result of composing primitives. [sent-91, score-0.821]
30 , the values of the ith primitive for R1 and R2, we define an algebra for PRi1 ◦ PRi2 , i. [sent-94, score-0.248]
31 Consider, for example, the Intrinsic primitive: if both relations are intrinsic (+), the composition is intrinsic (+); else if intrinsic does not apply to either relation (0), the primitive does not apply to the composition either (0); else the composition is not intrinsic (−). [sent-99, score-0.8]
32 3 Inference Axioms Semantic relations are composed using inference axioms. [sent-100, score-0.254]
33 An axiom is defined by using the composi1458 premises R1 and R2. [sent-101, score-0.338]
34 tion operator ‘◦’ ; it combines two relations called premises aatnodr yields a ocomnbclinuession tw. [sent-103, score-0.342]
35 o W ree dlaetnioontse an axiom as R1(x, y) ◦ R2(y, z) → R3(x, z), where R1 and R2 are the premises and R3 the conclusion. [sent-104, score-0.338]
36 In order to instantiate an axiom, the premises must form a chain by having argument y in common. [sent-105, score-0.17]
37 The most interesting axioms fall into category (a) and there are ? [sent-113, score-0.592]
38 Depending on n, the number of potential axioms to consider can be significantly large. [sent-118, score-0.592]
39 For n = 20, there are 820 axioms to explore and for n = 30, 1,830. [sent-119, score-0.592]
40 We avoid this by using the extended definition and the algebra for composing primitives. [sent-121, score-0.418]
41 1 Necessary Conditions for Composing Semantic Relations There are two necessary conditions for composing R1 and R2: • They have to be compatible. [sent-123, score-0.174]
42 = = = CSR: An Algorithm for Composing Semantic Relations Consider any set of relations R defined using the extended definition. [sent-132, score-0.247]
43 One can obtain inference axioms using the following algorithm: For (R1, R2) ∈ R R: For (Ri, Rj) ∈ [(R1 , R2), (R1−1 , R2), (R2, R1), (R2, R1 −1)]: 1. [sent-133, score-0.647]
44 Conclusion m)a ∩tc Dh Repeat for R3 ∈ possible conc(R, Ri , Rj) : (a) If DOMAIN(R3) ∩ DOMAIN(Ri) = ∅ or RANGE(R3) ∩ RANGE(Rj) = ∅), b =r e∅a okr (b) If consisten)t ∩(P RR3 , PRi ◦ PRj =), ∅ axioms += Ri (x, y) ◦ Rj ( Py, z) → R3 (x, z) Given R, R−1 can be automatically obtained (Sec- × tion 2). [sent-135, score-0.592]
45 3 An Example: Agent and Purpose We present an example of applying the CSR algorithm by inspecting the potential axiom AGENT(x, y) ◦ PURPOSE−1 (y, z) → R3(x, z), where x is the agent ofy, and action y h→as as its purpose z. [sent-141, score-0.376]
46 A statement instantiating the premises is [Mary]x [came]y to [talk]z about the issue. [sent-142, score-0.175]
47 Given PAGENT and PPURPOSE−1 , we obtain PAGENT ◦ PPURPOSE−1 using the algebra: PAGENT ◦ {+,+, −,+, PPURPOSE−1 PAGENT = 0, = {+,−,−,+,+,−,−,−,−,0,+} PPURPOSE−1 = {+,+,−,+,+,−,−,+,−,0,+} −, −,+,−,0, 0} Out of all relations (Section 4), AGENT and INTENT−1 fit the conclusion match. [sent-147, score-0.199]
48 First, their domains and ranges are compatible with the composition (Step 2a). [sent-148, score-0.246]
49 An empty cell indicates that R1 and R2 do not have compatible domains and ranges; ‘:’ ◦that →the composition is prohibited; and ‘-’ that a relation R3 such that PR3 is consistent with PR1 ◦ PR2 could not be found. [sent-153, score-0.302]
50 These two relations are valid but most probably ignored by a role labeler since Mary is not an argument of talk. [sent-155, score-0.199]
51 It is out of the scope of this paper to explain in detail the semantics of each relation or their detection. [sent-157, score-0.17]
52 Our goal is to obtain inference axioms and, taking for granted that annotation is available, evaluate their accuracy. [sent-158, score-0.647]
53 The only requirement for the CSR algorithm is to define semantic relations using the extended definition (Table 4). [sent-159, score-0.402]
54 The meaning of each relations is as follows: • CAU(x, y) encodes a relation between two situations, where the existence of y is due to the previous existence of x, e. [sent-163, score-0.301]
55 • INT(x, y) links an animate concrete object and the situations he wants to become true, e. [sent-166, score-0.171]
56 1 Inference Axioms Automatically Obtained After applying the CSR algorithm over the relations in Table 4, we obtain 78 unique inference axioms (Table 5). [sent-199, score-0.846]
57 Each sub table must be indexed with the first and second premises as row and column respectively. [sent-200, score-0.188]
58 The table on the left summarizes axioms R1 ◦ R2 → R3 and R2 ◦ R1 → R3, the one in the middle axiom→ R1−1 ◦ R2 → R3 and the one on the right axiom R2 ◦ R1−1◦ → R3. [sent-201, score-0.787]
59 The CS◦R algorithm identifies several correct axioms and accurately marks as prohibited several combinations that would lead to wrong inferences: • For CAUSE, the inherent transitivity is detected (a ◦ a → a). [sent-202, score-0.672]
60 • (Tah e◦ aloca→tion :) and temporal information of concept y iast ioinnhe arintded t by iotsra cause, intention, purpose, agent and manner (sub table on the left, f and g columns). [sent-205, score-0.183]
61 • As expected, axioms involving SYNONYMY as one eoxfp tehcetierd premises yield vthineg other premise as their conclusion (all sub tables). [sent-206, score-0.78]
62 It is important to note that domain and range restrictions are not sufficient to identify inference axioms; they only filter out pairs of not compatible relations. [sent-211, score-0.257]
63 The algebra to compose primitives is used to detect prohibited combinations of relations based on semantic grounds and identify the conclusion of composing them. [sent-212, score-1.094]
64 Without primitives, the cells in Table 5 would be either empty (marking the pair as not compatible) or would simply indicate that the pair has compatible domain and range (without identifying the conclusion). [sent-213, score-0.16]
65 Since more than one conclusion might be detected for the same pair of premises, 78 inference axioms are ultimately identified. [sent-225, score-0.647]
66 This simplified ontology does not aim at defining domains and ranges for any relation set; it is a simplification to fit the eight relations we work with. [sent-249, score-0.461]
67 Because the number of axioms is large we have focused on a subset of them (Table 6). [sent-251, score-0.592]
68 The 31 axioms having SYN as premise are intuitively correct: since synonymous concepts are interchangeable, given veracious annotation they perform valid inferences. [sent-252, score-0.592]
69 First, all instantiations of axiom PRP ◦ MNR−1 → MNR−1 were manually checked. [sent-255, score-0.227]
70 Table 6 depicts the total number of instantiations for each axiom and its accuracy (columns 3 and 4). [sent-261, score-0.227]
71 90, showing that the plausibility of an axiom depends on the axiom. [sent-264, score-0.219]
72 The average accuracy for axioms involving CAU is 0. [sent-265, score-0.592]
73 Other axioms are less productive but have a greater relative impact and accu- added on top the ones already present). [sent-272, score-0.592]
74 For example, axiom PRP ◦ MNR−1 → MNR−1, only yields 71 new MNR, and yet it is adding 3. [sent-281, score-0.195]
75 Overall, applying the seven axioms adds 923 relations on top of the ones already present (2. [sent-284, score-0.791]
76 Figure 2 shows examples of inferences using axioms 1–3. [sent-287, score-0.633]
77 1 Error Analysis Because of the low accuracy of axiom 1, an error analysis was performed. [sent-289, score-0.195]
78 We found that unlike other axioms, this axiom often yield a relation type that is already present in the semantic representation. [sent-290, score-0.411]
79 We use the following heuristic in order to improve accuracy: do not instantiate an axiom R1(x, y) ◦ R2(y, z) → R3(x, z) if a relation of the form R3(x’, z) is already known. [sent-292, score-0.324]
80 1462 6 Comparison with Previous Work There have been many proposals to detect semantic relations from text without composition. [sent-295, score-0.313]
81 , CAUSE (Chang and Choi, 2006; Bethard and Martin, 2008)), relations within noun phrases (Nulty, 2007), named entities (Hirano et al. [sent-298, score-0.199]
82 In contrast to all the above references and the state of the art, the proposed framework obtains axioms that take as input semantic relations produced by others and output more relations: it adds an extra layer of semantics previously ignored. [sent-312, score-0.998]
83 Previous research has exploited the idea of using semantic primitives to define and classify semantic relations under the names of relation elements, deep structure, aspects and primitives. [sent-313, score-0.854]
84 The first attempt on describing semantic relations using primitives was made by Chaffin and Herrmann (1987); they differentiate 3 1 relations using 30 relation elements clustered into five groups (intensional force, dimension, agreement, propositional and part-whole inclusion). [sent-314, score-0.939]
85 There has not been much work on composing relations in the field of computational linguistics. [sent-321, score-0.373]
86 Previous research has manually extracted plausible inference axioms for WordNet relations (Harabagiu and Moldovan, 1998) and transformed chains of relations into theoretical axioms (Helbig, 2005). [sent-330, score-1.663]
87 Composing relations has been proposed before within knowledge bases. [sent-332, score-0.199]
88 Cohen and Losielle (1988) combines a set of nine fairly specific relations (e. [sent-333, score-0.199]
89 The key to determine plausibility is the transitivity characteristic of the aspects: two relations shall not combine if they have contradictory values for any aspect. [sent-336, score-0.223]
90 The first algebra to compose semantic primitives was proposed by Huhns and Stephens (1989). [sent-337, score-0.641]
91 Their relations are not linguistically motivated and ten of them map to some sort of PART-WHOLE (e. [sent-338, score-0.199]
92 Unlike (Cohen and Losielle, 1988; Huhns and Stephens, 1989), we use typical relations that encode the semantics of natural language, propose a method to automatically obtain the inverse of a relation and empirically test the validity of the axioms obtained. [sent-341, score-0.987]
93 7 Conclusions Going beyond current research, in this paper we investigate the composition of semantic relations. [sent-342, score-0.2]
94 The proposed CSR algorithm obtains inference axioms that take as their input semantic relations and output a relation previously ignored. [sent-343, score-1.062]
95 Regardless of the set of relations and annotation scheme, an additional layer of semantics is created on top of the already existing relations. [sent-344, score-0.292]
96 An extended definition for semantic relations is proposed, including restrictions on their domains and ranges as well as values for semantic primitives. [sent-345, score-0.635]
97 An algebra for composing semantic primitives is defined, allowing to automatically determine the primitives values for the composition of any two relations. [sent-347, score-1.179]
98 The CSR algorithm makes use of the extended definition and algebra to discover inference axioms in an unsupervised manner. [sent-348, score-0.891]
99 32% of relations in relative terms with an overall accuracy of 0. [sent-351, score-0.199]
100 This is a novel way of retrieving semantic relations in the field of computational linguistics. [sent-355, score-0.313]
wordName wordTfidf (topN-words)
[('axioms', 0.592), ('primitives', 0.325), ('csr', 0.208), ('relations', 0.199), ('axiom', 0.195), ('composing', 0.174), ('algebra', 0.155), ('rj', 0.146), ('premises', 0.143), ('agent', 0.137), ('semantic', 0.114), ('agt', 0.112), ('huhns', 0.112), ('relation', 0.102), ('mnr', 0.099), ('pagent', 0.096), ('stephens', 0.096), ('primitive', 0.093), ('composition', 0.086), ('compatible', 0.083), ('ppurpose', 0.08), ('prohibited', 0.08), ('talk', 0.074), ('ri', 0.072), ('semantics', 0.068), ('cau', 0.065), ('losielle', 0.064), ('mary', 0.061), ('animate', 0.056), ('inference', 0.055), ('holds', 0.052), ('went', 0.048), ('extended', 0.048), ('chaffin', 0.048), ('helbig', 0.048), ('concrete', 0.048), ('compose', 0.047), ('prp', 0.047), ('ranges', 0.046), ('temporal', 0.046), ('sub', 0.045), ('eight', 0.044), ('purpose', 0.044), ('came', 0.043), ('restrictions', 0.042), ('szpakowicz', 0.042), ('blanco', 0.042), ('situations', 0.042), ('definition', 0.041), ('inferences', 0.041), ('range', 0.041), ('barker', 0.039), ('stan', 0.039), ('ontology', 0.039), ('objects', 0.037), ('intrinsic', 0.037), ('syn', 0.037), ('eduardo', 0.037), ('yesterday', 0.037), ('domain', 0.036), ('moldovan', 0.035), ('situation', 0.035), ('arguments', 0.033), ('aco', 0.032), ('conc', 0.032), ('garage', 0.032), ('herrmann', 0.032), ('homeomerous', 0.032), ('ioms', 0.032), ('primes', 0.032), ('instantiating', 0.032), ('instantiations', 0.032), ('domains', 0.031), ('propbank', 0.03), ('cohen', 0.029), ('loc', 0.029), ('intent', 0.028), ('hands', 0.028), ('hirano', 0.028), ('hendrickx', 0.028), ('anchez', 0.028), ('compositionality', 0.028), ('inherited', 0.028), ('achievements', 0.028), ('opposite', 0.027), ('instantiate', 0.027), ('inverse', 0.026), ('harabagiu', 0.026), ('ruppenhofer', 0.026), ('bethard', 0.026), ('separable', 0.026), ('dowty', 0.026), ('plausible', 0.026), ('compositional', 0.026), ('int', 0.026), ('layer', 0.025), ('links', 0.025), ('winston', 0.024), ('plausibility', 0.024), ('chklovski', 0.024)]
simIndex simValue paperId paperTitle
same-paper 1 0.99999976 322 acl-2011-Unsupervised Learning of Semantic Relation Composition
Author: Eduardo Blanco ; Dan Moldovan
Abstract: This paper presents an unsupervised method for deriving inference axioms by composing semantic relations. The method is independent of any particular relation inventory. It relies on describing semantic relations using primitives and manipulating these primitives according to an algebra. The method was tested using a set of eight semantic relations yielding 78 inference axioms which were evaluated over PropBank.
2 0.10978208 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
3 0.08988367 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 0.0831149 114 acl-2011-End-to-End Relation Extraction Using Distant Supervision from External Semantic Repositories
Author: Truc Vien T. Nguyen ; Alessandro Moschitti
Abstract: In this paper, we extend distant supervision (DS) based on Wikipedia for Relation Extraction (RE) by considering (i) relations defined in external repositories, e.g. YAGO, and (ii) any subset of Wikipedia documents. We show that training data constituted by sentences containing pairs of named entities in target relations is enough to produce reliable supervision. Our experiments with state-of-the-art relation extraction models, trained on the above data, show a meaningful F1 of 74.29% on a manually annotated test set: this highly improves the state-of-art in RE using DS. Additionally, our end-to-end experiments demonstrated that our extractors can be applied to any general text document.
5 0.081931718 262 acl-2011-Relation Guided Bootstrapping of Semantic Lexicons
Author: Tara McIntosh ; Lars Yencken ; James R. Curran ; Timothy Baldwin
Abstract: State-of-the-art bootstrapping systems rely on expert-crafted semantic constraints such as negative categories to reduce semantic drift. Unfortunately, their use introduces a substantial amount of supervised knowledge. We present the Relation Guided Bootstrapping (RGB) algorithm, which simultaneously extracts lexicons and open relationships to guide lexicon growth and reduce semantic drift. This removes the necessity for manually crafting category and relationship constraints, and manually generating negative categories.
6 0.073394366 324 acl-2011-Unsupervised Semantic Role Induction via Split-Merge Clustering
7 0.072316602 86 acl-2011-Coreference for Learning to Extract Relations: Yes Virginia, Coreference Matters
8 0.0677054 277 acl-2011-Semi-supervised Relation Extraction with Large-scale Word Clustering
9 0.0641063 269 acl-2011-Scaling up Automatic Cross-Lingual Semantic Role Annotation
10 0.063901722 174 acl-2011-Insights from Network Structure for Text Mining
11 0.059283789 190 acl-2011-Knowledge-Based Weak Supervision for Information Extraction of Overlapping Relations
12 0.058511484 79 acl-2011-Confidence Driven Unsupervised Semantic Parsing
13 0.057015471 167 acl-2011-Improving Dependency Parsing with Semantic Classes
14 0.056397241 273 acl-2011-Semantic Representation of Negation Using Focus Detection
15 0.054389436 294 acl-2011-Temporal Evaluation
16 0.05293275 53 acl-2011-Automatically Evaluating Text Coherence Using Discourse Relations
17 0.050341673 293 acl-2011-Template-Based Information Extraction without the Templates
18 0.048432216 334 acl-2011-Which Noun Phrases Denote Which Concepts?
19 0.046335969 200 acl-2011-Learning Dependency-Based Compositional Semantics
20 0.045983966 149 acl-2011-Hierarchical Reinforcement Learning and Hidden Markov Models for Task-Oriented Natural Language Generation
topicId topicWeight
[(0, 0.121), (1, 0.029), (2, -0.078), (3, -0.02), (4, 0.034), (5, 0.027), (6, 0.025), (7, -0.001), (8, -0.053), (9, -0.056), (10, 0.039), (11, -0.053), (12, 0.013), (13, 0.018), (14, -0.067), (15, -0.055), (16, -0.05), (17, -0.104), (18, -0.01), (19, 0.003), (20, -0.019), (21, 0.065), (22, -0.047), (23, -0.051), (24, 0.037), (25, -0.033), (26, 0.014), (27, 0.004), (28, 0.046), (29, 0.002), (30, -0.05), (31, 0.053), (32, 0.048), (33, -0.018), (34, -0.022), (35, 0.022), (36, -0.118), (37, 0.031), (38, -0.005), (39, -0.066), (40, -0.047), (41, -0.016), (42, -0.005), (43, -0.062), (44, -0.02), (45, -0.035), (46, -0.027), (47, -0.023), (48, 0.066), (49, 0.044)]
simIndex simValue paperId paperTitle
same-paper 1 0.93725932 322 acl-2011-Unsupervised Learning of Semantic Relation Composition
Author: Eduardo Blanco ; Dan Moldovan
Abstract: This paper presents an unsupervised method for deriving inference axioms by composing semantic relations. The method is independent of any particular relation inventory. It relies on describing semantic relations using primitives and manipulating these primitives according to an algebra. The method was tested using a set of eight semantic relations yielding 78 inference axioms which were evaluated over PropBank.
2 0.75443119 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
3 0.7225855 294 acl-2011-Temporal Evaluation
Author: Naushad UzZaman ; James Allen
Abstract: In this paper we propose a new method for evaluating systems that extract temporal information from text. It uses temporal closure1 to reward relations that are equivalent but distinct. Our metric measures the overall performance of systems with a single score, making comparison between different systems straightforward. Our approach is easy to implement, intuitive, accurate, scalable and computationally inexpensive. 1
4 0.70979887 262 acl-2011-Relation Guided Bootstrapping of Semantic Lexicons
Author: Tara McIntosh ; Lars Yencken ; James R. Curran ; Timothy Baldwin
Abstract: State-of-the-art bootstrapping systems rely on expert-crafted semantic constraints such as negative categories to reduce semantic drift. Unfortunately, their use introduces a substantial amount of supervised knowledge. We present the Relation Guided Bootstrapping (RGB) algorithm, which simultaneously extracts lexicons and open relationships to guide lexicon growth and reduce semantic drift. This removes the necessity for manually crafting category and relationship constraints, and manually generating negative categories.
5 0.67429715 138 acl-2011-French TimeBank: An ISO-TimeML Annotated Reference Corpus
Author: Andre Bittar ; Pascal Amsili ; Pascal Denis ; Laurence Danlos
Abstract: This article presents the main points in the creation of the French TimeBank (Bittar, 2010), a reference corpus annotated according to the ISO-TimeML standard for temporal annotation. A number of improvements were made to the markup language to deal with linguistic phenomena not yet covered by ISO-TimeML, including cross-language modifications and others specific to French. An automatic preannotation system was used to speed up the annotation process. A preliminary evaluation of the methodology adopted for this project yields positive results in terms of data quality and annotation time.
6 0.66598368 114 acl-2011-End-to-End Relation Extraction Using Distant Supervision from External Semantic Repositories
7 0.66346449 190 acl-2011-Knowledge-Based Weak Supervision for Information Extraction of Overlapping Relations
8 0.63954949 200 acl-2011-Learning Dependency-Based Compositional Semantics
9 0.63758558 86 acl-2011-Coreference for Learning to Extract Relations: Yes Virginia, Coreference Matters
10 0.60589772 277 acl-2011-Semi-supervised Relation Extraction with Large-scale Word Clustering
11 0.57299817 53 acl-2011-Automatically Evaluating Text Coherence Using Discourse Relations
12 0.5505209 40 acl-2011-An Error Analysis of Relation Extraction in Social Media Documents
13 0.52736473 239 acl-2011-P11-5002 k2opt.pdf
14 0.52693182 174 acl-2011-Insights from Network Structure for Text Mining
15 0.52645785 3 acl-2011-A Bayesian Model for Unsupervised Semantic Parsing
16 0.52273232 324 acl-2011-Unsupervised Semantic Role Induction via Split-Merge Clustering
17 0.50468141 79 acl-2011-Confidence Driven Unsupervised Semantic Parsing
18 0.496223 215 acl-2011-MACAON An NLP Tool Suite for Processing Word Lattices
19 0.49189058 317 acl-2011-Underspecifying and Predicting Voice for Surface Realisation Ranking
20 0.4898068 68 acl-2011-Classifying arguments by scheme
topicId topicWeight
[(5, 0.029), (17, 0.044), (26, 0.016), (37, 0.063), (39, 0.025), (41, 0.044), (53, 0.015), (55, 0.012), (59, 0.51), (72, 0.02), (91, 0.026), (96, 0.096), (97, 0.01)]
simIndex simValue paperId paperTitle
same-paper 1 0.90420431 322 acl-2011-Unsupervised Learning of Semantic Relation Composition
Author: Eduardo Blanco ; Dan Moldovan
Abstract: This paper presents an unsupervised method for deriving inference axioms by composing semantic relations. The method is independent of any particular relation inventory. It relies on describing semantic relations using primitives and manipulating these primitives according to an algebra. The method was tested using a set of eight semantic relations yielding 78 inference axioms which were evaluated over PropBank.
2 0.84524894 102 acl-2011-Does Size Matter - How Much Data is Required to Train a REG Algorithm?
Author: Mariet Theune ; Ruud Koolen ; Emiel Krahmer ; Sander Wubben
Abstract: In this paper we investigate how much data is required to train an algorithm for attribute selection, a subtask of Referring Expressions Generation (REG). To enable comparison between different-sized training sets, a systematic training method was developed. The results show that depending on the complexity of the domain, training on 10 to 20 items may already lead to a good performance.
3 0.84240568 279 acl-2011-Semi-supervised latent variable models for sentence-level sentiment analysis
Author: Oscar Tackstrom ; Ryan McDonald
Abstract: We derive two variants of a semi-supervised model for fine-grained sentiment analysis. Both models leverage abundant natural supervision in the form of review ratings, as well as a small amount of manually crafted sentence labels, to learn sentence-level sentiment classifiers. The proposed model is a fusion of a fully supervised structured conditional model and its partially supervised counterpart. This allows for highly efficient estimation and inference algorithms with rich feature definitions. We describe the two variants as well as their component models and verify experimentally that both variants give significantly improved results for sentence-level sentiment analysis compared to all baselines. 1 Sentence-level sentiment analysis In this paper, we demonstrate how combining coarse-grained and fine-grained supervision benefits sentence-level sentiment analysis an important task in the field of opinion classification and retrieval (Pang and Lee, 2008). Typical supervised learning approaches to sentence-level sentiment analysis rely on sentence-level supervision. While such fine-grained supervision rarely exist naturally, and thus requires labor intensive manual annotation effort (Wiebe et al., 2005), coarse-grained supervision is naturally abundant in the form of online review ratings. This coarse-grained supervision is, of course, less informative compared to fine-grained supervision, however, by combining a small amount of sentence-level supervision with a large amount of document-level supervision, we are able to substantially improve on the sentence-level classification task. Our work combines two strands of research: models for sentiment analysis that take document structure into account; – 569 Ryan McDonald Google, Inc., New York ryanmcd@ google com . and models that use latent variables to learn unobserved phenomena from that which can be observed. Exploiting document structure for sentiment analysis has attracted research attention since the early work of Pang and Lee (2004), who performed minimal cuts in a sentence graph to select subjective sentences. McDonald et al. (2007) later showed that jointly learning fine-grained (sentence) and coarsegrained (document) sentiment improves predictions at both levels. More recently, Yessenalina et al. (2010) described how sentence-level latent variables can be used to improve document-level prediction and Nakagawa et al. (2010) used latent variables over syntactic dependency trees to improve sentence-level prediction, using only labeled sentences for training. In a similar vein, Sauper et al. (2010) integrated generative content structure models with discriminative models for multi-aspect sentiment summarization and ranking. These approaches all rely on the availability of fine-grained annotations, but Ta¨ckstro¨m and McDonald (201 1) showed that latent variables can be used to learn fine-grained sentiment using only coarse-grained supervision. While this model was shown to beat a set of natural baselines with quite a wide margin, it has its shortcomings. Most notably, due to the loose constraints provided by the coarse supervision, it tends to only predict the two dominant fine-grained sentiment categories well for each document sentiment category, so that almost all sentences in positive documents are deemed positive or neutral, and vice versa for negative documents. As a way of overcoming these shortcomings, we propose to fuse a coarsely supervised model with a fully supervised model. Below, we describe two ways of achieving such a combined model in the framework of structured conditional latent variable models. Contrary to (generative) topic models (Mei et al., 2007; Titov and Proceedings ofP thoer t4l9atnhd A, Onrnuegaoln M,e Jeuntineg 19 o-f2 t4h,e 2 A0s1s1o.c?i ac t2io0n11 fo Ar Cssoocmiaptuiotanti foonra Clo Lminpguutiast i ocns:aslh Loirntpgaupisetrics , pages 569–574, Figure 1: a) Factor graph of the fully observed graphical model. b) Factor graph of the corresponding latent variable model. During training, shaded nodes are observed, while non-shaded nodes are unobserved. The input sentences si are always observed. Note that there are no factors connecting the document node, yd, with the input nodes, s, so that the sentence-level variables, ys, in effect form a bottleneck between the document sentiment and the input sentences. McDonald, 2008; Lin and He, 2009), structured conditional models can handle rich and overlapping features and allow for exact inference and simple gradient based estimation. The former models are largely orthogonal to the one we propose in this work and combining their merits might be fruitful. As shown by Sauper et al. (2010), it is possible to fuse generative document structure models and task specific structured conditional models. While we do model document structure in terms of sentiment transitions, we do not model topical structure. An interesting avenue for future work would be to extend the model of Sauper et al. (2010) to take coarse-grained taskspecific supervision into account, while modeling fine-grained task-specific aspects with latent variables. Note also that the proposed approach is orthogonal to semi-supervised and unsupervised induction of context independent (prior polarity) lexicons (Turney, 2002; Kim and Hovy, 2004; Esuli and Sebastiani, 2009; Rao and Ravichandran, 2009; Velikovich et al., 2010). The output of such models could readily be incorporated as features in the proposed model. 1.1 Preliminaries Let d be a document consisting of n sentences, s = (si)in=1, with a document–sentence-sequence pair denoted d = (d, s). Let yd = (yd, ys) denote random variables1 the document level sentiment, yd, and the sequence of sentence level sentiment, = (ysi)in=1 . – ys 1We are abusing notation throughout by using the same symbols to refer to random variables and their particular assignments. 570 In what follows, we assume that we have access to two training sets: a small set of fully labeled instances, DF = {(dj, and a large set of ydj)}jm=f1, coarsely labeled instances DC = {(dj, yjd)}jm=fm+fm+c1. Furthermore, we assume that yd and all yis take values in {POS, NEG, NEU}. We focus on structured conditional models in the exponential family, with the standard parametrization pθ(yd,ys|s) = expnhφ(yd,ys,s),θi − Aθ(s)o
4 0.83577067 293 acl-2011-Template-Based Information Extraction without the Templates
Author: Nathanael Chambers ; Dan Jurafsky
Abstract: Standard algorithms for template-based information extraction (IE) require predefined template schemas, and often labeled data, to learn to extract their slot fillers (e.g., an embassy is the Target of a Bombing template). This paper describes an approach to template-based IE that removes this requirement and performs extraction without knowing the template structure in advance. Our algorithm instead learns the template structure automatically from raw text, inducing template schemas as sets of linked events (e.g., bombings include detonate, set off, and destroy events) associated with semantic roles. We also solve the standard IE task, using the induced syntactic patterns to extract role fillers from specific documents. We evaluate on the MUC-4 terrorism dataset and show that we induce template structure very similar to handcreated gold structure, and we extract role fillers with an F1 score of .40, approaching the performance of algorithms that require full knowledge of the templates.
5 0.78796333 224 acl-2011-Models and Training for Unsupervised Preposition Sense Disambiguation
Author: Dirk Hovy ; Ashish Vaswani ; Stephen Tratz ; David Chiang ; Eduard Hovy
Abstract: We present a preliminary study on unsupervised preposition sense disambiguation (PSD), comparing different models and training techniques (EM, MAP-EM with L0 norm, Bayesian inference using Gibbs sampling). To our knowledge, this is the first attempt at unsupervised preposition sense disambiguation. Our best accuracy reaches 56%, a significant improvement (at p <.001) of 16% over the most-frequent-sense baseline.
6 0.75259191 329 acl-2011-Using Deep Morphology to Improve Automatic Error Detection in Arabic Handwriting Recognition
7 0.68295294 51 acl-2011-Automatic Headline Generation using Character Cross-Correlation
8 0.56588936 262 acl-2011-Relation Guided Bootstrapping of Semantic Lexicons
9 0.56324905 164 acl-2011-Improving Arabic Dependency Parsing with Form-based and Functional Morphological Features
10 0.52620125 7 acl-2011-A Corpus for Modeling Morpho-Syntactic Agreement in Arabic: Gender, Number and Rationality
11 0.52503085 170 acl-2011-In-domain Relation Discovery with Meta-constraints via Posterior Regularization
12 0.51554877 324 acl-2011-Unsupervised Semantic Role Induction via Split-Merge Clustering
13 0.49388048 269 acl-2011-Scaling up Automatic Cross-Lingual Semantic Role Annotation
14 0.49275213 167 acl-2011-Improving Dependency Parsing with Semantic Classes
15 0.48297161 3 acl-2011-A Bayesian Model for Unsupervised Semantic Parsing
16 0.47679669 244 acl-2011-Peeling Back the Layers: Detecting Event Role Fillers in Secondary Contexts
17 0.47450802 229 acl-2011-NULEX: An Open-License Broad Coverage Lexicon
18 0.47391349 190 acl-2011-Knowledge-Based Weak Supervision for Information Extraction of Overlapping Relations
19 0.47312394 198 acl-2011-Latent Semantic Word Sense Induction and Disambiguation
20 0.47191095 174 acl-2011-Insights from Network Structure for Text Mining