emnlp emnlp2012 emnlp2012-80 knowledge-graph by maker-knowledge-mining
Source: pdf
Author: Hila Weisman ; Jonathan Berant ; Idan Szpektor ; Ido Dagan
Abstract: Learning inference relations between verbs is at the heart of many semantic applications. However, most prior work on learning such rules focused on a rather narrow set of information sources: mainly distributional similarity, and to a lesser extent manually constructed verb co-occurrence patterns. In this paper, we claim that it is imperative to utilize information from various textual scopes: verb co-occurrence within a sentence, verb cooccurrence within a document, as well as overall corpus statistics. To this end, we propose a much richer novel set of linguistically motivated cues for detecting entailment between verbs and combine them as features in a supervised classification framework. We empirically demonstrate that our model significantly outperforms previous methods and that information from each textual scope contributes to the verb entailment learning task.
Reference: text
sentIndex sentText sentNum sentScore
1 However, most prior work on learning such rules focused on a rather narrow set of information sources: mainly distributional similarity, and to a lesser extent manually constructed verb co-occurrence patterns. [sent-14, score-0.646]
2 In this paper, we claim that it is imperative to utilize information from various textual scopes: verb co-occurrence within a sentence, verb cooccurrence within a document, as well as overall corpus statistics. [sent-15, score-0.857]
3 To this end, we propose a much richer novel set of linguistically motivated cues for detecting entailment between verbs and combine them as features in a supervised classification framework. [sent-16, score-0.93]
4 We empirically demonstrate that our model significantly outperforms previous methods and that information from each textual scope contributes to the verb entailment learning task. [sent-17, score-0.944]
5 , 2006), which refers to such rules as entailment rules. [sent-22, score-0.544]
6 In this work we focus on one 194 of the most important rule types, namely, lexical entailment rules between verbs (verb entailment), e. [sent-23, score-0.811]
7 , ‘whisper → talk’, ‘win → play’ and ‘buy → T‘wheh significance ,o ‘fw wsuinch → → ru plelasy h’a asn lde ‘db utoy a →ctiv oew research in automatic learning of entailment rules between verbs or verb-like structures (Zanzotto et al. [sent-25, score-0.754]
8 Most prior efforts to learn verb entailment rules from large corpora employed distributional similarity methods, assuming that verbs are semantically similar if they occur in similar contexts (Lin, 1998; Berant et al. [sent-30, score-1.466]
9 Fewer works, such as VerbOcean (Chklovski and Pantel, 2004), focused on identifying verb entailment through verb instantiation of manually constructed patterns. [sent-33, score-1.249]
10 In this paper, we claim that on top of standard pattern-based and distributional similarity methods, corpus-based learning of verb entailment can greatly benefit from exploiting additional linguisticallymotivated cues that are specific to verbs. [sent-37, score-1.207]
11 For instance, when verbs co-occur in different clauses of scare’ the same sentence, the syntactic relation between the clauses can be viewed as a proxy for the semantic relation between the verbs. [sent-38, score-0.68]
12 First, we suggest a novel set of entailment indicators that help to detect the likelihood of verb entailment. [sent-43, score-1.054]
13 This results in a supervised corpus-based learning method that combines verb entailment information at the sentence, document and corpus levels. [sent-46, score-0.973]
14 This analysis reveals that using a rich and diverse set of indicators that capture sentence-level interactions between verbs substantially improves verb entailment detection. [sent-49, score-1.227]
15 2 Background The main approach for learning entailment rules between verbs and verb-like structures has employed the distributional hypothesis, which assumes that words with similar meanings appear in similar contexts. [sent-50, score-0.974]
16 A far less explored direction for learning verb entailment involves exploiting verb co-occurrence in a sentence or a document. [sent-54, score-1.249]
17 Starting with candidate verb pairs based on a distributional similarity measure, the patterns are used to choose a semantic relation per verb pair based on the different patterns this pair instantiates. [sent-58, score-1.488]
18 This method is more precise than distributional similarity approaches, but it is highly susceptible to sparseness issues, since verbs do not typically co-occur within rigid patterns. [sent-59, score-0.579]
19 Utilizing verb co-occurrence at the document level, Chambers and Jurafsky (2008) estimate whether a pair of verbs is narratively related by counting the number of times the verbs share an argument in the same document. [sent-60, score-0.972]
20 In a similar manner, Pekar (2008) detects entailment rules between templates from shared arguments within discourse- related clauses in the same document. [sent-61, score-0.636]
21 (2006) introduced a system for learning entailment rules between nouns (e. [sent-64, score-0.544]
22 (2012) utilized various distributional similarity features to identify entailment between lexical-syntactic predicates. [sent-74, score-0.876]
23 In this paper, we follow the supervised approach for semantic relation detection in order to identify verb entailment. [sent-75, score-0.591]
24 While we utilize and adapt useful features from prior work, we introduce a diverse set of novel features for the task, effectively combining verb co-occurrence information at the sentence, doc- ument, and corpus levels. [sent-76, score-0.506]
25 In this section, we introduce linguistically motivated indicators that are specific to verbs and may signal the semantic relation between verb pairs. [sent-78, score-0.926]
26 We suggest that these markers can indicate semantic relations between the main verbs of the connected clauses. [sent-85, score-0.452]
27 For example, in dependency parsing the relation can be captured by labeled dependency edges expressing that one clause is an adverbial adjunct of the other, or that two clauses are coordinated. [sent-88, score-0.522]
28 This can indicate the existence (or lack) of entailment between verbs. [sent-89, score-0.495]
29 For instance, in the sentence “When I walked into the room, he was working out”, the verb ‘walk’ is an adverbial adjunct of the verb ‘work out’. [sent-90, score-0.888]
30 One of the most general verb classes are stative vs. [sent-93, score-0.485]
31 We hypothesize that verb classes are relevant for determining entailment, for example, that stative verbs are not likely to entail event verbs. [sent-97, score-0.729]
32 Detecting verb generality can help us tackle an infamous property of distributional similarity methods, namely, the difficulty in detecting the direction of entailment (Berant et al. [sent-102, score-1.301]
33 For example, the verb ’cover’ appears with many different particles such as ’up ’ and ’for’, while the verb ’coat’ does not. [sent-104, score-0.754]
34 Thus, assuming we have evidence for an entailment relation between the two verbs, this indicator can help us discern the direction of entailment and determine that ‘coat → cover’. [sent-105, score-1.171]
35 Typed Distributional Similarity As discussed in section 2, distributional similarity is the most com- mon source of information for learning semantic relations between verbs. [sent-106, score-0.426]
36 If a verb appears with a small set of adverbs, it is more likely to be a specific verb that already conveys a specific action or state, making an additional adverb redundant. [sent-111, score-0.802]
37 For example, the verb ‘whisper’ conveys a specific manner of talking and will probably not appear with the adverb ‘loudly’, while the verb ‘talk’ is more likely to appear with such an adverb. [sent-112, score-0.802]
38 4 Supervised Entailment Detection In the previous section, we discussed linguistic observations regarding novel indicators that may help in detecting entailment relations between verbs. [sent-114, score-0.788]
39 We next describe how to incorporate these indicators as features within a supervised framework for learning lexical entailment rules between verbs. [sent-115, score-0.79]
40 Specifically, given an ordered verb pair (v1, v2) as input, we learn a classifier that detects whether the entailment relation ‘v1 → v2’ holds for this pair. [sent-119, score-1.059]
41 1 Entailment features Most of our features are based on information extracted from the target verb pair co-occurring within varying textual scopes (sentence, document, corpus). [sent-123, score-0.598]
42 1 Sentence-level co-occurrence We next detail features that address co-occurrence of the target verb pair within a sentence. [sent-131, score-0.483]
43 Discourse markers As discussed in Section 3, discourse markers may signal relations between the main verbs of adjacent clauses. [sent-133, score-0.55]
44 For a target verb pair (v1, v2) and each discourse relation r, we count the number of times that v1 is the main verb in the main clause, v2 is the main verb in the subordinate clause, and the clauses are connected via a marker mapped to r. [sent-137, score-1.705]
45 For example, given the sentence “You must enroll in the competition be197 fore you can participate in it”, the verb pair ( ‘enroll’, ‘participate ’) appears in the ’Temporal’ relation, indicated by the marker ‘before ’, where ‘enroll’ is in the main clause. [sent-138, score-0.537]
46 Dependency relations between clauses As noted in Section 3, the syntactic structure of verb cooccurrence can indicate the existence or lack of entailment. [sent-144, score-0.598]
47 For example, in “it surprised me that the lizard could talk” the verb pair ( ‘surprise ’, ‘talk’) is connected by the ‘obj’ relation. [sent-152, score-0.485]
48 The second relation is the adverbial adjunct relation ‘adv’, in which the subordinate clause is adverbial and describes the time, place, manner, etc. [sent-153, score-0.697]
49 , ‘v1-adv-v2 ’ refers to v1 being in the main clause and connected to the subordinate clause via an adverbial adjunct. [sent-163, score-0.462]
50 tgan,daifntegrw,athrdosug,hmeanwhile Pattern-based We follow Chklovski and Pantel (2004) and extract occurrences of VerbOcean patterns that are instantiated by the target verb pair. [sent-171, score-0.437]
51 Since the corpus pattern counts were very sparse, we defined for a target verb pair (v1, v2) two binary features: the first denotes whether the verb pair instantiates at least one positive pattern, and the second denotes whether the verb pair instantiates at least one negative pattern. [sent-179, score-1.448]
52 For example, given the aforementioned sentences, the value of the positive feature for the verb pair ( ‘startle ’, ‘scare ’) is ‘ 1’ . [sent-180, score-0.437]
53 For example, in “he didn ’t say why he left”, the verb ’say ’ appears in negative polarity and the verb ’leave ’ in positive polarity. [sent-183, score-0.787]
54 For each verb pair co-occurrence, we extract the verbs’ tenses and order them as follows: past < present < future. [sent-186, score-0.437]
55 , if tense-v1 >tense-v2, the verb pair is less likely to entail. [sent-194, score-0.437]
56 Co-reference Following Tremper (2010), in every co-occurrence of (v1,v2) we extract for each verb the set of arguments at either the subject or object positions, denoted A1 and A2 (for v1 and v2, respectively). [sent-195, score-0.433]
57 cTchuer intuition, which is similar to distributional similarity, is that semantically related verbs tend to share arguments. [sent-199, score-0.43]
58 2 Document-level co-occurrence This group of features addresses co-occurrence of a target verb pair within the same document. [sent-208, score-0.483]
59 (1) in their paper) that estimates whether a pair consisting of a verb and one of its dependency relations (v1, r1) is narrativelyrelated to another such pair (v2, r2). [sent-212, score-0.588]
60 Their estimation is based on quantifying the likelihood that two verbs will share an argument that instantiates both the dependency position (v1, r1) and (v2, r2) within documents in which the two verbs co-occur. [sent-213, score-0.573]
61 Such narrative relations may provide cues to the semantic relatedness of the verb pair. [sent-217, score-0.565]
62 We compute for every target verb pair nine features using their narrative score. [sent-218, score-0.58]
63 199 apply three state-of-the-art distributional similarity measures, Lin (Lin, 1998), Weeds precision (Weeds and Weir, 2003) and BInc (Szpektor and Dagan, 2008), to compute for every verb pair a similarity score between each of the five count vectors4. [sent-232, score-0.887]
64 Verb classes Following our discussion in Section 3, we first measure for each target verb v a “stative” feature f by computing the proportion of times it appears in progressive tense, since stative verbs usually do not appear in the progressive tense (e. [sent-236, score-0.751]
65 Then, given a verb pair (v1,v2) and their corresponding stative features f1and f2, we add two features f1 ·f2 and between the· fverb classes of the two verbs. [sent-239, score-0.637]
66 Then, given a verb pair (v1,v2) and their corresponding features f1 and f2, we add the feature We expect that when is high, v1 is more general than v2, which is a negative entailment indicator. [sent-241, score-1.011]
67 Since our model contains many novel features, it is important to investigate their utility for detecting verb entailment. [sent-246, score-0.466]
68 4We employ the common practice of using the pmi between a verb and an argument rather than the argument count as the argument’s weight. [sent-251, score-0.609]
69 1 Experimental Setting To evaluate our proposed supervised model, we constructed a dataset containing labeled verb pairs. [sent-253, score-0.432]
70 Next, we extracted the 20 most similar verbs to each seed verb according to the Lin similarity measure (Lin, 1998), which was computed on the RCV1 corpus. [sent-255, score-0.702]
71 Then, for each seed verb vs and one of its extracted similar verbs vsi we generated the two directed pairs (vs, vsi) and (vis, vs), which represent the candidate rules ‘vs → vsi’ and ‘vis → vs’ respectively. [sent-256, score-0.72]
72 To reduce noise, we vfiltered out v→erb v pairs where one of the verbs is an auxiliary or a light verb such as ’do ’, ’get’ and ’have ’. [sent-257, score-0.587]
73 This step resulted in 812 verb pairs as our dataset6, which were manually annotated by the authors as representing a valid entailment rule or not. [sent-258, score-0.929]
74 To annotate these pairs, we generally followed the rule-based approach for entailment rule annotation, where a rule ‘v1 → v2’ is considered as correct if the annotator coul→d th vink of reasonable contexts under which the rule holds (Dekang and Pantel, 2001 ; Szpektor et al. [sent-259, score-0.666]
75 In total 225 verb pairs were labeled as entailing (the rule ‘v1 → v2’ was judged as correct) and 587 verb pairs were →lab veled as non-entailing (the rule ‘v1 → v2’ was judged as incorrect). [sent-261, score-0.899]
76 Looking more closely, our suggestion for typed distributional similarity proved to be useful, and indeed most of the highly correlated distributional similarity features are typed measures. [sent-282, score-0.944]
77 The table also indicates that document-level cooccurrence contributes positively to entailment detection. [sent-285, score-0.605]
78 Again, we point at the significant correlation of our novel typed measures with verb entailment, in this case the typed narrative measure. [sent-287, score-0.679]
79 For example, verbs connected via an adverbial adjunct ( ‘v2adverb-v1 ’) or an object complement ( ‘v1-obj-v2 ’) are negatively correlated with entailment. [sent-289, score-0.567]
80 This shows that encoding various aspects of verb co-occurrence at the sentence level can lead to better prediction of verb entailment. [sent-292, score-0.754]
81 Finally, we note that PMI at the sentence level is highly correlated with entailment even more than at the document level, since the local textual scope is more indicative, though sparser. [sent-293, score-0.673]
82 To conclude, our feature analysis shows that feaRankTop PositiveTop Negative tures at all levels: sentence, document and corpus, contain useful information for entailment detection, both positive and negative, and should be combined together. [sent-294, score-0.541]
83 Moreover, many of our novel features are among the highly correlated features, showing that devising a rich set ofverb-specific and linguisticallymotivated features provides better discriminative evidence for entailment detection. [sent-295, score-0.684]
84 3 Results and Analysis We compared our method to the following baselines which were mostly taken from or inspired by prior work: Random: A simple pair (v1, v2), randomly probability equal to the pairs out of all verb pairs = 0. [sent-297, score-0.437]
85 TDS: Include only the 15 distributional similarity features in our supervised model. [sent-304, score-0.436]
86 classifier over several distributional similarity features, and provides an evaluation of the discrimina- tive power of distributional similarity alone, without co-occurrence features. [sent-311, score-0.67]
87 TDS+VO: Include only the 15 typed distributional similarity features and the two VerbOcean features in our supervised model. [sent-312, score-0.566]
88 (2006), who combined distributional similarity features and Hearst patterns (Hearst, 1992) for learning entailment between nouns. [sent-314, score-0.936]
89 Yet, VerbOcean positive and negative patterns do add some discriminative power over only distributional similarity measures, as seen by the improvement of TDS+VO over TDS in all criteria. [sent-333, score-0.428]
90 In addition, sentence-level features alone (Sent-level) provide the best discriminative power for verb entailment, compared to document and corpus levels, which include distributional similarity features. [sent-344, score-0.804]
91 As an additional insight from Table 4, we point out that document-level features are not good entailment indicators by themselves (Doc-level in Table 4), and they perform worse than the distributional similarity baseline (TDS at Table 3). [sent-347, score-1.021]
92 As a final analysis, we randomly sampled correct entailment rules learned by our model but missed by the typed distributional similarity classifier (TDS). [sent-350, score-0.963]
93 Our overall impression is that employing co-occurrence information helps to better cap- ture entailment relations other than synonymy and troponymy. [sent-351, score-0.554]
94 e 6 Conclusions and Future Work We presented a supervised classification model for detecting lexical entailment between verbs. [sent-353, score-0.602]
95 At the heart of our model stand novel linguistically motivated indicators that capture positive and negative entailment information. [sent-354, score-0.745]
96 These indicators encompass co-occurrence relationships between verbs at the sentence, document and corpus level, as well as more fine-grained typed distributional similarity measures. [sent-355, score-0.82]
97 Our model incorporates these novel indicators together with useful features from prior work, combining co-occurrence and distributional similarity information about verb pairs. [sent-356, score-0.94]
98 Further feature analysis indicated that our novel indicators contribute greatly to the performance of the model, and that co-occurrence at multiple levels, combined with distributional similarity features, is necessary to achieve the model’s best performance. [sent-358, score-0.517]
99 Integrating pattern-based and distributional similarity methods for lexical entailment acquisition. [sent-454, score-0.83]
100 Discovering asymmetric entailment relations between verbs using selectional preferences. [sent-509, score-0.764]
wordName wordTfidf (topN-words)
[('entailment', 0.495), ('verb', 0.377), ('distributional', 0.22), ('verbs', 0.21), ('verbocean', 0.171), ('indicators', 0.145), ('relation', 0.127), ('tds', 0.126), ('subordinate', 0.126), ('szpektor', 0.117), ('similarity', 0.115), ('stative', 0.108), ('weeds', 0.108), ('clause', 0.105), ('markers', 0.103), ('narrative', 0.097), ('dagan', 0.096), ('pmi', 0.094), ('clauses', 0.092), ('typed', 0.084), ('ido', 0.083), ('adverbial', 0.078), ('mirkin', 0.078), ('discourse', 0.075), ('berant', 0.073), ('presupposition', 0.072), ('tremper', 0.072), ('idan', 0.07), ('cooccurrence', 0.07), ('argument', 0.069), ('directional', 0.065), ('chklovski', 0.062), ('termed', 0.062), ('pair', 0.06), ('patterns', 0.06), ('correlated', 0.06), ('relations', 0.059), ('complement', 0.059), ('rule', 0.057), ('object', 0.056), ('ukwac', 0.056), ('adjunct', 0.056), ('tense', 0.056), ('supervised', 0.055), ('discern', 0.054), ('enroll', 0.054), ('geffet', 0.054), ('hagiwara', 0.054), ('kotlerman', 0.054), ('lindsay', 0.054), ('maayan', 0.054), ('scare', 0.054), ('startle', 0.054), ('vo', 0.054), ('detecting', 0.052), ('hearst', 0.052), ('instantiates', 0.052), ('rules', 0.049), ('adverb', 0.048), ('connected', 0.048), ('coat', 0.047), ('guyon', 0.047), ('vsi', 0.047), ('features', 0.046), ('pantel', 0.046), ('marker', 0.046), ('document', 0.046), ('generality', 0.042), ('positively', 0.04), ('scope', 0.039), ('vs', 0.037), ('talk', 0.037), ('novel', 0.037), ('baroni', 0.036), ('scopes', 0.036), ('temporal', 0.036), ('methodp', 0.036), ('scared', 0.036), ('schoenmackers', 0.036), ('sleeps', 0.036), ('startled', 0.036), ('vis', 0.036), ('whisper', 0.036), ('linguistically', 0.035), ('chambers', 0.035), ('rigid', 0.034), ('pennacchiotti', 0.034), ('event', 0.034), ('loose', 0.033), ('negative', 0.033), ('textual', 0.033), ('semantic', 0.032), ('dependency', 0.032), ('yes', 0.031), ('knott', 0.031), ('entailing', 0.031), ('auc', 0.031), ('bins', 0.031), ('zanzotto', 0.031), ('preposition', 0.031)]
simIndex simValue paperId paperTitle
same-paper 1 1.0000002 80 emnlp-2012-Learning Verb Inference Rules from Linguistically-Motivated Evidence
Author: Hila Weisman ; Jonathan Berant ; Idan Szpektor ; Ido Dagan
Abstract: Learning inference relations between verbs is at the heart of many semantic applications. However, most prior work on learning such rules focused on a rather narrow set of information sources: mainly distributional similarity, and to a lesser extent manually constructed verb co-occurrence patterns. In this paper, we claim that it is imperative to utilize information from various textual scopes: verb co-occurrence within a sentence, verb cooccurrence within a document, as well as overall corpus statistics. To this end, we propose a much richer novel set of linguistically motivated cues for detecting entailment between verbs and combine them as features in a supervised classification framework. We empirically demonstrate that our model significantly outperforms previous methods and that information from each textual scope contributes to the verb entailment learning task.
2 0.14463577 16 emnlp-2012-Aligning Predicates across Monolingual Comparable Texts using Graph-based Clustering
Author: Michael Roth ; Anette Frank
Abstract: Generating coherent discourse is an important aspect in natural language generation. Our aim is to learn factors that constitute coherent discourse from data, with a focus on how to realize predicate-argument structures in a model that exceeds the sentence level. We present an important subtask for this overall goal, in which we align predicates across comparable texts, admitting partial argument structure correspondence. The contribution of this work is two-fold: We first construct a large corpus resource of comparable texts, including an evaluation set with manual predicate alignments. Secondly, we present a novel approach for aligning predicates across comparable texts using graph-based clustering with Mincuts. Our method significantly outperforms other alignment techniques when applied to this novel alignment task, by a margin of at least 6.5 percentage points in F1-score.
3 0.11163209 40 emnlp-2012-Ensemble Semantics for Large-scale Unsupervised Relation Extraction
Author: Bonan Min ; Shuming Shi ; Ralph Grishman ; Chin-Yew Lin
Abstract: Discovering significant types of relations from the web is challenging because of its open nature. Unsupervised algorithms are developed to extract relations from a corpus without knowing the relations in advance, but most of them rely on tagging arguments of predefined types. Recently, a new algorithm was proposed to jointly extract relations and their argument semantic classes, taking a set of relation instances extracted by an open IE algorithm as input. However, it cannot handle polysemy of relation phrases and fails to group many similar (“synonymous”) relation instances because of the sparseness of features. In this paper, we present a novel unsupervised algorithm that provides a more general treatment of the polysemy and synonymy problems. The algorithm incorporates various knowledge sources which we will show to be very effective for unsupervised extraction. Moreover, it explicitly disambiguates polysemous relation phrases and groups synonymous ones. While maintaining approximately the same precision, the algorithm achieves significant improvement on recall compared to the previous method. It is also very efficient. Experiments on a realworld dataset show that it can handle 14.7 million relation instances and extract a very large set of relations from the web. Ralph Grishman1 Chin-Yew Lin2 2Microsoft Research Asia Beijing, China { shumings cyl } @mi cro s o ft . com , that has many applications in answering factoid questions, building knowledge bases and improving search engine relevance. The web has become a massive potential source of such relations. However, its open nature brings an open-ended set of relation types. To extract these relations, a system should not assume a fixed set of relation types, nor rely on a fixed set of relation argument types. The past decade has seen some promising solutions, unsupervised relation extraction (URE) algorithms that extract relations from a corpus without knowing the relations in advance. However, most algorithms (Hasegawa et al., 2004, Shinyama and Sekine, 2006, Chen et. al, 2005) rely on tagging predefined types of entities as relation arguments, and thus are not well-suited for the open domain. Recently, Kok and Domingos (2008) proposed Semantic Network Extractor (SNE), which generates argument semantic classes and sets of synonymous relation phrases at the same time, thus avoiding the requirement of tagging relation arguments of predefined types. However, SNE has 2 limitations: 1) Following previous URE algorithms, it only uses features from the set of input relation instances for clustering. Empirically we found that it fails to group many relevant relation instances. These features, such as the surface forms of arguments and lexical sequences in between, are very sparse in practice. In contrast, there exist several well-known corpus-level semantic resources that can be automatically derived from a source corpus and are shown to be useful for generating the key elements of a relation: its 2 argument semantic classes and a set of synonymous phrases. For example, semantic classes can be derived from a source corpus with contextual distributional simi1 Introduction Relation extraction aims at discovering semantic larity and web table co-occurrences. The “synonymy” 1 problem for clustering relation instances relations between entities. It is an important task * Work done during an internship at Microsoft Research Asia 1027 LParnogcue agdein Lgesa ornf tihneg, 2 p0a1g2e Jso 1in02t C7–o1n0f3e7re,n Jce ju on Is Elanmdp,ir Kicoarlea M,e 1t2h–o1d4s J iunly N 2a0tu1r2a.l ? Lc a2n0g1u2ag Aes Psorcoicaetsiosin fgo arn Cdo Cmopmutpauti oantiaoln Lailn Ngautiustriacls could potentially be better solved by adding these resources. 2) SNE assumes that each entity or relation phrase belongs to exactly one cluster, thus is not able to effectively handle polysemy of relation phrases2. An example of a polysemous phrase is be the currency of as in 2 triples
4 0.10716435 103 emnlp-2012-PATTY: A Taxonomy of Relational Patterns with Semantic Types
Author: Ndapandula Nakashole ; Gerhard Weikum ; Fabian Suchanek
Abstract: This paper presents PATTY: a large resource for textual patterns that denote binary relations between entities. The patterns are semantically typed and organized into a subsumption taxonomy. The PATTY system is based on efficient algorithms for frequent itemset mining and can process Web-scale corpora. It harnesses the rich type system and entity population of large knowledge bases. The PATTY taxonomy comprises 350,569 pattern synsets. Random-sampling-based evaluation shows a pattern accuracy of 84.7%. PATTY has 8,162 subsumptions, with a random-sampling-based precision of 75%. The PATTY resource is freely available for interactive access and download.
5 0.10414266 95 emnlp-2012-N-gram-based Tense Models for Statistical Machine Translation
Author: Zhengxian Gong ; Min Zhang ; Chew Lim Tan ; Guodong Zhou
Abstract: Tense is a small element to a sentence, however, error tense can raise odd grammars and result in misunderstanding. Recently, tense has drawn attention in many natural language processing applications. However, most of current Statistical Machine Translation (SMT) systems mainly depend on translation model and language model. They never consider and make full use of tense information. In this paper, we propose n-gram-based tense models for SMT and successfully integrate them into a state-of-the-art phrase-based SMT system via two additional features. Experimental results on the NIST Chinese-English translation task show that our proposed tense models are very effective, contributing performance improvement by 0.62 BLUE points over a strong baseline. 1
6 0.09650401 4 emnlp-2012-A Comparison of Vector-based Representations for Semantic Composition
7 0.088139921 112 emnlp-2012-Resolving Complex Cases of Definite Pronouns: The Winograd Schema Challenge
8 0.080479644 135 emnlp-2012-Using Discourse Information for Paraphrase Extraction
9 0.079309657 62 emnlp-2012-Identifying Constant and Unique Relations by using Time-Series Text
10 0.071932152 136 emnlp-2012-Weakly Supervised Training of Semantic Parsers
11 0.071576409 100 emnlp-2012-Open Language Learning for Information Extraction
12 0.067886256 7 emnlp-2012-A Novel Discriminative Framework for Sentence-Level Discourse Analysis
13 0.066066138 72 emnlp-2012-Joint Inference for Event Timeline Construction
14 0.064612143 21 emnlp-2012-Assessment of ESL Learners' Syntactic Competence Based on Similarity Measures
15 0.062419124 93 emnlp-2012-Multi-instance Multi-label Learning for Relation Extraction
16 0.059264299 12 emnlp-2012-A Transition-Based System for Joint Part-of-Speech Tagging and Labeled Non-Projective Dependency Parsing
17 0.05915539 14 emnlp-2012-A Weakly Supervised Model for Sentence-Level Semantic Orientation Analysis with Multiple Experts
18 0.058476243 29 emnlp-2012-Concurrent Acquisition of Word Meaning and Lexical Categories
19 0.058209565 116 emnlp-2012-Semantic Compositionality through Recursive Matrix-Vector Spaces
20 0.057982702 123 emnlp-2012-Syntactic Transfer Using a Bilingual Lexicon
topicId topicWeight
[(0, 0.216), (1, 0.103), (2, -0.046), (3, -0.01), (4, 0.121), (5, 0.09), (6, 0.096), (7, 0.066), (8, -0.065), (9, 0.044), (10, -0.159), (11, -0.06), (12, -0.088), (13, 0.093), (14, -0.031), (15, -0.124), (16, -0.225), (17, -0.072), (18, 0.078), (19, 0.05), (20, -0.01), (21, -0.068), (22, 0.074), (23, 0.152), (24, -0.03), (25, -0.079), (26, -0.03), (27, -0.08), (28, 0.005), (29, -0.027), (30, -0.045), (31, -0.121), (32, -0.019), (33, 0.029), (34, 0.139), (35, 0.034), (36, -0.084), (37, -0.023), (38, -0.224), (39, 0.059), (40, -0.05), (41, -0.028), (42, 0.141), (43, -0.099), (44, -0.073), (45, 0.121), (46, 0.007), (47, -0.149), (48, -0.021), (49, -0.112)]
simIndex simValue paperId paperTitle
same-paper 1 0.97187734 80 emnlp-2012-Learning Verb Inference Rules from Linguistically-Motivated Evidence
Author: Hila Weisman ; Jonathan Berant ; Idan Szpektor ; Ido Dagan
Abstract: Learning inference relations between verbs is at the heart of many semantic applications. However, most prior work on learning such rules focused on a rather narrow set of information sources: mainly distributional similarity, and to a lesser extent manually constructed verb co-occurrence patterns. In this paper, we claim that it is imperative to utilize information from various textual scopes: verb co-occurrence within a sentence, verb cooccurrence within a document, as well as overall corpus statistics. To this end, we propose a much richer novel set of linguistically motivated cues for detecting entailment between verbs and combine them as features in a supervised classification framework. We empirically demonstrate that our model significantly outperforms previous methods and that information from each textual scope contributes to the verb entailment learning task.
2 0.60895473 95 emnlp-2012-N-gram-based Tense Models for Statistical Machine Translation
Author: Zhengxian Gong ; Min Zhang ; Chew Lim Tan ; Guodong Zhou
Abstract: Tense is a small element to a sentence, however, error tense can raise odd grammars and result in misunderstanding. Recently, tense has drawn attention in many natural language processing applications. However, most of current Statistical Machine Translation (SMT) systems mainly depend on translation model and language model. They never consider and make full use of tense information. In this paper, we propose n-gram-based tense models for SMT and successfully integrate them into a state-of-the-art phrase-based SMT system via two additional features. Experimental results on the NIST Chinese-English translation task show that our proposed tense models are very effective, contributing performance improvement by 0.62 BLUE points over a strong baseline. 1
3 0.54122704 62 emnlp-2012-Identifying Constant and Unique Relations by using Time-Series Text
Author: Yohei Takaku ; Nobuhiro Kaji ; Naoki Yoshinaga ; Masashi Toyoda
Abstract: Because the real world evolves over time, numerous relations between entities written in presently available texts are already obsolete or will potentially evolve in the future. This study aims at resolving the intricacy in consistently compiling relations extracted from text, and presents a method for identifying constancy and uniqueness of the relations in the context of supervised learning. We exploit massive time-series web texts to induce features on the basis of time-series frequency and linguistic cues. Experimental results confirmed that the time-series frequency distributions contributed much to the recall of constancy identification and the precision of the uniqueness identification.
4 0.52490145 103 emnlp-2012-PATTY: A Taxonomy of Relational Patterns with Semantic Types
Author: Ndapandula Nakashole ; Gerhard Weikum ; Fabian Suchanek
Abstract: This paper presents PATTY: a large resource for textual patterns that denote binary relations between entities. The patterns are semantically typed and organized into a subsumption taxonomy. The PATTY system is based on efficient algorithms for frequent itemset mining and can process Web-scale corpora. It harnesses the rich type system and entity population of large knowledge bases. The PATTY taxonomy comprises 350,569 pattern synsets. Random-sampling-based evaluation shows a pattern accuracy of 84.7%. PATTY has 8,162 subsumptions, with a random-sampling-based precision of 75%. The PATTY resource is freely available for interactive access and download.
5 0.49552026 16 emnlp-2012-Aligning Predicates across Monolingual Comparable Texts using Graph-based Clustering
Author: Michael Roth ; Anette Frank
Abstract: Generating coherent discourse is an important aspect in natural language generation. Our aim is to learn factors that constitute coherent discourse from data, with a focus on how to realize predicate-argument structures in a model that exceeds the sentence level. We present an important subtask for this overall goal, in which we align predicates across comparable texts, admitting partial argument structure correspondence. The contribution of this work is two-fold: We first construct a large corpus resource of comparable texts, including an evaluation set with manual predicate alignments. Secondly, we present a novel approach for aligning predicates across comparable texts using graph-based clustering with Mincuts. Our method significantly outperforms other alignment techniques when applied to this novel alignment task, by a margin of at least 6.5 percentage points in F1-score.
6 0.48032477 100 emnlp-2012-Open Language Learning for Information Extraction
7 0.45963454 40 emnlp-2012-Ensemble Semantics for Large-scale Unsupervised Relation Extraction
8 0.42905611 38 emnlp-2012-Employing Compositional Semantics and Discourse Consistency in Chinese Event Extraction
9 0.3665863 7 emnlp-2012-A Novel Discriminative Framework for Sentence-Level Discourse Analysis
10 0.29361555 112 emnlp-2012-Resolving Complex Cases of Definite Pronouns: The Winograd Schema Challenge
11 0.28593129 135 emnlp-2012-Using Discourse Information for Paraphrase Extraction
12 0.28547689 107 emnlp-2012-Polarity Inducing Latent Semantic Analysis
13 0.28445432 29 emnlp-2012-Concurrent Acquisition of Word Meaning and Lexical Categories
14 0.27958134 3 emnlp-2012-A Coherence Model Based on Syntactic Patterns
15 0.2795434 46 emnlp-2012-Exploiting Reducibility in Unsupervised Dependency Parsing
16 0.27713963 4 emnlp-2012-A Comparison of Vector-based Representations for Semantic Composition
17 0.27382779 26 emnlp-2012-Building a Lightweight Semantic Model for Unsupervised Information Extraction on Short Listings
18 0.26570898 79 emnlp-2012-Learning Syntactic Categories Using Paradigmatic Representations of Word Context
19 0.26321512 85 emnlp-2012-Local and Global Context for Supervised and Unsupervised Metonymy Resolution
20 0.26093492 97 emnlp-2012-Natural Language Questions for the Web of Data
topicId topicWeight
[(2, 0.018), (16, 0.031), (25, 0.013), (34, 0.035), (45, 0.01), (60, 0.075), (63, 0.052), (64, 0.482), (65, 0.047), (70, 0.014), (73, 0.019), (74, 0.05), (76, 0.038), (80, 0.02), (86, 0.011), (95, 0.025)]
simIndex simValue paperId paperTitle
same-paper 1 0.88834858 80 emnlp-2012-Learning Verb Inference Rules from Linguistically-Motivated Evidence
Author: Hila Weisman ; Jonathan Berant ; Idan Szpektor ; Ido Dagan
Abstract: Learning inference relations between verbs is at the heart of many semantic applications. However, most prior work on learning such rules focused on a rather narrow set of information sources: mainly distributional similarity, and to a lesser extent manually constructed verb co-occurrence patterns. In this paper, we claim that it is imperative to utilize information from various textual scopes: verb co-occurrence within a sentence, verb cooccurrence within a document, as well as overall corpus statistics. To this end, we propose a much richer novel set of linguistically motivated cues for detecting entailment between verbs and combine them as features in a supervised classification framework. We empirically demonstrate that our model significantly outperforms previous methods and that information from each textual scope contributes to the verb entailment learning task.
2 0.81531888 38 emnlp-2012-Employing Compositional Semantics and Discourse Consistency in Chinese Event Extraction
Author: Peifeng Li ; Guodong Zhou ; Qiaoming Zhu ; Libin Hou
Abstract: Current Chinese event extraction systems suffer much from two problems in trigger identification: unknown triggers and word segmentation errors to known triggers. To resolve these problems, this paper proposes two novel inference mechanisms to explore special characteristics in Chinese via compositional semantics inside Chinese triggers and discourse consistency between Chinese trigger mentions. Evaluation on the ACE 2005 Chinese corpus justifies the effectiveness of our approach over a strong baseline. 1
3 0.73053873 34 emnlp-2012-Do Neighbours Help? An Exploration of Graph-based Algorithms for Cross-domain Sentiment Classification
Author: Natalia Ponomareva ; Mike Thelwall
Abstract: This paper presents a comparative study of graph-based approaches for cross-domain sentiment classification. In particular, the paper analyses two existing methods: an optimisation problem and a ranking algorithm. We compare these graph-based methods with each other and with the other state-ofthe-art approaches and conclude that graph domain representations offer a competitive solution to the domain adaptation problem. Analysis of the best parameters for graphbased algorithms reveals that there are no optimal values valid for all domain pairs and that these values are dependent on the characteristics of corresponding domains.
4 0.59225518 129 emnlp-2012-Type-Supervised Hidden Markov Models for Part-of-Speech Tagging with Incomplete Tag Dictionaries
Author: Dan Garrette ; Jason Baldridge
Abstract: Past work on learning part-of-speech taggers from tag dictionaries and raw data has reported good results, but the assumptions made about those dictionaries are often unrealistic: due to historical precedents, they assume access to information about labels in the raw and test sets. Here, we demonstrate ways to learn hidden Markov model taggers from incomplete tag dictionaries. Taking the MINGREEDY algorithm (Ravi et al., 2010) as a starting point, we improve it with several intuitive heuristics. We also define a simple HMM emission initialization that takes advantage of the tag dictionary and raw data to capture both the openness of a given tag and its estimated prevalence in the raw data. Altogether, our augmentations produce improvements to per- formance over the original MIN-GREEDY algorithm for both English and Italian data.
5 0.39307982 14 emnlp-2012-A Weakly Supervised Model for Sentence-Level Semantic Orientation Analysis with Multiple Experts
Author: Lizhen Qu ; Rainer Gemulla ; Gerhard Weikum
Abstract: We propose the weakly supervised MultiExperts Model (MEM) for analyzing the semantic orientation of opinions expressed in natural language reviews. In contrast to most prior work, MEM predicts both opinion polarity and opinion strength at the level of individual sentences; such fine-grained analysis helps to understand better why users like or dislike the entity under review. A key challenge in this setting is that it is hard to obtain sentence-level training data for both polarity and strength. For this reason, MEM is weakly supervised: It starts with potentially noisy indicators obtained from coarse-grained training data (i.e., document-level ratings), a small set of diverse base predictors, and, if available, small amounts of fine-grained training data. We integrate these noisy indicators into a unified probabilistic framework using ideas from ensemble learning and graph-based semi-supervised learning. Our experiments indicate that MEM outperforms state-of-the-art methods by a significant margin.
6 0.38574517 95 emnlp-2012-N-gram-based Tense Models for Statistical Machine Translation
7 0.36629152 92 emnlp-2012-Multi-Domain Learning: When Do Domains Matter?
8 0.36517808 30 emnlp-2012-Constructing Task-Specific Taxonomies for Document Collection Browsing
9 0.36460978 26 emnlp-2012-Building a Lightweight Semantic Model for Unsupervised Information Extraction on Short Listings
10 0.3624534 71 emnlp-2012-Joint Entity and Event Coreference Resolution across Documents
11 0.35589069 72 emnlp-2012-Joint Inference for Event Timeline Construction
12 0.34320778 138 emnlp-2012-Wiki-ly Supervised Part-of-Speech Tagging
13 0.33637619 124 emnlp-2012-Three Dependency-and-Boundary Models for Grammar Induction
14 0.3320488 27 emnlp-2012-Characterizing Stylistic Elements in Syntactic Structure
15 0.33018324 136 emnlp-2012-Weakly Supervised Training of Semantic Parsers
16 0.32999286 103 emnlp-2012-PATTY: A Taxonomy of Relational Patterns with Semantic Types
17 0.32707915 77 emnlp-2012-Learning Constraints for Consistent Timeline Extraction
18 0.32350892 64 emnlp-2012-Improved Parsing and POS Tagging Using Inter-Sentence Consistency Constraints
19 0.32276624 114 emnlp-2012-Revisiting the Predictability of Language: Response Completion in Social Media
20 0.32237086 135 emnlp-2012-Using Discourse Information for Paraphrase Extraction