emnlp emnlp2010 emnlp2010-103 knowledge-graph by maker-knowledge-mining
Source: pdf
Author: Roi Reichart ; Ari Rappoport
Abstract: Polysemy is a major characteristic of natural languages. Like words, syntactic forms can have several meanings. Understanding the correct meaning of a syntactic form is of great importance to many NLP applications. In this paper we address an important type of syntactic polysemy the multiple possible senses of tense syntactic forms. We make our discussion concrete by introducing the task of Tense Sense Disambiguation (TSD): given a concrete tense syntactic form present in a sentence, select its appropriate sense among a set of possible senses. Using English grammar textbooks, we compiled a syntactic sense dictionary comprising common tense syntactic forms and semantic senses for each. We annotated thousands of BNC sentences using the – defined senses. We describe a supervised TSD algorithm trained on these annotations, which outperforms a strong baseline for the task.
Reference: text
sentIndex sentText sentNum sentScore
1 Understanding the correct meaning of a syntactic form is of great importance to many NLP applications. [sent-6, score-0.091]
2 In this paper we address an important type of syntactic polysemy the multiple possible senses of tense syntactic forms. [sent-7, score-0.56]
3 We make our discussion concrete by introducing the task of Tense Sense Disambiguation (TSD): given a concrete tense syntactic form present in a sentence, select its appropriate sense among a set of possible senses. [sent-8, score-0.642]
4 Using English grammar textbooks, we compiled a syntactic sense dictionary comprising common tense syntactic forms and semantic senses for each. [sent-9, score-0.696]
5 1 Introduction The function of syntax is to combine words to express meanings, using syntactic devices such as word order, auxiliary words, and morphology (Goldberg, 1995). [sent-12, score-0.15]
6 Like words, concrete syntactic forms (the sentence words generated by specific syntactic devices) can have several meanings. [sent-16, score-0.249]
7 Both contain the concrete syntactic form ‘are play- ing’, generated by the abstract syntactic form usually known as ‘present progressive’ (am/is/are + Ving). [sent-22, score-0.258]
8 Note that the polysemy is of the syntactic form as a unit, not of individual words. [sent-24, score-0.131]
9 In particular, the verb ‘play’ is used in the same sense in both cases. [sent-25, score-0.141]
10 In this paper we address a prominent type of syntactic form polysemy: the multiple possible senses that tense syntactic forms can have. [sent-26, score-0.577]
11 Disambiguating the polysemy of tense forms is of theoretical and practical importance (Section 2). [sent-27, score-0.373]
12 To make our discussion concrete, we introduce the task of Tense Sense Disambiguation (TSD): given a concrete tense syntactic form in a sentence, select its correct sense among a given set of possible senses (Section 3). [sent-28, score-0.641]
13 The disambiguation of polysemy is a fundamental problem in NLP. [sent-29, score-0.113]
14 For experimenting with the TSD task, we com- piled an English syntactic sense dictionary based on a thorough study of three major English grammar projects (Section 4). [sent-32, score-0.259]
15 We selected 3000 sentences from the British National Corpus containing 4702 concrete syntactic forms, and annotated each of these by its sense (Section 5). [sent-33, score-0.267]
16 2 TSD Motivation In this work we follow linguistics theories that posit that tense does not directly reflect conceptual time as one might think. [sent-41, score-0.279]
17 Dinsmore (1991) and Cutrer (1994) explain that the same tense may end up indicating very different objective time relations relative to the sentence production time. [sent-42, score-0.279]
18 In the following sentences, the present tense corresponds to the future time: (1) The boat leaves next week. [sent-44, score-0.256]
19 In contrast, the following present tense sentences talk about events that happened in the past: (1) Iam walking down the street one day when suddenly this guy walks up to me. [sent-47, score-0.279]
20 According to this model, the grammar specifies partial constraints on time and fact/prediction status that hold locally between mental spaces within a discourse configuration. [sent-60, score-0.094]
21 Accordingly, the same tense may end up indicating very different objective time relations relative to the 326 speech event. [sent-62, score-0.279]
22 Words, multiword expressions, and syntactic forms are all valid constructions. [sent-65, score-0.125]
23 It is thus very natural to address the sense disambiguation problem for all of these. [sent-66, score-0.174]
24 For many NLP applications, it is very important to disambiguate the tense forms of the sentence. [sent-68, score-0.33]
25 The most common devices in English are word order, morphology, and the usage of auxiliary words. [sent-77, score-0.102]
26 A Concrete Syntactic Form (CSF) is a concrete set of words generated by an ASF for expressing a certain meaning in an utterance1. [sent-79, score-0.118]
27 The TSD task is to disambiguate the semantic sense of a tense syntactic form. [sent-91, score-0.444]
28 While the verb ‘arrive’ has two main senses: ‘reach a place’, and ‘begin’, as in ‘Summer has arrived’, in that example we focused on the disambiguation of the tense sense of the ‘arrives’ construction. [sent-95, score-0.449]
29 In all cases, we need to disambiguate the sense of the ASFs. [sent-99, score-0.14]
30 4 The Syntactic Sense Dictionary A prerequisite to any concrete experimentation with the TSD task is a syntactic sense dictionary. [sent-103, score-0.267]
31 Based on a thorough examination of three major English grammar projects, we compiled a set of 18 common English tense ASFs and their possible senses. [sent-104, score-0.322]
32 As in any sense dictionary, in many cases it is hard to draw the line between senses. [sent-108, score-0.122]
33 In order to be able to explore the computational limits of the task, we have adopted a policy of fine sense granularity. [sent-109, score-0.141]
34 For example, senses 1and 3 of the ‘present simple’ ASF in Table 1 can be argued to be quite similar to each other, having a very fine semantic distinction. [sent-110, score-0.115]
35 A specific application may choose to collapse some senses into one. [sent-111, score-0.096]
36 , the ‘present simple’ ASF can be used to refer to future, not present, events, as in Table 1, sense 4). [sent-114, score-0.122]
37 Note that the first four ASFs are not direct tense forms; we include them because they involve tensed sub-sentences whose disambiguation is necessary for disambigua- tion of the whole ASF. [sent-116, score-0.332]
38 The total number of possible senses for these 18 ASFs is 103. [sent-117, score-0.096]
39 Table 1 shows the complete senses set for the ‘present simple’ and ‘be + to + infinitive’ ASFs, plus an example sentence for each sense. [sent-118, score-0.096]
40 Space limitations prevent us from listing all form senses here; we will make the listing available online. [sent-119, score-0.118]
41 To select the 3000 sentences, we randomly sampled sentences from the various written and spoken sections of the and ‘be + to + infinitive’ abstract syntactic forms (ASFs), with an example for each. [sent-123, score-0.104]
42 To make sure that our definition of auxiliary words does not skew the sampling process, and to obtain ASFs that do not have clear auxiliary words, we have also added 1000 random sentences. [sent-128, score-0.118]
43 All 328 senses are represented; the number of senses represented by at least 15 CSFs is 77 (out of 103, average number of CSFs per sense is 45. [sent-130, score-0.314]
44 We implemented an interactive application that displays a sentence and asks an annotator to (1) mark words that participate in the CSFs contained in the sentence; (2) specify the ASF(s) of these CSFs; and (3) select the appropriate ASF sense from the set of possible senses. [sent-132, score-0.122]
45 7%, and the inter-annotator agreement for the senses was 84. [sent-138, score-0.096]
46 First, note that the syntactic sense is not easy to deduce from readily computable annotations such as the sentence’s POS tagging, dependency structure, or parse tree (see Section 8). [sent-142, score-0.17]
47 The possible verb tags in this tagset are: VB for the base form, VBD for past tense, VBN for past participle, VBG for a present participle or gerund (-ing), VBP for present tense that is not 3rd person singular, and VBZ for present simple 3rd person singular. [sent-168, score-0.349]
48 A{(sx noted in }Section 3, there are two versions of the task, one in which Ci includes the totality of sense labels, and one in which it includes only the labels associated with a particular ASF. [sent-177, score-0.164]
49 For the task version in which the ASF 5These are all in a past form due to the semantics of the reported speech form. [sent-181, score-0.079]
50 The classifier decides what is the type of an ASF according to the POS tag of its verb and to its auxiliary words (given in the annotation). [sent-192, score-0.187]
51 For example, if we see the auxiliary phrase ‘had been’ and the verb POS is not VBG, then the ASF is ‘past perfect simple’ . [sent-193, score-0.123]
52 In this scenario, when given a test CSF, Xn+1, its set of possible labels Cn+1 is defined by the classifier output. [sent-197, score-0.08]
53 In the features in which ASF type is used (see table 2), it is taken from the classifier output in this case. [sent-198, score-0.131]
54 In addition we used this data to design the rules ofthe ASF type classifier (which is not statistical and does not have a training phase). [sent-217, score-0.109]
55 These two conditions differ in whether the type is taken from the gold standard annotation of the test sentences (TypeKnown), or from the output of the simple rule-based classifier (TypeClassifier, see Section 6). [sent-225, score-0.132]
56 For both conditions, the results reported below are when both ASF type features and possible labels sets are provided during training by the manual annotation. [sent-226, score-0.095]
57 %7n%own Table 3: Performance of our algorithm and of the MFS baseline where at test time ASF type is known (right), unknown (left) or given by a simple rule-based classifier (middle). [sent-240, score-0.153]
58 When an unconstrained classifier is used, POS features affect the results both when ASF type features are used and when they are not (see discussion in the text). [sent-247, score-0.21]
59 In the condition where the ASF type is not known at test time, MFS gives each form in the test set the sense that was the overall most frequent in the training set. [sent-250, score-0.26]
60 When the ASF type is known at test time, MFS gives each test CSF the most frequent sense of that ASF type in the training set. [sent-252, score-0.245]
61 That is, in this case all CSFs having the same ASF type get the same sense, and forms of different types are guaranteed to get different senses. [sent-253, score-0.107]
62 Recall that the condition where ASF type is known at test time is further divided to two conditions. [sent-254, score-0.139]
63 In the TypeKnown condition, MFS selects the most frequent sense of the manually created ASF type, while in the TypeClassifier condition it selects the most frequent sense of the type decided by the rule-based classifier. [sent-255, score-0.339]
64 Note that a random baseline which selects a sense for every test CSF from a uniform distribution over the possible senses (103 in our case) would score very poorly. [sent-257, score-0.218]
65 Results are shown where ASF type is not known at test time 331 (left), when it is decided at test time by a rule-based classifier (middle) and when it is known at test time (right). [sent-260, score-0.22]
66 We use the learning model of Even-Zohar and Roth (2001), which allows us to constrain the possible senses an input vector can get to the senses of its ASF type. [sent-272, score-0.192]
67 In this case, the only difference between the TypeKnown and the TypeUnknown conditions is whether ASF type features are encoded at test time. [sent-274, score-0.118]
68 9% (when using training and test time constraints and ASF type features) to 53% (when using only ASF type features but no constraints). [sent-276, score-0.173]
69 03% (when neither constraints nor ASF type features are used). [sent-279, score-0.099]
70 Note 10Recall that the performance of the rule-based ASF type classifier on test data is not 100% but 91. [sent-280, score-0.109]
71 that the difference between the constrained model and the unconstrained model is quite large. [sent-283, score-0.082]
72 Thus, the algorithm outperforms the baseline both when the constrained model is used and when an unconstrained multi-class classifier is used. [sent-287, score-0.14]
73 Note also that when constraints on the possible labels are available at training time, test time constraints and ASF type features (whose inclusion is the difference between the TypeKnown and TypeUnknown) have a minor effect on the results (57. [sent-288, score-0.17]
74 However, when training time constraints on the possible labels are not available at training time, ASF type features alone do have a significant effect on the result (53% for TypeKnown compared to 48. [sent-291, score-0.144]
75 Verb forms provide some partial information corresponding to the ASF type features encoded at the TypeKnown scenario. [sent-296, score-0.151]
76 Table 4 shows that when both label constraints and ASF type features are used, POS features have almost no impact on the final results. [sent-297, score-0.146]
77 When the constrained model is used but ASF type features are not encoded, POS features have an effect on the results. [sent-298, score-0.12]
78 We conclude that when using the constrained model, POS features are important mainly for ASF type information. [sent-299, score-0.098]
79 When the unconstrained classifier is used, POS features have an effect on performance whether ASF type features are encoded or not. [sent-300, score-0.232]
80 In other words, when using an unconstrained classifier, POS features give more than ASF type information to to the model. [sent-302, score-0.13]
81 The conditionals and ‘wish’ features have a more substantial impact on the results, as they have a role in defining the overall syntactic structure of the sentence. [sent-311, score-0.131]
82 4% degradation in model accuracy when using the constrained and unconstrained models respectively. [sent-313, score-0.082]
83 The references above also describe some unsupervised word sense induction algorithms. [sent-325, score-0.122]
84 For each tensed clause, PUNDIT first decides whether it refers to an actual time (as in ‘We flew TWA to Boston’) or not (as in ‘Tourists flew TWA to Boston’, or ‘John always flew his own plane to Boston’). [sent-331, score-0.155]
85 The temporal structure of actual time clauses is then further analyzed. [sent-332, score-0.08]
86 We are not aware of further research that followed their sense disambiguation direction. [sent-336, score-0.174]
87 Current temporal reasoning research focuses on temporal ordering of events (e. [sent-337, score-0.159]
88 This direction is very different from TSD, which deals with the semantics of individual concrete tense syntactic forms. [sent-343, score-0.421]
89 A potential application of TSD is machine trans- lation where it can assist in translating tense and aspect. [sent-345, score-0.256]
90 Indeed several papers have explored tense and aspect in the MT context. [sent-346, score-0.256]
91 Dorr (1992) explored the integration of tense and aspect information with lexical semantics for machine translation. [sent-347, score-0.276]
92 Schiehlen (2000) analyzed the effect tense understanding has on MT. [sent-348, score-0.256]
93 Ye and Zhang (2005) explored tense tagging in a cross-lingual context. [sent-349, score-0.256]
94 , (2006) extracted features for tense translation between Chinese and English. [sent-351, score-0.278]
95 , (2007) compared the performance of several MT systems in translating tense and aspect and found that various ML techniques perform better on the task. [sent-353, score-0.256]
96 9 Conclusion and Future Work In this paper we introduced the Tense Sense Disambiguation (TSD) task, defined as selecting the correct sense of a concrete tense syntactic form in a sentence among the senses of abstract syntactic forms 333 in a syntactic sense dictionary. [sent-360, score-0.915]
97 Unlike in other semantic disambiguation tasks, the sense to be disambiguated is not lexical but of a syntactic structure. [sent-361, score-0.222]
98 We prepared a syntactic sense dictionary, annotated a corpus by it, and developed a supervised classifier for sense disambiguation that outperformed a strong baseline. [sent-362, score-0.402]
99 For example, we saw that seeing the full paragraph containing a sentence helps human annotators decide on the appropriate sense which implies that using larger contexts may improve the algorithm. [sent-364, score-0.122]
100 In fact, TSD can assist textual entailment as well, since the sense of a tense form may provide substantial information about the relations entailed from the sentence. [sent-367, score-0.45]
wordName wordTfidf (topN-words)
[('asf', 0.716), ('tsd', 0.343), ('tense', 0.256), ('csf', 0.206), ('typeknown', 0.17), ('csfs', 0.146), ('mfs', 0.125), ('sense', 0.122), ('asfs', 0.097), ('typeunknown', 0.097), ('concrete', 0.097), ('senses', 0.096), ('progressive', 0.065), ('typeclassifier', 0.061), ('polysemy', 0.061), ('auxiliary', 0.059), ('classifier', 0.058), ('temporal', 0.057), ('unconstrained', 0.057), ('forms', 0.056), ('pos', 0.055), ('disambiguation', 0.052), ('wsd', 0.051), ('type', 0.051), ('pundit', 0.049), ('snow', 0.048), ('syntactic', 0.048), ('grammar', 0.045), ('perfect', 0.045), ('condition', 0.044), ('devices', 0.043), ('arrives', 0.042), ('past', 0.037), ('conditionals', 0.036), ('flew', 0.036), ('willis', 0.036), ('agirre', 0.032), ('infinitive', 0.032), ('edmonds', 0.031), ('ensemble', 0.029), ('naive', 0.028), ('entailment', 0.026), ('winnow', 0.026), ('constraints', 0.026), ('comprising', 0.025), ('ye', 0.025), ('impact', 0.025), ('sequential', 0.025), ('constrained', 0.025), ('cn', 0.024), ('biber', 0.024), ('brother', 0.024), ('chess', 0.024), ('cobuild', 0.024), ('evening', 0.024), ('fauconnier', 0.024), ('harpercollins', 0.024), ('jerusalem', 0.024), ('longman', 0.024), ('mackay', 0.024), ('martinez', 0.024), ('murata', 0.024), ('tempeval', 0.024), ('tensed', 0.024), ('twa', 0.024), ('vbg', 0.024), ('verhagen', 0.024), ('goldberg', 0.024), ('textual', 0.024), ('classifiers', 0.023), ('conditions', 0.023), ('events', 0.023), ('experimenting', 0.023), ('time', 0.023), ('encoded', 0.022), ('form', 0.022), ('meanings', 0.022), ('features', 0.022), ('reasoning', 0.022), ('labels', 0.022), ('essential', 0.021), ('bayes', 0.021), ('known', 0.021), ('meaning', 0.021), ('dave', 0.021), ('penntreebank', 0.021), ('thorough', 0.021), ('bnc', 0.021), ('multiword', 0.021), ('semantics', 0.02), ('versions', 0.02), ('verb', 0.019), ('fine', 0.019), ('baldwin', 0.019), ('eneko', 0.019), ('navigli', 0.019), ('murphy', 0.019), ('tanaka', 0.019), ('disambiguate', 0.018), ('raymond', 0.018)]
simIndex simValue paperId paperTitle
same-paper 1 0.99999899 103 emnlp-2010-Tense Sense Disambiguation: A New Syntactic Polysemy Task
Author: Roi Reichart ; Ari Rappoport
Abstract: Polysemy is a major characteristic of natural languages. Like words, syntactic forms can have several meanings. Understanding the correct meaning of a syntactic form is of great importance to many NLP applications. In this paper we address an important type of syntactic polysemy the multiple possible senses of tense syntactic forms. We make our discussion concrete by introducing the task of Tense Sense Disambiguation (TSD): given a concrete tense syntactic form present in a sentence, select its appropriate sense among a set of possible senses. Using English grammar textbooks, we compiled a syntactic sense dictionary comprising common tense syntactic forms and semantic senses for each. We annotated thousands of BNC sentences using the – defined senses. We describe a supervised TSD algorithm trained on these annotations, which outperforms a strong baseline for the task.
2 0.056587767 124 emnlp-2010-Word Sense Induction Disambiguation Using Hierarchical Random Graphs
Author: Ioannis Klapaftis ; Suresh Manandhar
Abstract: Graph-based methods have gained attention in many areas of Natural Language Processing (NLP) including Word Sense Disambiguation (WSD), text summarization, keyword extraction and others. Most of the work in these areas formulate their problem in a graph-based setting and apply unsupervised graph clustering to obtain a set of clusters. Recent studies suggest that graphs often exhibit a hierarchical structure that goes beyond simple flat clustering. This paper presents an unsupervised method for inferring the hierarchical grouping of the senses of a polysemous word. The inferred hierarchical structures are applied to the problem of word sense disambiguation, where we show that our method performs sig- nificantly better than traditional graph-based methods and agglomerative clustering yielding improvements over state-of-the-art WSD systems based on sense induction.
3 0.05414471 122 emnlp-2010-WikiWars: A New Corpus for Research on Temporal Expressions
Author: Pawel Mazur ; Robert Dale
Abstract: The reliable extraction of knowledge from text requires an appropriate treatment of the time at which reported events take place. Unfortunately, there are very few annotated data sets that support the development of techniques for event time-stamping and tracking the progression of time through a narrative. In this paper, we present a new corpus of temporally-rich documents sourced from English Wikipedia, which we have annotated with TIMEX2 tags. The corpus contains around 120000 tokens, and 2600 TIMEX2 expressions, thus comparing favourably in size to other existing corpora used in these areas. We describe the prepa- ration of the corpus, and compare the profile of the data with other existing temporally annotated corpora. We also report the results obtained when we use DANTE, our temporal expression tagger, to process this corpus, and point to where further work is required. The corpus is publicly available for research purposes.
4 0.049530119 66 emnlp-2010-Inducing Word Senses to Improve Web Search Result Clustering
Author: Roberto Navigli ; Giuseppe Crisafulli
Abstract: In this paper, we present a novel approach to Web search result clustering based on the automatic discovery of word senses from raw text, a task referred to as Word Sense Induction (WSI). We first acquire the senses (i.e., meanings) of a query by means of a graphbased clustering algorithm that exploits cycles (triangles and squares) in the co-occurrence graph of the query. Then we cluster the search results based on their semantic similarity to the induced word senses. Our experiments, conducted on datasets of ambiguous queries, show that our approach improves search result clustering in terms of both clustering quality and degree of diversification.
5 0.049159192 77 emnlp-2010-Measuring Distributional Similarity in Context
Author: Georgiana Dinu ; Mirella Lapata
Abstract: The computation of meaning similarity as operationalized by vector-based models has found widespread use in many tasks ranging from the acquisition of synonyms and paraphrases to word sense disambiguation and textual entailment. Vector-based models are typically directed at representing words in isolation and thus best suited for measuring similarity out of context. In his paper we propose a probabilistic framework for measuring similarity in context. Central to our approach is the intuition that word meaning is represented as a probability distribution over a set of latent senses and is modulated by context. Experimental results on lexical substitution and word similarity show that our algorithm outperforms previously proposed models.
6 0.042510696 114 emnlp-2010-Unsupervised Parse Selection for HPSG
7 0.037626483 33 emnlp-2010-Cross Language Text Classification by Model Translation and Semi-Supervised Learning
8 0.036199484 7 emnlp-2010-A Mixture Model with Sharing for Lexical Semantics
9 0.035960283 61 emnlp-2010-Improving Gender Classification of Blog Authors
10 0.034675293 9 emnlp-2010-A New Approach to Lexical Disambiguation of Arabic Text
11 0.033512436 95 emnlp-2010-SRL-Based Verb Selection for ESL
12 0.031677637 64 emnlp-2010-Incorporating Content Structure into Text Analysis Applications
13 0.031025674 107 emnlp-2010-Towards Conversation Entailment: An Empirical Investigation
14 0.030739682 21 emnlp-2010-Automatic Discovery of Manner Relations and its Applications
15 0.030656528 106 emnlp-2010-Top-Down Nearly-Context-Sensitive Parsing
16 0.030455818 111 emnlp-2010-Two Decades of Unsupervised POS Induction: How Far Have We Come?
17 0.030144319 97 emnlp-2010-Simple Type-Level Unsupervised POS Tagging
18 0.029343231 11 emnlp-2010-A Semi-Supervised Approach to Improve Classification of Infrequent Discourse Relations Using Feature Vector Extension
19 0.029200824 116 emnlp-2010-Using Universal Linguistic Knowledge to Guide Grammar Induction
20 0.029040098 71 emnlp-2010-Latent-Descriptor Clustering for Unsupervised POS Induction
topicId topicWeight
[(0, 0.116), (1, 0.058), (2, 0.006), (3, 0.049), (4, -0.021), (5, 0.006), (6, -0.048), (7, -0.026), (8, 0.037), (9, 0.016), (10, 0.023), (11, -0.016), (12, 0.031), (13, -0.035), (14, -0.007), (15, -0.077), (16, -0.041), (17, 0.035), (18, -0.044), (19, 0.029), (20, 0.073), (21, -0.015), (22, -0.034), (23, -0.138), (24, -0.001), (25, -0.006), (26, 0.021), (27, -0.051), (28, 0.026), (29, 0.078), (30, -0.217), (31, -0.015), (32, -0.037), (33, 0.021), (34, 0.041), (35, -0.0), (36, 0.188), (37, -0.121), (38, 0.381), (39, 0.073), (40, -0.319), (41, 0.235), (42, 0.167), (43, -0.112), (44, -0.152), (45, 0.231), (46, 0.018), (47, 0.099), (48, -0.232), (49, -0.074)]
simIndex simValue paperId paperTitle
same-paper 1 0.94296169 103 emnlp-2010-Tense Sense Disambiguation: A New Syntactic Polysemy Task
Author: Roi Reichart ; Ari Rappoport
Abstract: Polysemy is a major characteristic of natural languages. Like words, syntactic forms can have several meanings. Understanding the correct meaning of a syntactic form is of great importance to many NLP applications. In this paper we address an important type of syntactic polysemy the multiple possible senses of tense syntactic forms. We make our discussion concrete by introducing the task of Tense Sense Disambiguation (TSD): given a concrete tense syntactic form present in a sentence, select its appropriate sense among a set of possible senses. Using English grammar textbooks, we compiled a syntactic sense dictionary comprising common tense syntactic forms and semantic senses for each. We annotated thousands of BNC sentences using the – defined senses. We describe a supervised TSD algorithm trained on these annotations, which outperforms a strong baseline for the task.
2 0.44793078 122 emnlp-2010-WikiWars: A New Corpus for Research on Temporal Expressions
Author: Pawel Mazur ; Robert Dale
Abstract: The reliable extraction of knowledge from text requires an appropriate treatment of the time at which reported events take place. Unfortunately, there are very few annotated data sets that support the development of techniques for event time-stamping and tracking the progression of time through a narrative. In this paper, we present a new corpus of temporally-rich documents sourced from English Wikipedia, which we have annotated with TIMEX2 tags. The corpus contains around 120000 tokens, and 2600 TIMEX2 expressions, thus comparing favourably in size to other existing corpora used in these areas. We describe the prepa- ration of the corpus, and compare the profile of the data with other existing temporally annotated corpora. We also report the results obtained when we use DANTE, our temporal expression tagger, to process this corpus, and point to where further work is required. The corpus is publicly available for research purposes.
3 0.38335428 124 emnlp-2010-Word Sense Induction Disambiguation Using Hierarchical Random Graphs
Author: Ioannis Klapaftis ; Suresh Manandhar
Abstract: Graph-based methods have gained attention in many areas of Natural Language Processing (NLP) including Word Sense Disambiguation (WSD), text summarization, keyword extraction and others. Most of the work in these areas formulate their problem in a graph-based setting and apply unsupervised graph clustering to obtain a set of clusters. Recent studies suggest that graphs often exhibit a hierarchical structure that goes beyond simple flat clustering. This paper presents an unsupervised method for inferring the hierarchical grouping of the senses of a polysemous word. The inferred hierarchical structures are applied to the problem of word sense disambiguation, where we show that our method performs sig- nificantly better than traditional graph-based methods and agglomerative clustering yielding improvements over state-of-the-art WSD systems based on sense induction.
4 0.25170374 61 emnlp-2010-Improving Gender Classification of Blog Authors
Author: Arjun Mukherjee ; Bing Liu
Abstract: The problem of automatically classifying the gender of a blog author has important applications in many commercial domains. Existing systems mainly use features such as words, word classes, and POS (part-ofspeech) n-grams, for classification learning. In this paper, we propose two new techniques to improve the current result. The first technique introduces a new class of features which are variable length POS sequence patterns mined from the training data using a sequence pattern mining algorithm. The second technique is a new feature selection method which is based on an ensemble of several feature selection criteria and approaches. Empirical evaluation using a real-life blog data set shows that these two techniques improve the classification accuracy of the current state-ofthe-art methods significantly.
5 0.20979014 21 emnlp-2010-Automatic Discovery of Manner Relations and its Applications
Author: Eduardo Blanco ; Dan Moldovan
Abstract: This paper presents a method for the automatic discovery of MANNER relations from text. An extended definition of MANNER is proposed, including restrictions on the sorts of concepts that can be part of its domain and range. The connections with other relations and the lexico-syntactic patterns that encode MANNER are analyzed. A new feature set specialized on MANNER detection is depicted and justified. Experimental results show improvement over previous attempts to extract MANNER. Combinations of MANNER with other semantic relations are also discussed.
6 0.20793539 77 emnlp-2010-Measuring Distributional Similarity in Context
7 0.1853516 9 emnlp-2010-A New Approach to Lexical Disambiguation of Arabic Text
8 0.18458426 66 emnlp-2010-Inducing Word Senses to Improve Web Search Result Clustering
9 0.16903229 114 emnlp-2010-Unsupervised Parse Selection for HPSG
10 0.15869983 105 emnlp-2010-Title Generation with Quasi-Synchronous Grammar
11 0.15725105 54 emnlp-2010-Generating Confusion Sets for Context-Sensitive Error Correction
12 0.15233839 95 emnlp-2010-SRL-Based Verb Selection for ESL
13 0.14952499 81 emnlp-2010-Modeling Perspective Using Adaptor Grammars
14 0.14867343 113 emnlp-2010-Unsupervised Induction of Tree Substitution Grammars for Dependency Parsing
15 0.14447711 107 emnlp-2010-Towards Conversation Entailment: An Empirical Investigation
17 0.13982567 120 emnlp-2010-What's with the Attitude? Identifying Sentences with Attitude in Online Discussions
18 0.13477667 118 emnlp-2010-Utilizing Extra-Sentential Context for Parsing
19 0.13461554 117 emnlp-2010-Using Unknown Word Techniques to Learn Known Words
20 0.13071693 123 emnlp-2010-Word-Based Dialect Identification with Georeferenced Rules
topicId topicWeight
[(3, 0.022), (10, 0.014), (12, 0.036), (17, 0.014), (29, 0.112), (30, 0.018), (32, 0.023), (38, 0.305), (52, 0.016), (56, 0.071), (62, 0.018), (66, 0.1), (72, 0.063), (76, 0.036), (79, 0.012), (87, 0.013)]
simIndex simValue paperId paperTitle
1 0.79162759 68 emnlp-2010-Joint Inference for Bilingual Semantic Role Labeling
Author: Tao Zhuang ; Chengqing Zong
Abstract: We show that jointly performing semantic role labeling (SRL) on bitext can improve SRL results on both sides. In our approach, we use monolingual SRL systems to produce argument candidates for predicates in bitext at first. Then, we simultaneously generate SRL results for two sides of bitext using our joint inference model. Our model prefers the bilingual SRL result that is not only reasonable on each side of bitext, but also has more consistent argument structures between two sides. To evaluate the consistency between two argument structures, we also formulate a log-linear model to compute the probability of aligning two arguments. We have experimented with our model on Chinese-English parallel PropBank data. Using our joint inference model, F1 scores of SRL results on Chinese and English text achieve 79.53% and 77.87% respectively, which are 1.52 and 1.74 points higher than the results of baseline monolingual SRL combination systems respectively.
same-paper 2 0.7138654 103 emnlp-2010-Tense Sense Disambiguation: A New Syntactic Polysemy Task
Author: Roi Reichart ; Ari Rappoport
Abstract: Polysemy is a major characteristic of natural languages. Like words, syntactic forms can have several meanings. Understanding the correct meaning of a syntactic form is of great importance to many NLP applications. In this paper we address an important type of syntactic polysemy the multiple possible senses of tense syntactic forms. We make our discussion concrete by introducing the task of Tense Sense Disambiguation (TSD): given a concrete tense syntactic form present in a sentence, select its appropriate sense among a set of possible senses. Using English grammar textbooks, we compiled a syntactic sense dictionary comprising common tense syntactic forms and semantic senses for each. We annotated thousands of BNC sentences using the – defined senses. We describe a supervised TSD algorithm trained on these annotations, which outperforms a strong baseline for the task.
3 0.49385437 78 emnlp-2010-Minimum Error Rate Training by Sampling the Translation Lattice
Author: Samidh Chatterjee ; Nicola Cancedda
Abstract: Minimum Error Rate Training is the algorithm for log-linear model parameter training most used in state-of-the-art Statistical Machine Translation systems. In its original formulation, the algorithm uses N-best lists output by the decoder to grow the Translation Pool that shapes the surface on which the actual optimization is performed. Recent work has been done to extend the algorithm to use the entire translation lattice built by the decoder, instead of N-best lists. We propose here a third, intermediate way, consisting in growing the translation pool using samples randomly drawn from the translation lattice. We empirically measure a systematic im- provement in the BLEU scores compared to training using N-best lists, without suffering the increase in computational complexity associated with operating with the whole lattice.
4 0.49263656 69 emnlp-2010-Joint Training and Decoding Using Virtual Nodes for Cascaded Segmentation and Tagging Tasks
Author: Xian Qian ; Qi Zhang ; Yaqian Zhou ; Xuanjing Huang ; Lide Wu
Abstract: Many sequence labeling tasks in NLP require solving a cascade of segmentation and tagging subtasks, such as Chinese POS tagging, named entity recognition, and so on. Traditional pipeline approaches usually suffer from error propagation. Joint training/decoding in the cross-product state space could cause too many parameters and high inference complexity. In this paper, we present a novel method which integrates graph structures of two subtasks into one using virtual nodes, and performs joint training and decoding in the factorized state space. Experimental evaluations on CoNLL 2000 shallow parsing data set and Fourth SIGHAN Bakeoff CTB POS tagging data set demonstrate the superiority of our method over cross-product, pipeline and candidate reranking approaches.
5 0.49178743 65 emnlp-2010-Inducing Probabilistic CCG Grammars from Logical Form with Higher-Order Unification
Author: Tom Kwiatkowksi ; Luke Zettlemoyer ; Sharon Goldwater ; Mark Steedman
Abstract: This paper addresses the problem of learning to map sentences to logical form, given training data consisting of natural language sentences paired with logical representations of their meaning. Previous approaches have been designed for particular natural languages or specific meaning representations; here we present a more general method. The approach induces a probabilistic CCG grammar that represents the meaning of individual words and defines how these meanings can be combined to analyze complete sentences. We use higher-order unification to define a hypothesis space containing all grammars consistent with the training data, and develop an online learning algorithm that efficiently searches this space while simultaneously estimating the parameters of a log-linear parsing model. Experiments demonstrate high accuracy on benchmark data sets in four languages with two different meaning representations.
6 0.49064893 82 emnlp-2010-Multi-Document Summarization Using A* Search and Discriminative Learning
7 0.49037683 105 emnlp-2010-Title Generation with Quasi-Synchronous Grammar
8 0.49002627 87 emnlp-2010-Nouns are Vectors, Adjectives are Matrices: Representing Adjective-Noun Constructions in Semantic Space
9 0.48722562 7 emnlp-2010-A Mixture Model with Sharing for Lexical Semantics
10 0.48571843 18 emnlp-2010-Assessing Phrase-Based Translation Models with Oracle Decoding
11 0.48559758 32 emnlp-2010-Context Comparison of Bursty Events in Web Search and Online Media
12 0.48558024 86 emnlp-2010-Non-Isomorphic Forest Pair Translation
13 0.48455271 60 emnlp-2010-Improved Fully Unsupervised Parsing with Zoomed Learning
14 0.48427236 34 emnlp-2010-Crouching Dirichlet, Hidden Markov Model: Unsupervised POS Tagging with Context Local Tag Generation
15 0.48405975 84 emnlp-2010-NLP on Spoken Documents Without ASR
16 0.48388374 89 emnlp-2010-PEM: A Paraphrase Evaluation Metric Exploiting Parallel Texts
17 0.48283041 35 emnlp-2010-Discriminative Sample Selection for Statistical Machine Translation
18 0.48078591 63 emnlp-2010-Improving Translation via Targeted Paraphrasing
19 0.47954661 57 emnlp-2010-Hierarchical Phrase-Based Translation Grammars Extracted from Alignment Posterior Probabilities
20 0.47928855 45 emnlp-2010-Evaluating Models of Latent Document Semantics in the Presence of OCR Errors