emnlp emnlp2011 emnlp2011-92 knowledge-graph by maker-knowledge-mining
Source: pdf
Author: Quang Do ; Yee Seng Chan ; Dan Roth
Abstract: This paper develops a minimally supervised approach, based on focused distributional similarity methods and discourse connectives, for identifying of causality relations between events in context. While it has been shown that distributional similarity can help identifying causality, we observe that discourse connectives and the particular discourse relation they evoke in context provide additional information towards determining causality between events. We show that combining discourse relation predictions and distributional similarity methods in a global inference procedure provides additional improvements towards determining event causality.
Reference: text
sentIndex sentText sentNum sentScore
1 edu l ino s Abstract This paper develops a minimally supervised approach, based on focused distributional similarity methods and discourse connectives, for identifying of causality relations between events in context. [sent-2, score-1.185]
2 While it has been shown that distributional similarity can help identifying causality, we observe that discourse connectives and the particular discourse relation they evoke in context provide additional information towards determining causality between events. [sent-3, score-1.484]
3 We show that combining discourse relation predictions and distributional similarity methods in a global inference procedure provides additional improvements towards determining event causality. [sent-4, score-0.777]
4 1 Introduction An important part of text understanding arises from understanding the semantics of events described in the narrative, such as identifying the events that are mentioned and how they are related semantically. [sent-5, score-0.372]
5 For instance, when given a sentence “The police arrested him because he killed someone. [sent-6, score-0.389]
6 ”, humans understand that there are two events, triggered by the words “arrested” and “killed”, and that there is a causality relationship between these two events. [sent-7, score-0.557]
7 Besides being an important component of discourse understanding, automatically identifying causal relations between events is important for various natural language processing (NLP) applications such as question answering, etc. [sent-8, score-0.737]
8 In this work, we automatically detect and extract causal relations between events in text. [sent-9, score-0.377]
9 294 Despite its importance, prior work on event causality extraction in context in the NLP literature is relatively sparse. [sent-10, score-0.766]
10 In (Girju, 2003), the author used noun-verb-noun lexico-syntactic patterns to learn that “mosquitoes cause malaria”, where the cause and effect mentions are nominals and not necessarily event evoking words. [sent-11, score-0.482]
11 , 2007), the authors focused on detecting causality between search query pairs in temporal query logs. [sent-13, score-0.51]
12 (Beamer and Girju, 2009) tried to detect causal relations between verbs in a corpus of screen plays, but limited themselves to consecutive, or adjacent verb pairs. [sent-14, score-0.262]
13 In (Riaz and Girju, 2010), the authors first cluster sentences into topic-specific scenarios, and then focus on building a dataset of causal text spans, where each span is headed by a verb. [sent-15, score-0.227]
14 Thus, their focus was not on identifying causal relations between events in a given text document. [sent-16, score-0.443]
15 In this paper, given a text document, we first identify events and their associated arguments. [sent-17, score-0.181]
16 We then identify causality or relatedness relations between event pairs. [sent-18, score-0.858]
17 To do this, we develop a minimally supervised approach using focused distributional similarity methods, such as co-occurrence counts of events collected automatically from an unannotated corpus, to measure and predict existence of causality relations between event pairs. [sent-19, score-1.182]
18 Then, we build on the observation that discourse connectives and the particular discourse relation they evoke in context provide additional information towards determining causality between events. [sent-20, score-1.389]
19 However, knowing that the connective “because” evokes a contingency discourse relation between the text spans “The police arrested him” and “he killed someone” provides further evidence towards predicting causality. [sent-24, score-1.078]
20 The contributions of this paper are summarized below: • • Our focus is on identifying causality between eOvuernt f pairs sin o nco indteenxtitf. [sent-25, score-0.502]
21 nSgin cceau esvaleintyts b are eoefnten triggered by either verbs (e. [sent-26, score-0.131]
22 “explosion”), we allow for detection of causality between verb-verb, verb-noun, and noun-noun triggered event pairs. [sent-30, score-0.859]
23 We developed a minimally supervised approach fvoelro othpeed dtas ak using foalcluyse sdu pdeisrtvriisbeudtion apalsimilarity methods that are automatically collected from an unannotated corpus. [sent-32, score-0.084]
24 • 2 We leverage on the interactions between event causality prediction taenrda tdioisncso burestwe er enla etivoennst prediction. [sent-34, score-0.793]
25 This allows us to easily define appropriate constraints to ensure that the causality and discourse predictions are coherent with each other, thereby improving the performance ofcausality identification. [sent-36, score-0.786]
26 Event Causality In this work, we define an event as an action or occurrence that happens with associated participants or arguments. [sent-37, score-0.333]
27 Examples of predicates could be verbs such as “attacked”, “employs”, nouns such as “explosion”, 295 “protest”, etc. [sent-45, score-0.247]
28 , and examples of the arguments of “attacked” could be its subject and object nouns. [sent-46, score-0.107]
29 To measure the causality association between a pair of events ei and ej (in general, ei and ej could be extracted from the same or different documents), we should use information gathered about their predicates and arguments. [sent-47, score-1.577]
30 However, this leads to very sparse counts as the predicate pi with its list of arguments ai1, . [sent-55, score-0.25]
31 , ain would rarely co-occur (within some reasonable context distance) with predicate pj and its entire list of arguments ,. [sent-58, score-0.321]
32 , Hence, in this work, we measure causality association using three separate components and focused distributional similarity methods collected about event pairs as described in the rest of this section. [sent-61, score-0.848]
33 In our work, we regard each event e as being triggered and rooted at a predicate p. [sent-65, score-0.458]
34 1 Predicate-Predicate Association We define spp as follows: spp(ei, ej) = PMI(pi,pj) ×IDF(pi,pj) max(ui, uj) Dist(pi,pj) (2) which takes into account the PMI between predicates pi and pj of events ei and ej respectively, as well as various other pieces of information. [sent-68, score-1.022]
35 In Suppes’ Probabilistic theory of Casuality (Suppes, 1970), he highlighted that event e is a possible cause of event e0, if e0 happens more frequently with e than × 1PMI is frequently used to measure association between variables. [sent-69, score-0.757]
36 Next, we build on the intuition that event predicates appearing in a large number of documents are probably not important or discriminative. [sent-74, score-0.43]
37 Thus, we penalize these predicates when calculating spp by adopting the inverse document frequency (idf): IDF(pi,pj) = idf(pi) idf(pj) idf(pi,pj), where idf(p) = log1+DN, D is the total number of documents in the collection and N is the number of documents that p occurs in. [sent-75, score-0.242]
38 If pi and pj are drawn from the same sentence, the numerator of the above fraction will return 1. [sent-77, score-0.2]
39 In our work, we set ws to 3 and thus, if pi occurs in sentence k, the furthest sentence that pj will be drawn from, is sentence k 2. [sent-78, score-0.229]
40 The final component of Equation 2, max(ui, uj), takes into account whether predicates (events) pi and pj appear most frequently with each other. [sent-79, score-0.338]
41 ui and uj are defined as follows: + ui=maxk[P(pi,Ppk(p)]i −,pj P)(pi,pj) + ? [sent-80, score-0.167]
42 ui will be maximized if there is no other predicate pk having a higher co-occurrence proba- bility with pi, i. [sent-85, score-0.169]
43 2 Predicate-Argument and Argument-Argument Association We define spa as follows: spa(ei,ej) =|A1ej|a∈XAejPMI(pi,a) +|A1ei|a∈XAeiPMI(pj,a), where Aei and Aej are the sets of arguments of and ej respectively. [sent-91, score-0.463]
44 Finally, we define saa as follows: saa(ei, ej) = |Aei|1|Aej|a∈XAeia0X∈AejPMI(a,a0) (3) ei (4) Together, spa and saa provide additional contexts and robustness (in addition to spp) for measuring the cause-effect association between events ei and ej . [sent-92, score-1.025]
45 , where ECD(a,b) measures the causality between two events a and b (headed by verbs), and the second component in the ECD equation is similar to Dist(pi, pj). [sent-97, score-0.617]
46 So far, our definitions in this section are generic and allow for any list of event argument types. [sent-99, score-0.327]
47 We describe how we extract event predicates and their associated arguments in the section below. [sent-101, score-0.488]
48 3 Verbal and Nominal Predicates We consider that events are not only triggered by verbs but also by nouns. [sent-102, score-0.284]
49 On the other hand, since events are also frequently triggered by nominal predicates, it is important to identify an appropriate list of event triggering nouns. [sent-104, score-0.642]
50 In our work, we gathered such a list using the following approach: • • • We first gather a list of deverbal nouns from the sWete ofifr smt goastth frequently occurring (in sth fero Gigaword corpus) 3,000 verbal predicate types. [sent-105, score-0.444]
51 For each verb type v, we go through all its WordNet2 senses and gather all its derivationally related nouns Nv 3. [sent-106, score-0.132]
52 From Nv, we heuristically remove nouns that are lmess N than three characters in length. [sent-107, score-0.128]
53 We also remove nouns whose first three characters are different from the first three characters ofv. [sent-108, score-0.152]
54 For each of the remaining nouns in Nv, we measeuacrehd oitfs Lheev reenmsateiinni (edit) dnisst ainnc Ne from v and keep the noun(s) with the minimum distance. [sent-109, score-0.104]
55 When multiple nouns have the same minimum distance from v, we keep all of them. [sent-110, score-0.104]
56 To further prune the list of nouns, we next re- Tmoo vfeudrt haellr nouns ending oinf “er”, “or”, or “ee”, as these nouns typically refer to a person, e. [sent-111, score-0.233]
57 We also remove nouns that are not hyponyms (children) of the first WordNet sense of the noun “event”4. [sent-114, score-0.129]
58 • Since we are concerned with nouns denoting events, Fera amreeN coent (Ruppenhofer uetn al. [sent-115, score-0.144]
59 As part of the FN resource, each FN frame consists of a list of lexical units (mainly verbs and nouns) representing the semantics of the frame. [sent-118, score-0.089]
60 Various frame-to-frame relations are also defined (in particular the inheritance relation). [sent-119, score-0.092]
61 Hence, we gathered all the children frames of the FN frame “Event”. [sent-120, score-0.148]
62 From these children frames, we then gathered all their noun lexical units (words) and add them to our list of 2http://wordnet. [sent-121, score-0.132]
63 Finally, we also add a few nouns denoting natural disaster from Wikipedia5. [sent-127, score-0.19]
64 Using the above approach, we gathered a list of about 2,000 noun types. [sent-128, score-0.107]
65 This current approach is heuristics based which we intend to improve in the future, and any such improvements should subsequently improve the performance of our causality identification approach. [sent-129, score-0.487]
66 Event triggering deverbal nouns could have associated arguments (for instance, acting as subject, object of the deverbal noun). [sent-130, score-0.405]
67 4 Discourse and Causality Discourse connectives are important for relating dif- ferent text spans, helping us to understand a piece of text in relation to its context: [The police arrested him] because [he killed someone]. [sent-136, score-0.653]
68 In the example sentence above, the discourse connective (“because”) and the discourse relation it evokes (in this case, the Cause relation) allows readers to relate its two associated text spans, “The police arrested him” and “he killed someone”. [sent-137, score-1.25]
69 Also, notice that the verbs “arrested” and “killed”, which cross the two text spans, are causally related. [sent-138, score-0.125]
70 To aid in extracting causal relations, we leverage on the identification of discourse relations to provide additional contextual information. [sent-139, score-0.618]
71 To identify discourse relations, we use the Penn Discourse Treebank (PDTB) (Prasad et al. [sent-140, score-0.322]
72 , 2007), which contains annotations of discourse relations in context. [sent-141, score-0.436]
73 The annotations are done over the Wall Street Journal corpus and the PDTB adopts a predicate-argument view of discourse relations. [sent-142, score-0.344]
74 In the rest of this section, we briefly describe the discourse relations in PDTB and highlight how we might leverage them to aid in determining event causality. [sent-146, score-0.824]
75 1 Discourse Relations PDTB contains annotations for four coarse-grained discourse relation types, as shown in the left column of Table 1. [sent-151, score-0.403]
76 Each of these are further refined into several fine-grained discourse relations, as shown in the right column of the table. [sent-152, score-0.322]
77 6 Next, we briefly describe these relations, highlighting those that could potentially help to determine event causality. [sent-153, score-0.324]
78 Comparison A Comparison discourse relation between two text spans highlights prominent differences between the situations described in the text spans. [sent-154, score-0.595]
79 , 2007), the truth of both spans is independent of the established discourse relation. [sent-157, score-0.431]
80 This means that the text spans are not causally related and thus, the existence of a Comparison relation should imply that there is no causality relation across the two text spans. [sent-158, score-0.836]
81 Contingency A Contingency relation between two text spans indicates that the situation described in one text span causally influences the situation in the other. [sent-159, score-0.314]
82 Existence of a Contingency relation potentially implies that there exists at least one causal event pair crossing the two text spans. [sent-161, score-0.521]
83 The PDTB annotation manual states that while the Cause and Condition discourse relations indicate casual influence in their text spans, there is no causal influence in the text spans of the Pragmatic-cause and Pragmatic-condition relations. [sent-162, score-0.711]
84 For instance, Pragmatic-condition indicates that one span pro6PDTB further refines these fine-grained relations into a final third level of relations, but we do not use them in this work. [sent-163, score-0.123]
85 Hence, there is a need to also identify fine-grained discourse relations. [sent-165, score-0.322]
86 Expansion Connectives evoking Expansion discourse relations expand the discourse, such as by providing additional information, illustrating alternative situations, etc. [sent-166, score-0.46]
87 Most of the Expansion fine-grained relations (except for Conjunction, which could connect arbitrary pieces of text spans) should not contain causality relations across its text spans. [sent-168, score-0.732]
88 Temporal These indicate that the situations described in the text spans are related temporally. [sent-169, score-0.186]
89 Temporal precedence of the (cause) event over the (effect) event is a necessary, but not sufficient requisite for causality. [sent-171, score-0.604]
90 Hence by itself, Temporal re- lations are probably not discriminative enough for determining event causality. [sent-172, score-0.362]
91 2 Discourse Relation Extraction System Our work follows the approach and features described in the state-of-the-art Ruby-based discourse system of (Lin et al. [sent-174, score-0.322]
92 , 2010), to build an inhouse Java-based discourse relation extraction system. [sent-175, score-0.381]
93 Our system identifies explicit connectives in text, predict their discourse relations, as well as their associated text spans. [sent-176, score-0.499]
94 , 2010), we achieved a competitive performance of slightly over 80% F1-score in identifying fine-grained relations for explicit connectives. [sent-178, score-0.13]
95 In the example sentences given thus far in this section, all the connectives were explicit, as they appear in the texts. [sent-182, score-0.149]
96 Identifying implicit connectives is a harder task and × incorporating these is a possible future work. [sent-184, score-0.149]
97 1) and the discourse system are forced to cohere with each other. [sent-188, score-0.322]
98 More importantly, this should improve the performance of using only CEA to extract causal event pairs. [sent-189, score-0.434]
99 To the best of our knowledge, this approach for causality extraction is novel. [sent-190, score-0.464]
100 1 CEA & Discourse: Implementation Details Let E denote the set of event mentions in a docuLmeetnt E. [sent-192, score-0.324]
wordName wordTfidf (topN-words)
[('causality', 0.464), ('discourse', 0.322), ('event', 0.302), ('ej', 0.268), ('arrested', 0.16), ('killed', 0.158), ('events', 0.153), ('connectives', 0.149), ('pdtb', 0.138), ('cea', 0.137), ('ecd', 0.137), ('spp', 0.137), ('causal', 0.132), ('ei', 0.131), ('pj', 0.119), ('saa', 0.114), ('spa', 0.114), ('spans', 0.109), ('idf', 0.109), ('predicates', 0.105), ('nouns', 0.104), ('triggered', 0.093), ('relations', 0.092), ('uj', 0.089), ('arguments', 0.081), ('pi', 0.081), ('deverbal', 0.079), ('ui', 0.078), ('police', 0.071), ('girju', 0.07), ('riaz', 0.069), ('contingency', 0.066), ('sent', 0.064), ('predicate', 0.063), ('pmi', 0.059), ('relation', 0.059), ('causally', 0.059), ('connective', 0.059), ('minimally', 0.059), ('gathered', 0.057), ('cause', 0.056), ('fn', 0.054), ('someone', 0.054), ('nv', 0.05), ('situations', 0.049), ('temporal', 0.046), ('aei', 0.046), ('aej', 0.046), ('attacked', 0.046), ('disaster', 0.046), ('evokes', 0.046), ('evoking', 0.046), ('gurevich', 0.046), ('maxk', 0.046), ('suppes', 0.046), ('dist', 0.044), ('denoting', 0.04), ('frames', 0.04), ('identifying', 0.038), ('verbs', 0.038), ('roth', 0.037), ('determining', 0.037), ('headed', 0.036), ('evoke', 0.036), ('maxt', 0.036), ('triggering', 0.036), ('distributional', 0.035), ('ain', 0.033), ('prasad', 0.033), ('frequently', 0.033), ('span', 0.031), ('happens', 0.031), ('expansion', 0.031), ('verbal', 0.03), ('existence', 0.03), ('yih', 0.029), ('ws', 0.029), ('pk', 0.028), ('gather', 0.028), ('pieces', 0.028), ('text', 0.028), ('leverage', 0.027), ('object', 0.026), ('frame', 0.026), ('children', 0.025), ('collected', 0.025), ('list', 0.025), ('noun', 0.025), ('readers', 0.025), ('ilp', 0.025), ('resource', 0.024), ('characters', 0.024), ('explosion', 0.024), ('probably', 0.023), ('identification', 0.023), ('mentions', 0.022), ('similarity', 0.022), ('briefly', 0.022), ('annotations', 0.022), ('aid', 0.022)]
simIndex simValue paperId paperTitle
same-paper 1 1.0 92 emnlp-2011-Minimally Supervised Event Causality Identification
Author: Quang Do ; Yee Seng Chan ; Dan Roth
Abstract: This paper develops a minimally supervised approach, based on focused distributional similarity methods and discourse connectives, for identifying of causality relations between events in context. While it has been shown that distributional similarity can help identifying causality, we observe that discourse connectives and the particular discourse relation they evoke in context provide additional information towards determining causality between events. We show that combining discourse relation predictions and distributional similarity methods in a global inference procedure provides additional improvements towards determining event causality.
2 0.28465772 94 emnlp-2011-Modelling Discourse Relations for Arabic
Author: Amal Al-Saif ; Katja Markert
Abstract: We present the first algorithms to automatically identify explicit discourse connectives and the relations they signal for Arabic text. First we show that, for Arabic news, most adjacent sentences are connected via explicit connectives in contrast to English, making the treatment of explicit discourse connectives for Arabic highly important. We also show that explicit Arabic discourse connectives are far more ambiguous than English ones, making their treatment challenging. In the second part of the paper, we present supervised algorithms to address automatic discourse connective identification and discourse relation recognition. Our connective identifier based on gold standard syntactic features achieves almost human performance. In addition, an identifier based solely on simple lexical and automatically derived morphological and POS features performs with high reliability, essential for languages that do not have high-quality parsers yet. Our algorithm for recognizing discourse relations performs significantly better than a baseline based on the connective surface string alone and therefore reduces the ambiguity in explicit connective interpretation.
3 0.20342599 142 emnlp-2011-Unsupervised Discovery of Discourse Relations for Eliminating Intra-sentence Polarity Ambiguities
Author: Lanjun Zhou ; Binyang Li ; Wei Gao ; Zhongyu Wei ; Kam-Fai Wong
Abstract: Polarity classification of opinionated sentences with both positive and negative sentiments1 is a key challenge in sentiment analysis. This paper presents a novel unsupervised method for discovering intra-sentence level discourse relations for eliminating polarity ambiguities. Firstly, a discourse scheme with discourse constraints on polarity was defined empirically based on Rhetorical Structure Theory (RST). Then, a small set of cuephrase-based patterns were utilized to collect a large number of discourse instances which were later converted to semantic sequential representations (SSRs). Finally, an unsupervised method was adopted to generate, weigh and filter new SSRs without cue phrases for recognizing discourse relations. Experimental results showed that the proposed methods not only effectively recognized the defined discourse relations but also achieved significant improvement by integrating discourse information in sentence-level polarity classification.
4 0.14723738 59 emnlp-2011-Fast and Robust Joint Models for Biomedical Event Extraction
Author: Sebastian Riedel ; Andrew McCallum
Abstract: Extracting biomedical events from literature has attracted much recent attention. The bestperforming systems so far have been pipelines of simple subtask-specific local classifiers. A natural drawback of such approaches are cascading errors introduced in early stages of the pipeline. We present three joint models of increasing complexity designed to overcome this problem. The first model performs joint trigger and argument extraction, and lends itself to a simple, efficient and exact inference algorithm. The second model captures correlations between events, while the third model ensures consistency between arguments of the same event. Inference in these models is kept tractable through dual decomposition. The first two models outperform the previous best joint approaches and are very competitive with respect to the current state-of-theart. The third model yields the best results reported so far on the BioNLP 2009 shared task, the BioNLP 2011 Genia task and the BioNLP 2011Infectious Diseases task.
5 0.14356372 113 emnlp-2011-Relation Acquisition using Word Classes and Partial Patterns
Author: Stijn De Saeger ; Kentaro Torisawa ; Masaaki Tsuchida ; Jun'ichi Kazama ; Chikara Hashimoto ; Ichiro Yamada ; Jong Hoon Oh ; Istvan Varga ; Yulan Yan
Abstract: This paper proposes a semi-supervised relation acquisition method that does not rely on extraction patterns (e.g. “X causes Y” for causal relations) but instead learns a combination of indirect evidence for the target relation semantic word classes and partial patterns. This method can extract long tail instances of semantic relations like causality from rare and complex expressions in a large Japanese Web corpus in extreme cases, patterns that occur only once in the entire corpus. Such patterns are beyond the reach ofcurrent pattern based methods. We show that our method performs on par with state-of-the-art pattern based methods, and maintains a reasonable level of accuracy even for instances — — acquired from infrequent patterns. This ability to acquire long tail instances is crucial for risk management and innovation, where an exhaustive database of high-level semantic relations like causation is of vital importance.
6 0.077527359 8 emnlp-2011-A Model of Discourse Predictions in Human Sentence Processing
7 0.07373973 14 emnlp-2011-A generative model for unsupervised discovery of relations and argument classes from clinical texts
8 0.068539731 49 emnlp-2011-Entire Relaxation Path for Maximum Entropy Problems
9 0.065933205 88 emnlp-2011-Linear Text Segmentation Using Affinity Propagation
10 0.062857799 7 emnlp-2011-A Joint Model for Extended Semantic Role Labeling
11 0.062134154 61 emnlp-2011-Generating Aspect-oriented Multi-Document Summarization with Event-aspect model
12 0.061925724 114 emnlp-2011-Relation Extraction with Relation Topics
13 0.060336228 128 emnlp-2011-Structured Relation Discovery using Generative Models
14 0.055797644 81 emnlp-2011-Learning General Connotation of Words using Graph-based Algorithms
15 0.053866848 105 emnlp-2011-Predicting Thread Discourse Structure over Technical Web Forums
16 0.052279867 144 emnlp-2011-Unsupervised Learning of Selectional Restrictions and Detection of Argument Coercions
17 0.047612015 57 emnlp-2011-Extreme Extraction - Machine Reading in a Week
18 0.046143547 84 emnlp-2011-Learning the Information Status of Noun Phrases in Spoken Dialogues
19 0.044972483 78 emnlp-2011-Large-Scale Noun Compound Interpretation Using Bootstrapping and the Web as a Corpus
20 0.044393543 145 emnlp-2011-Unsupervised Semantic Role Induction with Graph Partitioning
topicId topicWeight
[(0, 0.147), (1, -0.109), (2, -0.147), (3, 0.042), (4, 0.022), (5, -0.24), (6, 0.152), (7, 0.068), (8, -0.127), (9, -0.43), (10, -0.151), (11, -0.06), (12, 0.106), (13, -0.023), (14, 0.069), (15, -0.148), (16, -0.045), (17, 0.089), (18, 0.005), (19, 0.049), (20, 0.083), (21, -0.065), (22, 0.152), (23, 0.148), (24, 0.034), (25, -0.161), (26, -0.009), (27, -0.005), (28, 0.076), (29, -0.034), (30, 0.01), (31, -0.07), (32, 0.037), (33, 0.032), (34, -0.011), (35, 0.015), (36, -0.03), (37, -0.065), (38, 0.053), (39, -0.055), (40, -0.064), (41, 0.053), (42, 0.03), (43, -0.04), (44, 0.022), (45, 0.037), (46, -0.014), (47, -0.01), (48, -0.001), (49, -0.038)]
simIndex simValue paperId paperTitle
same-paper 1 0.97245497 92 emnlp-2011-Minimally Supervised Event Causality Identification
Author: Quang Do ; Yee Seng Chan ; Dan Roth
Abstract: This paper develops a minimally supervised approach, based on focused distributional similarity methods and discourse connectives, for identifying of causality relations between events in context. While it has been shown that distributional similarity can help identifying causality, we observe that discourse connectives and the particular discourse relation they evoke in context provide additional information towards determining causality between events. We show that combining discourse relation predictions and distributional similarity methods in a global inference procedure provides additional improvements towards determining event causality.
2 0.8781296 94 emnlp-2011-Modelling Discourse Relations for Arabic
Author: Amal Al-Saif ; Katja Markert
Abstract: We present the first algorithms to automatically identify explicit discourse connectives and the relations they signal for Arabic text. First we show that, for Arabic news, most adjacent sentences are connected via explicit connectives in contrast to English, making the treatment of explicit discourse connectives for Arabic highly important. We also show that explicit Arabic discourse connectives are far more ambiguous than English ones, making their treatment challenging. In the second part of the paper, we present supervised algorithms to address automatic discourse connective identification and discourse relation recognition. Our connective identifier based on gold standard syntactic features achieves almost human performance. In addition, an identifier based solely on simple lexical and automatically derived morphological and POS features performs with high reliability, essential for languages that do not have high-quality parsers yet. Our algorithm for recognizing discourse relations performs significantly better than a baseline based on the connective surface string alone and therefore reduces the ambiguity in explicit connective interpretation.
Author: Lanjun Zhou ; Binyang Li ; Wei Gao ; Zhongyu Wei ; Kam-Fai Wong
Abstract: Polarity classification of opinionated sentences with both positive and negative sentiments1 is a key challenge in sentiment analysis. This paper presents a novel unsupervised method for discovering intra-sentence level discourse relations for eliminating polarity ambiguities. Firstly, a discourse scheme with discourse constraints on polarity was defined empirically based on Rhetorical Structure Theory (RST). Then, a small set of cuephrase-based patterns were utilized to collect a large number of discourse instances which were later converted to semantic sequential representations (SSRs). Finally, an unsupervised method was adopted to generate, weigh and filter new SSRs without cue phrases for recognizing discourse relations. Experimental results showed that the proposed methods not only effectively recognized the defined discourse relations but also achieved significant improvement by integrating discourse information in sentence-level polarity classification.
4 0.41142681 59 emnlp-2011-Fast and Robust Joint Models for Biomedical Event Extraction
Author: Sebastian Riedel ; Andrew McCallum
Abstract: Extracting biomedical events from literature has attracted much recent attention. The bestperforming systems so far have been pipelines of simple subtask-specific local classifiers. A natural drawback of such approaches are cascading errors introduced in early stages of the pipeline. We present three joint models of increasing complexity designed to overcome this problem. The first model performs joint trigger and argument extraction, and lends itself to a simple, efficient and exact inference algorithm. The second model captures correlations between events, while the third model ensures consistency between arguments of the same event. Inference in these models is kept tractable through dual decomposition. The first two models outperform the previous best joint approaches and are very competitive with respect to the current state-of-theart. The third model yields the best results reported so far on the BioNLP 2009 shared task, the BioNLP 2011 Genia task and the BioNLP 2011Infectious Diseases task.
5 0.34108904 8 emnlp-2011-A Model of Discourse Predictions in Human Sentence Processing
Author: Amit Dubey ; Frank Keller ; Patrick Sturt
Abstract: This paper introduces a psycholinguistic model of sentence processing which combines a Hidden Markov Model noun phrase chunker with a co-reference classifier. Both models are fully incremental and generative, giving probabilities of lexical elements conditional upon linguistic structure. This allows us to compute the information theoretic measure of surprisal, which is known to correlate with human processing effort. We evaluate our surprisal predictions on the Dundee corpus of eye-movement data show that our model achieve a better fit with human reading times than a syntax-only model which does not have access to co-reference information.
6 0.31051081 113 emnlp-2011-Relation Acquisition using Word Classes and Partial Patterns
7 0.29202551 14 emnlp-2011-A generative model for unsupervised discovery of relations and argument classes from clinical texts
8 0.23870264 114 emnlp-2011-Relation Extraction with Relation Topics
9 0.20949979 88 emnlp-2011-Linear Text Segmentation Using Affinity Propagation
10 0.20912176 57 emnlp-2011-Extreme Extraction - Machine Reading in a Week
11 0.2071908 70 emnlp-2011-Identifying Relations for Open Information Extraction
12 0.195649 78 emnlp-2011-Large-Scale Noun Compound Interpretation Using Bootstrapping and the Web as a Corpus
13 0.18405668 61 emnlp-2011-Generating Aspect-oriented Multi-Document Summarization with Event-aspect model
14 0.17694257 144 emnlp-2011-Unsupervised Learning of Selectional Restrictions and Detection of Argument Coercions
15 0.17664941 49 emnlp-2011-Entire Relaxation Path for Maximum Entropy Problems
16 0.17437898 7 emnlp-2011-A Joint Model for Extended Semantic Role Labeling
17 0.1656417 53 emnlp-2011-Experimental Support for a Categorical Compositional Distributional Model of Meaning
18 0.16510756 128 emnlp-2011-Structured Relation Discovery using Generative Models
19 0.16503276 81 emnlp-2011-Learning General Connotation of Words using Graph-based Algorithms
20 0.15805173 105 emnlp-2011-Predicting Thread Discourse Structure over Technical Web Forums
topicId topicWeight
[(14, 0.429), (15, 0.014), (23, 0.107), (36, 0.021), (37, 0.02), (45, 0.061), (57, 0.015), (62, 0.023), (64, 0.016), (66, 0.028), (69, 0.038), (79, 0.041), (82, 0.012), (90, 0.012), (96, 0.044), (98, 0.011)]
simIndex simValue paperId paperTitle
same-paper 1 0.71637803 92 emnlp-2011-Minimally Supervised Event Causality Identification
Author: Quang Do ; Yee Seng Chan ; Dan Roth
Abstract: This paper develops a minimally supervised approach, based on focused distributional similarity methods and discourse connectives, for identifying of causality relations between events in context. While it has been shown that distributional similarity can help identifying causality, we observe that discourse connectives and the particular discourse relation they evoke in context provide additional information towards determining causality between events. We show that combining discourse relation predictions and distributional similarity methods in a global inference procedure provides additional improvements towards determining event causality.
2 0.67877126 127 emnlp-2011-Structured Lexical Similarity via Convolution Kernels on Dependency Trees
Author: Danilo Croce ; Alessandro Moschitti ; Roberto Basili
Abstract: Alessandro Moschitti DISI University of Trento 38123 Povo (TN), Italy mo s chitt i di s i @ .unit n . it Roberto Basili DII University of Tor Vergata 00133 Roma, Italy bas i i info .uni roma2 . it l@ over semantic networks, e.g. (Cowie et al., 1992; Wu and Palmer, 1994; Resnik, 1995; Jiang and Conrath, A central topic in natural language processing is the design of lexical and syntactic fea- tures suitable for the target application. In this paper, we study convolution dependency tree kernels for automatic engineering of syntactic and semantic patterns exploiting lexical similarities. We define efficient and powerful kernels for measuring the similarity between dependency structures, whose surface forms of the lexical nodes are in part or completely different. The experiments with such kernels for question classification show an unprecedented results, e.g. 41% of error reduction of the former state-of-the-art. Additionally, semantic role classification confirms the benefit of semantic smoothing for dependency kernels.
3 0.40295681 147 emnlp-2011-Using Syntactic and Semantic Structural Kernels for Classifying Definition Questions in Jeopardy!
Author: Alessandro Moschitti ; Jennifer Chu-carroll ; Siddharth Patwardhan ; James Fan ; Giuseppe Riccardi
Abstract: The last decade has seen many interesting applications of Question Answering (QA) technology. The Jeopardy! quiz show is certainly one of the most fascinating, from the viewpoints of both its broad domain and the complexity of its language. In this paper, we study kernel methods applied to syntactic/semantic structures for accurate classification of Jeopardy! definition questions. Our extensive empirical analysis shows that our classification models largely improve on classifiers based on word-language models. Such classifiers are also used in the state-of-the-art QA pipeline constituting Watson, the IBM Jeopardy! system. Our experiments measuring their impact on Watson show enhancements in QA accuracy and a consequent increase in the amount of money earned in game-based evaluation.
4 0.33889848 68 emnlp-2011-Hypotheses Selection Criteria in a Reranking Framework for Spoken Language Understanding
Author: Marco Dinarelli ; Sophie Rosset
Abstract: Reranking models have been successfully applied to many tasks of Natural Language Processing. However, there are two aspects of this approach that need a deeper investigation: (i) Assessment of hypotheses generated for reranking at classification phase: baseline models generate a list of hypotheses and these are used for reranking without any assessment; (ii) Detection of cases where reranking models provide a worst result: the best hypothesis provided by the reranking model is assumed to be always the best result. In some cases the reranking model provides an incorrect hypothesis while the baseline best hypothesis is correct, especially when baseline models are accurate. In this paper we propose solutions for these two aspects: (i) a semantic inconsistency metric to select possibly more correct n-best hypotheses, from a large set generated by an SLU basiline model. The selected hypotheses are reranked applying a state-of-the-art model based on Partial Tree Kernels, which encode SLU hypotheses in Support Vector Machines with complex structured features; (ii) finally, we apply a decision strategy, based on confidence values, to select the final hypothesis between the first ranked hypothesis provided by the baseline SLU model and the first ranked hypothesis provided by the re-ranker. We show the effectiveness of these solutions presenting comparative results obtained reranking hypotheses generated by a very accurate Conditional Random Field model. We evaluate our approach on the French MEDIA corpus. The results show significant improvements with respect to current state-of-the-art and previous 1104 Sophie Rosset LIMSI-CNRS B.P. 133, 91403 Orsay Cedex France ro s set @ l ims i fr . re-ranking models.
5 0.31407255 108 emnlp-2011-Quasi-Synchronous Phrase Dependency Grammars for Machine Translation
Author: Kevin Gimpel ; Noah A. Smith
Abstract: We present a quasi-synchronous dependency grammar (Smith and Eisner, 2006) for machine translation in which the leaves of the tree are phrases rather than words as in previous work (Gimpel and Smith, 2009). This formulation allows us to combine structural components of phrase-based and syntax-based MT in a single model. We describe a method of extracting phrase dependencies from parallel text using a target-side dependency parser. For decoding, we describe a coarse-to-fine approach based on lattice dependency parsing of phrase lattices. We demonstrate performance improvements for Chinese-English and UrduEnglish translation over a phrase-based baseline. We also investigate the use of unsupervised dependency parsers, reporting encouraging preliminary results.
6 0.313512 128 emnlp-2011-Structured Relation Discovery using Generative Models
7 0.31226599 1 emnlp-2011-A Bayesian Mixture Model for PoS Induction Using Multiple Features
8 0.31124914 28 emnlp-2011-Closing the Loop: Fast, Interactive Semi-Supervised Annotation With Queries on Features and Instances
9 0.31079116 137 emnlp-2011-Training dependency parsers by jointly optimizing multiple objectives
10 0.31055641 109 emnlp-2011-Random Walk Inference and Learning in A Large Scale Knowledge Base
11 0.30976528 59 emnlp-2011-Fast and Robust Joint Models for Biomedical Event Extraction
12 0.30931267 8 emnlp-2011-A Model of Discourse Predictions in Human Sentence Processing
13 0.30793294 132 emnlp-2011-Syntax-Based Grammaticality Improvement using CCG and Guided Search
14 0.30759233 98 emnlp-2011-Named Entity Recognition in Tweets: An Experimental Study
15 0.30716035 136 emnlp-2011-Training a Parser for Machine Translation Reordering
16 0.30710599 35 emnlp-2011-Correcting Semantic Collocation Errors with L1-induced Paraphrases
17 0.30604643 126 emnlp-2011-Structural Opinion Mining for Graph-based Sentiment Representation
18 0.3059817 70 emnlp-2011-Identifying Relations for Open Information Extraction
19 0.30559823 65 emnlp-2011-Heuristic Search for Non-Bottom-Up Tree Structure Prediction
20 0.30482763 79 emnlp-2011-Lateen EM: Unsupervised Training with Multiple Objectives, Applied to Dependency Grammar Induction