emnlp emnlp2013 emnlp2013-121 knowledge-graph by maker-knowledge-mining
Source: pdf
Author: Swapna Gottipati ; Minghui Qiu ; Yanchuan Sim ; Jing Jiang ; Noah A. Smith
Abstract: We explore Debatepedia, a communityauthored encyclopedia of sociopolitical debates, as evidence for inferring a lowdimensional, human-interpretable representation in the domain of issues and positions. We introduce a generative model positing latent topics and cross-cutting positions that gives special treatment to person mentions and opinion words. We evaluate the resulting representation’s usefulness in attaching opinionated documents to arguments and its consistency with human judgments about positions.
Reference: text
sentIndex sentText sentNum sentScore
1 We introduce a generative model positing latent topics and cross-cutting positions that gives special treatment to person mentions and opinion words. [sent-12, score-0.475]
2 1 Introduction The social web has evolved into a forum for large portions of the population to discuss and debate complex issues of societal importance. [sent-14, score-0.506]
3 Websites like an online, community-authored encyclopedia of debates (§2), seek to organize some of Debatepedia,1 tchloisp exchange ibnattoe sstr (u§2c)t,ur seede kin tfoor omrgaatinoizn resources that summarize arguments and link externally to texts (editorials, blog posts, etc. [sent-15, score-0.4]
4 We view topics as lexicons, and propose that grounding a topic model with evidence beyond bags of words can lead to more lexicon-like repre- sentations. [sent-25, score-0.329]
5 Specifically, our generative topic model grounds topics using the hierarchical organization of arguments within Debatepedia. [sent-26, score-0.486]
6 Further, we use named entity recognition as a preprocessing step, an existing sentiment lexicon to construct an informed prior, and we incorporate a latent, discrete position variable that cuts across debates. [sent-27, score-0.25]
7 Subjectively, the model identifies reasonable topic a Snudb perspective terms, ealn did eitn atsifsioesci aretaess topics sensibly with important public figures. [sent-29, score-0.289]
8 In quantitative evaluations, we find the model’s representation superior to topics from vanilla latent Dirichlet allocation (Blei et al. [sent-30, score-0.194]
9 , 2003) and the joint sentiment topic model (Lin and He, 2009) in matching external texts to debates. [sent-31, score-0.273]
10 Further, the position variables can be used to infer the side of an argument within a debate; our model performs with an accuracy of 86% on position prediction of the debate argument. [sent-32, score-0.884]
11 The cross-cutting position variable is not especially consistent with human judgments, suggesting that further knowledge sources may be required to improve interpretability across issues. [sent-33, score-0.185]
12 2 Data Debatepedia, like Wikipedia, is constructed by volunteer contributors and has a system of community 2This variable might serve to cluster debate sides according to “abstract beliefs commonly shared by a group of people,” sometimes called ideologies (Van Dijk, 1998). [sent-34, score-0.691]
13 Many of the debate issues covered are controversial and salient in current public discourse. [sent-39, score-0.538]
14 Because it is primarily expressed as text, Debatepedia is a corpus of debate topics, but it is organized hierarchically, with multiple issues in each debate topic, questions within each issue, and arguments on two sides of each question. [sent-40, score-1.302]
15 An important feature of the corpus is the widespread quotation and linking to external articles on the web, including news stories, blog postings, wiki pages, and social media forums; here we use these external articles in evaluation (§4). [sent-41, score-0.351]
16 Table 1 shows excerpts from a debate page3 from Debatepedia. [sent-42, score-0.46]
17 Each debate contains “questions,” which reflect the different aspects of a debate. [sent-43, score-0.46]
18 Many of these arguments also contains links to online articles where the quotes are extracted from (not shown in Table 1). [sent-46, score-0.195]
19 4 Within a debate topic, the sides cut across different questions, aligning arguments together. [sent-48, score-0.704]
20 , the “Yes” sides of all the questions are consistent with the same side of the larger debate). [sent-59, score-0.253]
21 The example of Table 1deviates from this pattern, with the self-defense “Yes” arguing “no” to the high-level debate question—Should laws be passed to limit gun ownership further? [sent-60, score-0.532]
22 1 Preprocessing We scraped the Debatepedia website and extracted the debate, question, argument, and side structure of the debate topics. [sent-64, score-0.522]
23 We crawled the external web articles that were linked from the Debatepedia arguments. [sent-65, score-0.208]
24 Although our modeling approach ultimately treats texts as bags of terms (unigrams and bigrams), one important preprocessing step was taken to further improve the interpretability of the inferred representation: named entity mentions of persons. [sent-70, score-0.165]
25 2), we will sohuorw q uhaoliwta tthivise special str oeaft tmhee mnt oodfe person )m, weneti wonilsl enables the association of well-known individuals with debate topics. [sent-74, score-0.46]
26 Each term occurs in a context defined by the tuple hd, q, s, ai (respectively, a debate, a question ewi ttuhpilne th hde, debate, a side within the debate, and an argument). [sent-77, score-0.165]
27 At each level of the hierarchy is a different latent variable: • Each question q within debate d is associated Ewaicthh a duiessttriiobnuti qon w over topics, d den ioste adss θd,q. [sent-78, score-0.526]
28 8 • Each side s of the debate d is associated with a position, ede sn ooft ethde id,s aantde we posit a global tdhis atribution ι that cuts across different questions and arguments. [sent-79, score-0.614]
29 In our experiments, there are two positions, and the two sides of a debate are constrained to associate with opposing positions. [sent-80, score-0.592]
30 8In future work, more sharing across questions within a debate, or more differentiation among the topic distributions for arguments under a question, might be explored. [sent-89, score-0.349]
31 • Each term wd,q,s,a,n (n is the position index oEfa cthhe term wwithin an argument) is associated with one of five functional term types, denoted yd,q,s,a,n. [sent-99, score-0.344]
32 1 Priors Typical probabilistic topic models assume a symmetric Dirichlet prior over its term distributions or apply empirical Bayesian techniques to estimate the hyperparameters. [sent-107, score-0.249]
33 We encode this in ηb, the prior over the background term distribution, by setting each value to the logarithm of the term’s argument frequency. [sent-118, score-0.194]
34 Preliminary tests showed that final topics are relatively insensitive to the values of the hyperparameters. [sent-125, score-0.194]
35 3 T and K In all experiments, we use T = 40 topics and K = 2 positions. [sent-136, score-0.194]
36 4 Evaluation Recall that the aim of this work is to infer a lowdimensional representation of debate text. [sent-138, score-0.497]
37 We estimated our model on the Debatepedia debates (not including hyperlinked articles), and conducted several evaluations of the model, each considering a different aspect of the goal. [sent-139, score-0.341]
38 We exploit external articles hyperlinked from Debatepedia described in §2 as supporting dte fxrtosm mfor D arguments, treating dea icnh §o2ne a’ss association to an argument as variable to be predicted. [sent-140, score-0.415]
39 l8091 JS Divergence Figure 3: The distribution over Jensen-Shannon divergences between a hyperlinked article and the corresponding Debatepedia argument, n = 3, 352. [sent-149, score-0.213]
40 our model’s positional assignment of arguments to human annotated clusterings. [sent-150, score-0.201]
41 151 2O aur-r model can be used to infer the posterior over topics associated with such an article, and we compare that distribution to that of the Debatepedia article that links to it. [sent-156, score-0.308]
42 Calculating the similarity of these distributions, we get an estimate of how closely our model can associate text related to a debate with the specific argument that linked to it. [sent-157, score-0.6]
43 , 2003), which ignores sentiment, and the joint sentiment topic (JST) model (Lin and He, 2009), an unsupervised model that jointly captures sentiment and topic. [sent-159, score-0.301]
44 More challenging, of course, is selecting the argument to which an external article should be associated. [sent-164, score-0.209]
45 We used the Jensen-Shannon divergence between topic distributions of articles and arguments to rank the latter, for each article. [sent-165, score-0.383]
46 12JST multiplies topics out by the set of sentiment labels, assigning each token to both a topic and a sentment. [sent-167, score-0.392]
47 ∈It i5s likely possible to engineer more accurate models for attaching articles to arguments, but the attachment task is our aim only insofar as it contributes to an overall assessment of an inferred rep- resentation’s quality. [sent-178, score-0.242]
48 We next consider the JS divergences of position term distributions by topic; for each topic t, we consider the divergence between inferred values for φ1o,t and φ2o,t. [sent-182, score-0.527]
49 We tested our model’s ability to infer the positions of arguments. [sent-186, score-0.201]
50 The heldout arguments were selected so that every debate side maintained at least one argument whose inferred side could serve as the correct answer for the held-out argument. [sent-188, score-0.873]
51 We then inferred ifor each heldout argument from debate d and side s, given the parameters, and compared it with the value of id,s inferred during parameter estimation. [sent-189, score-0.786]
52 Note that JST does not provide a baseline for comparison, since it does not capture debate sides. [sent-191, score-0.46]
53 We can also use the model to predict the position adopted in an external text. [sent-194, score-0.176]
54 For articles linked from within Debatepedia, we have a gold standard: from which side of a debate was it linked? [sent-195, score-0.688]
55 After using the model to infer a position variable for such a text, we can check whether the inferred position variable matches that of the argument that links to it. [sent-196, score-0.508]
56 Note that our model is learned only from text within Debatepedia; it does not observe the text of external linked articles. [sent-206, score-0.158]
57 1863 Figure 5: Jensen-Shannon divergences between topicspecific positional term distributions, for each topic. [sent-208, score-0.24]
58 3 Comparison to Human Judgments of Positions We compared our model’s inferred positions to human judgments. [sent-212, score-0.251]
59 For each of the 11 topics in Table 8, we selected two associated debates with more arguments than average (24. [sent-213, score-0.592]
60 The debates were provided to each of three human who annotators,13 13All were native English-speaking American graduate students not otherwise involved in this research. [sent-215, score-0.253]
61 Each is known by the authors to have basic literacy with issues and debates in grJiSenocvSDe0 . [sent-216, score-0.299]
62 7 654blog(12)edit4wki(1)news(3)other(18)gov(12) Article type(% of articles) Figure 6: Position prediction on 500 hyperlinked articles by genre. [sent-218, score-0.171]
63 The labels and alignments between the two models’ topics were assigned manually. [sent-224, score-0.194]
64 ) Our model: topic-specific position bigrams which are ranked by comparing the log odds conditioned on the position and topic: log φio1 log φio2,t,w. [sent-226, score-0.263]
65 The instructions stated: Our goal is to see what you think about how the different sides of different debates can be lined up. [sent-237, score-0.385]
66 You might find it convenient to think of these in terms of political philosophies, contemporary political party platforms, or something else. [sent-238, score-0.198]
67 14 A1 used keyword lists to label items; we coarsened his labels manually by removing or merging less common keywords (resulting in: Republican, Democrat, science/environment, nanny, political reform, fiscal liberal, fiscal conservative, libertarian, Israel, Palestine, and one unlabeled side). [sent-241, score-0.179]
68 We used 100 samples from our Gibbs sampler to estimate posteriors for each id,s; these were always 99% or more in agreement, so we mapped each debate side into its single most probable cluster. [sent-247, score-0.522]
69 Recall that the two sides of each debate must be in different clusters. [sent-248, score-0.592]
70 The model agrees with A2’s coarse clustering most closely, and in fact is closer to A2’s clustering than A2 is to A3’s; it also agrees with A2’s coarse clustering better than A2’s coarse and fine clusterings agree (3. [sent-250, score-0.155]
71 Within debates and within topics, the model uses the position variable to distinguish sides well. [sent-254, score-0.565]
72 For external text, the model performs well on articles such as blogs and editorials but on others the positional categories do not seem meaning15This was determined using 1,000 samples. [sent-255, score-0.361]
73 Table 8: For 6 selected topics (labels assigned manually), top terms and person entities Bigrams were included but did not rank in the top five for these topics. [sent-257, score-0.194]
74 The model has conflated debates relating to same-sex marriage with the space program. [sent-258, score-0.301]
75 Noting the vast literature focusing on ideological positions expressed in text, we believe this failure suggests (i) that broadbased positions that hold across many topics may require richer textual representations (see, e. [sent-260, score-0.627]
76 Aside from those issues, a stronger theory of positions may be required. [sent-263, score-0.164]
77 Finally, exploiting explicitly ideological texts alongside the moderated arguments of Debatepedia might also help to identify textual associations with general positions (Sim et al. [sent-265, score-0.381]
78 2 Qualitative Analysis Of the T = 40 topics our model inferred, we subjectively judged 37 to be coherent; a glimpse of each is given in Figure 5. [sent-269, score-0.275]
79 We manually selected six of the most interpretable topics for further evaluation. [sent-270, score-0.194]
80 As a generative modeling approach, our model was designed for the purpose of reducing the dimensionality of the sociopolitical debate space, as evidenced by Debatepedia. [sent-271, score-0.613]
81 Table 6 compares the positional bigrams ofour model to the sentiments inferred by JST. [sent-273, score-0.237]
82 We observe the benefit of our model in identifying terms associated with positions on social issues, while JST selects more general sentiment terms. [sent-274, score-0.3]
83 Table 7 shows bigrams most strongly associated with general position distributions and selected topic-position distributions φo. [sent-275, score-0.295]
84 Although we have used frequent bigrams as a poor man’s approximation to multiword expression analysis, we find the topic-specific positions terms to be subjectively evocative. [sent-277, score-0.306]
85 The separation of personal name mentions into their own distributions, shown for some topics in Table 8, gives a distinctive characterization of topics based on relevant personalities. [sent-279, score-0.388]
86 Subjectively, the top individuals are relevant to the subject matter associated with each topic (though the topics are not always pure; same-sex marriage and the space program are merged, for example). [sent-280, score-0.37]
87 φi φi 5 Related Work Insofar as debates are subjective, our study is related to opinion mining. [sent-281, score-0.318]
88 Subjective text classification (Wiebe and Riloff, 2005) leads to opinion mining tasks such as opinion extraction (Dave et al. [sent-282, score-0.168]
89 The above studies are conducted mostly on product reviews, a domain with a simpler opinion landscape and more concrete rationales for those opinions, compared to sociopolitical debates. [sent-286, score-0.166]
90 Generative topic models have been successfully implemented in opinion mining tasks such as feature identification (Titov and McDonald, 2008), entitytopic extraction (Newman et al. [sent-287, score-0.198]
91 The JST model of Lin and He (2009) is an LDA-based topic model in which each word token is assigned both a sentiment and a topic; they exploited a sen16For more topics, please refer to the supplementary notes. [sent-294, score-0.198]
92 Unlike Lin and He’s sentiments, our model’s positions are associated with the two sides of a debate, and we incorporate topics at the level of questions within debates. [sent-297, score-0.615]
93 (201 1) used a debate corpus as a seed for extracting person-opinion-topic tuples from news and other web documents and in later work classified the quotations to specific topics and polarity using language models (Awadallah et al. [sent-309, score-0.654]
94 (201 1) were interested in ideological content in debates, relying on discourse structure and leveraging sentiment lexicons to recognize stances. [sent-312, score-0.208]
95 (2008) presented a statistical model for political discourse that incorporates both topics and ideologies; they used debates on the Israeli-Palestinian conflict. [sent-314, score-0.546]
96 , 2003); these were applied to debates and political blog datasets. [sent-319, score-0.387]
97 We go farther injointly modeling text across many debates evidenced by the structure of Debatepedia, thus grounding our models more solidly in familiar sociopolitical issues, and in making extensive use of existing NLP resources. [sent-321, score-0.354]
98 6 Conclusion Using text from Debatepedia, we inferred topics and position term lexicons in the domain of sociopolitical debates. [sent-322, score-0.553]
99 Our approach brings together tools from information extraction and sentiment analysis into a latent-variable topic model and exploits the hierarchical structure of the dataset. [sent-323, score-0.198]
100 Combining textual entailment and argumentation theory for supporting online debates interactions. [sent-365, score-0.321]
wordName wordTfidf (topN-words)
[('debate', 0.46), ('debatepedia', 0.423), ('debates', 0.253), ('topics', 0.194), ('jst', 0.181), ('positions', 0.164), ('sides', 0.132), ('arguments', 0.112), ('ideological', 0.105), ('sentiment', 0.103), ('position', 0.101), ('sociopolitical', 0.101), ('political', 0.099), ('topic', 0.095), ('argument', 0.09), ('positional', 0.089), ('hyperlinked', 0.088), ('inferred', 0.087), ('articles', 0.083), ('awadallah', 0.081), ('divergences', 0.081), ('editorials', 0.081), ('subjectively', 0.081), ('external', 0.075), ('term', 0.07), ('opinion', 0.065), ('side', 0.062), ('bigrams', 0.061), ('dasigi', 0.06), ('liberal', 0.06), ('questions', 0.059), ('ahmed', 0.054), ('ideologies', 0.053), ('generative', 0.052), ('distributions', 0.05), ('linked', 0.05), ('marriage', 0.048), ('opinionfinder', 0.048), ('issues', 0.046), ('variable', 0.046), ('stances', 0.045), ('article', 0.044), ('lin', 0.043), ('divergence', 0.043), ('conservative', 0.042), ('swapna', 0.042), ('xing', 0.042), ('yes', 0.042), ('coarse', 0.041), ('arguing', 0.04), ('confusi', 0.04), ('dbp', 0.04), ('fiscal', 0.04), ('insofar', 0.04), ('pradeep', 0.04), ('rawia', 0.04), ('bags', 0.04), ('somasundaran', 0.04), ('mining', 0.038), ('interpretability', 0.038), ('gibbs', 0.038), ('infer', 0.037), ('singapore', 0.036), ('wiebe', 0.036), ('blog', 0.035), ('maya', 0.035), ('ganapathibhotla', 0.035), ('boilerplate', 0.035), ('kohlsch', 0.035), ('tter', 0.035), ('argumentation', 0.035), ('fortuna', 0.035), ('poole', 0.035), ('subgroup', 0.035), ('yanchuan', 0.035), ('opinions', 0.034), ('mei', 0.034), ('prior', 0.034), ('blogs', 0.033), ('supporting', 0.033), ('switching', 0.033), ('blei', 0.033), ('within', 0.033), ('associated', 0.033), ('qualitative', 0.032), ('summarizing', 0.032), ('clusterings', 0.032), ('viewpoints', 0.032), ('cabrio', 0.032), ('attaching', 0.032), ('newman', 0.032), ('controversial', 0.032), ('dave', 0.032), ('gun', 0.032), ('lexi', 0.032), ('ramanath', 0.032), ('annotators', 0.032), ('proceedings', 0.031), ('contrastive', 0.031), ('subjective', 0.031)]
simIndex simValue paperId paperTitle
same-paper 1 1.000002 121 emnlp-2013-Learning Topics and Positions from Debatepedia
Author: Swapna Gottipati ; Minghui Qiu ; Yanchuan Sim ; Jing Jiang ; Noah A. Smith
Abstract: We explore Debatepedia, a communityauthored encyclopedia of sociopolitical debates, as evidence for inferring a lowdimensional, human-interpretable representation in the domain of issues and positions. We introduce a generative model positing latent topics and cross-cutting positions that gives special treatment to person mentions and opinion words. We evaluate the resulting representation’s usefulness in attaching opinionated documents to arguments and its consistency with human judgments about positions.
2 0.17094287 129 emnlp-2013-Measuring Ideological Proportions in Political Speeches
Author: Yanchuan Sim ; Brice D. L. Acree ; Justin H. Gross ; Noah A. Smith
Abstract: We seek to measure political candidates’ ideological positioning from their speeches. To accomplish this, we infer ideological cues from a corpus of political writings annotated with known ideologies. We then represent the speeches of U.S. Presidential candidates as sequences of cues and lags (filler distinguished only by its length in words). We apply a domain-informed Bayesian HMM to infer the proportions of ideologies each candidate uses in each campaign. The results are validated against a set of preregistered, domain expertauthored hypotheses.
3 0.14661793 77 emnlp-2013-Exploiting Domain Knowledge in Aspect Extraction
Author: Zhiyuan Chen ; Arjun Mukherjee ; Bing Liu ; Meichun Hsu ; Malu Castellanos ; Riddhiman Ghosh
Abstract: Aspect extraction is one of the key tasks in sentiment analysis. In recent years, statistical models have been used for the task. However, such models without any domain knowledge often produce aspects that are not interpretable in applications. To tackle the issue, some knowledge-based topic models have been proposed, which allow the user to input some prior domain knowledge to generate coherent aspects. However, existing knowledge-based topic models have several major shortcomings, e.g., little work has been done to incorporate the cannot-link type of knowledge or to automatically adjust the number of topics based on domain knowledge. This paper proposes a more advanced topic model, called MC-LDA (LDA with m-set and c-set), to address these problems, which is based on an Extended generalized Pólya urn (E-GPU) model (which is also proposed in this paper). Experiments on real-life product reviews from a variety of domains show that MCLDA outperforms the existing state-of-the-art models markedly.
4 0.12583579 143 emnlp-2013-Open Domain Targeted Sentiment
Author: Margaret Mitchell ; Jacqui Aguilar ; Theresa Wilson ; Benjamin Van Durme
Abstract: We propose a novel approach to sentiment analysis for a low resource setting. The intuition behind this work is that sentiment expressed towards an entity, targeted sentiment, may be viewed as a span of sentiment expressed across the entity. This representation allows us to model sentiment detection as a sequence tagging problem, jointly discovering people and organizations along with whether there is sentiment directed towards them. We compare performance in both Spanish and English on microblog data, using only a sentiment lexicon as an external resource. By leveraging linguisticallyinformed features within conditional random fields (CRFs) trained to minimize empirical risk, our best models in Spanish significantly outperform a strong baseline, and reach around 90% accuracy on the combined task of named entity recognition and sentiment prediction. Our models in English, trained on a much smaller dataset, are not yet statistically significant against their baselines.
5 0.11243428 47 emnlp-2013-Collective Opinion Target Extraction in Chinese Microblogs
Author: Xinjie Zhou ; Xiaojun Wan ; Jianguo Xiao
Abstract: Microblog messages pose severe challenges for current sentiment analysis techniques due to some inherent characteristics such as the length limit and informal writing style. In this paper, we study the problem of extracting opinion targets of Chinese microblog messages. Such fine-grained word-level task has not been well investigated in microblogs yet. We propose an unsupervised label propagation algorithm to address the problem. The opinion targets of all messages in a topic are collectively extracted based on the assumption that similar messages may focus on similar opinion targets. Topics in microblogs are identified by hashtags or using clustering algorithms. Experimental results on Chinese microblogs show the effectiveness of our framework and algorithms.
6 0.095091142 109 emnlp-2013-Is Twitter A Better Corpus for Measuring Sentiment Similarity?
7 0.093806639 100 emnlp-2013-Improvements to the Bayesian Topic N-Gram Models
8 0.092292264 16 emnlp-2013-A Unified Model for Topics, Events and Users on Twitter
9 0.084057108 133 emnlp-2013-Modeling Scientific Impact with Topical Influence Regression
10 0.081439123 81 emnlp-2013-Exploring Demographic Language Variations to Improve Multilingual Sentiment Analysis in Social Media
11 0.080545448 62 emnlp-2013-Detection of Product Comparisons - How Far Does an Out-of-the-Box Semantic Role Labeling System Take You?
12 0.075495355 148 emnlp-2013-Orthonormal Explicit Topic Analysis for Cross-Lingual Document Matching
13 0.073195517 147 emnlp-2013-Optimized Event Storyline Generation based on Mixture-Event-Aspect Model
14 0.071985051 99 emnlp-2013-Implicit Feature Detection via a Constrained Topic Model and SVM
16 0.069153026 63 emnlp-2013-Discourse Level Explanatory Relation Extraction from Product Reviews Using First-Order Logic
17 0.067266338 6 emnlp-2013-A Generative Joint, Additive, Sequential Model of Topics and Speech Acts in Patient-Doctor Communication
18 0.066381477 158 emnlp-2013-Recursive Deep Models for Semantic Compositionality Over a Sentiment Treebank
19 0.065915488 169 emnlp-2013-Semi-Supervised Representation Learning for Cross-Lingual Text Classification
20 0.065525472 120 emnlp-2013-Learning Latent Word Representations for Domain Adaptation using Supervised Word Clustering
topicId topicWeight
[(0, -0.208), (1, 0.103), (2, -0.138), (3, -0.052), (4, 0.077), (5, -0.05), (6, 0.008), (7, -0.001), (8, 0.052), (9, -0.014), (10, -0.05), (11, -0.125), (12, -0.083), (13, 0.061), (14, 0.015), (15, 0.141), (16, 0.144), (17, 0.043), (18, 0.044), (19, -0.065), (20, -0.011), (21, 0.075), (22, -0.096), (23, 0.076), (24, 0.07), (25, 0.004), (26, 0.019), (27, 0.046), (28, 0.069), (29, 0.074), (30, 0.024), (31, 0.004), (32, -0.12), (33, 0.085), (34, 0.031), (35, -0.065), (36, 0.026), (37, -0.053), (38, -0.025), (39, -0.001), (40, -0.132), (41, -0.037), (42, 0.046), (43, -0.152), (44, 0.064), (45, 0.159), (46, 0.088), (47, -0.064), (48, 0.067), (49, 0.093)]
simIndex simValue paperId paperTitle
same-paper 1 0.94807869 121 emnlp-2013-Learning Topics and Positions from Debatepedia
Author: Swapna Gottipati ; Minghui Qiu ; Yanchuan Sim ; Jing Jiang ; Noah A. Smith
Abstract: We explore Debatepedia, a communityauthored encyclopedia of sociopolitical debates, as evidence for inferring a lowdimensional, human-interpretable representation in the domain of issues and positions. We introduce a generative model positing latent topics and cross-cutting positions that gives special treatment to person mentions and opinion words. We evaluate the resulting representation’s usefulness in attaching opinionated documents to arguments and its consistency with human judgments about positions.
2 0.8070001 129 emnlp-2013-Measuring Ideological Proportions in Political Speeches
Author: Yanchuan Sim ; Brice D. L. Acree ; Justin H. Gross ; Noah A. Smith
Abstract: We seek to measure political candidates’ ideological positioning from their speeches. To accomplish this, we infer ideological cues from a corpus of political writings annotated with known ideologies. We then represent the speeches of U.S. Presidential candidates as sequences of cues and lags (filler distinguished only by its length in words). We apply a domain-informed Bayesian HMM to infer the proportions of ideologies each candidate uses in each campaign. The results are validated against a set of preregistered, domain expertauthored hypotheses.
3 0.643525 77 emnlp-2013-Exploiting Domain Knowledge in Aspect Extraction
Author: Zhiyuan Chen ; Arjun Mukherjee ; Bing Liu ; Meichun Hsu ; Malu Castellanos ; Riddhiman Ghosh
Abstract: Aspect extraction is one of the key tasks in sentiment analysis. In recent years, statistical models have been used for the task. However, such models without any domain knowledge often produce aspects that are not interpretable in applications. To tackle the issue, some knowledge-based topic models have been proposed, which allow the user to input some prior domain knowledge to generate coherent aspects. However, existing knowledge-based topic models have several major shortcomings, e.g., little work has been done to incorporate the cannot-link type of knowledge or to automatically adjust the number of topics based on domain knowledge. This paper proposes a more advanced topic model, called MC-LDA (LDA with m-set and c-set), to address these problems, which is based on an Extended generalized Pólya urn (E-GPU) model (which is also proposed in this paper). Experiments on real-life product reviews from a variety of domains show that MCLDA outperforms the existing state-of-the-art models markedly.
4 0.61019242 133 emnlp-2013-Modeling Scientific Impact with Topical Influence Regression
Author: James Foulds ; Padhraic Smyth
Abstract: When reviewing scientific literature, it would be useful to have automatic tools that identify the most influential scientific articles as well as how ideas propagate between articles. In this context, this paper introduces topical influence, a quantitative measure of the extent to which an article tends to spread its topics to the articles that cite it. Given the text of the articles and their citation graph, we show how to learn a probabilistic model to recover both the degree of topical influence of each article and the influence relationships between articles. Experimental results on corpora from two well-known computer science conferences are used to illustrate and validate the proposed approach.
5 0.59624803 100 emnlp-2013-Improvements to the Bayesian Topic N-Gram Models
Author: Hiroshi Noji ; Daichi Mochihashi ; Yusuke Miyao
Abstract: One of the language phenomena that n-gram language model fails to capture is the topic information of a given situation. We advance the previous study of the Bayesian topic language model by Wallach (2006) in two directions: one, investigating new priors to alleviate the sparseness problem caused by dividing all ngrams into exclusive topics, and two, developing a novel Gibbs sampler that enables moving multiple n-grams across different documents to another topic. Our blocked sampler can efficiently search for higher probability space even with higher order n-grams. In terms of modeling assumption, we found it is effective to assign a topic to only some parts of a document.
6 0.48707312 6 emnlp-2013-A Generative Joint, Additive, Sequential Model of Topics and Speech Acts in Patient-Doctor Communication
7 0.48472652 99 emnlp-2013-Implicit Feature Detection via a Constrained Topic Model and SVM
8 0.45143032 148 emnlp-2013-Orthonormal Explicit Topic Analysis for Cross-Lingual Document Matching
9 0.44949564 47 emnlp-2013-Collective Opinion Target Extraction in Chinese Microblogs
10 0.44862503 199 emnlp-2013-Using Topic Modeling to Improve Prediction of Neuroticism and Depression in College Students
11 0.39689243 144 emnlp-2013-Opinion Mining in Newspaper Articles by Entropy-Based Word Connections
12 0.39085594 35 emnlp-2013-Automatically Detecting and Attributing Indirect Quotations
13 0.38310346 26 emnlp-2013-Assembling the Kazakh Language Corpus
14 0.38117331 173 emnlp-2013-Simulating Early-Termination Search for Verbose Spoken Queries
15 0.37251571 143 emnlp-2013-Open Domain Targeted Sentiment
17 0.3467502 11 emnlp-2013-A Multimodal LDA Model integrating Textual, Cognitive and Visual Modalities
18 0.34656537 161 emnlp-2013-Rule-Based Information Extraction is Dead! Long Live Rule-Based Information Extraction Systems!
19 0.32946292 170 emnlp-2013-Sentiment Analysis: How to Derive Prior Polarities from SentiWordNet
20 0.32907176 63 emnlp-2013-Discourse Level Explanatory Relation Extraction from Product Reviews Using First-Order Logic
topicId topicWeight
[(3, 0.041), (9, 0.015), (18, 0.038), (22, 0.039), (30, 0.059), (31, 0.288), (47, 0.013), (50, 0.012), (51, 0.184), (53, 0.029), (66, 0.052), (71, 0.043), (75, 0.043), (77, 0.015), (90, 0.013), (96, 0.037)]
simIndex simValue paperId paperTitle
1 0.81845802 149 emnlp-2013-Overcoming the Lack of Parallel Data in Sentence Compression
Author: Katja Filippova ; Yasemin Altun
Abstract: A major challenge in supervised sentence compression is making use of rich feature representations because of very scarce parallel data. We address this problem and present a method to automatically build a compression corpus with hundreds of thousands of instances on which deletion-based algorithms can be trained. In our corpus, the syntactic trees of the compressions are subtrees of their uncompressed counterparts, and hence supervised systems which require a structural alignment between the input and output can be successfully trained. We also extend an existing unsupervised compression method with a learning module. The new system uses structured prediction to learn from lexical, syntactic and other features. An evaluation with human raters shows that the presented data harvesting method indeed produces a parallel corpus of high quality. Also, the supervised system trained on this corpus gets high scores both from human raters and in an automatic evaluation setting, significantly outperforming a strong baseline. 1 Introduction and related work Sentence compression is a paraphrasing task where the goal is to generate sentences shorter than given while preserving the essential content. A robust compression system would be useful for mobile devices as well as a module in an extractive summarization system (Mani, 2001). Although a compression may differ lexically and structurally from the source sentence, to date most systems are extractive and proceed by deleting words from the 1481 input (Knight & Marcu, 2000; Dorr et al., 2003; Turner & Charniak, 2005; Clarke & Lapata, 2008; Berg-Kirkpatrick et al., 2011, inter alia). To decide which words, dependencies or phrases can be dropped, (i) rule-based approaches (Grefenstette, 1998; Jing & McKeown, 2000; Dorr et al., 2003; Zajic et al., 2007), (ii) supervised models trained on parallel data (Knight & Marcu, 2000; Turner & Charniak, 2005; McDonald, 2006; Gillick & Favre, 2009; Galanis & Androutsopoulos, 2010, inter alia) and (iii) unsupervised methods which make use of statistics collected from non-parallel data (Hori & Furui, 2004; Zajic et al., 2007; Clarke & Lapata, 2008; Filippova & Strube, 2008) have been investigated. Since it is infeasible to manually devise a set of accurate deletion rules with high coverage, recent research has been devoted to developing statistical methods and possibly augmenting them with a few linguistic rules to improve output readability (Clarke & Lapata, 2008; Nomoto, 2009). Supervised models. A major problem for supervised deletion-based systems is very limited amount of parallel data. Many approaches make use of a small portion of the Ziff-Davis corpus which has about 1K sentence-compression pairs1 . Other main sources of training data are the two manually crafted compression corpora from the University of Edinburgh (“written” and “spoken”, each approx. 1.4K pairs). Galanis & Androutsopoulos (201 1) attempt at getting more parallel data by applying a deletionbased compressor together with an automatic para1The method of Galley & McKeown (2007) could benefit from a larger number of sentences. Proce Sdeiantgtlse o,f W thaesh 2i0n1gt3o nC,o UnSfeAre,n 1c8e- o2n1 E Omctpoibriecra 2l0 M13et.h ?oc d2s0 i1n3 N Aastusorcaila Ltiaon g fuoarg Ceo Pmrpoucetastsi on ga,l p Laignegsu 1is4t8ic1s–1491, phraser and generating multiple alternative compressions. To our knowledge, this extended data set has not yet been used for successful training of compression systems. Scarce parallel data makes it hard to go beyond a small set of features and explore lexicalization. For example, Knight & Marcu (2000) only induce nonlexicalized CFG rules, many of which occurred only once in the training data. The features of McDonald (2006) are formulated exclusively in terms of syntactic categories. Berg-Kirkpatrick et al. (201 1) have as few as 13 features to decide whether a constituent can be dropped. Galanis & Androutsopoulos (2010) use many features when deciding which branches of the input dependency tree can be pruned but require a reranker to select most fluent compressions from a pool of candidates generated in the pruning phase, many of which are ungrammatical. Even further data limitations exist for the algorithms which operate on syntactic trees and reformulate the compression task as a tree pruning one (Nomoto, 2008; Filippova & Strube, 2008; Cohn & Lapata, 2009; Galanis & Androutsopoulos, 2010, inter alia). These methods are sensitive to alignment errors, their performance degrades if the syntactic structure of the compression is very different from that of the input. For example, see Nomoto’s 2009 analysis of the poor performance of the T3 system of Cohn & Lapata (2009) when retrained on a corpus of loosely similar RSS feeds and news. Unsupervised models. Few approaches require no training data at all. The model of Hori & Furui (2004) combines scores estimated from monolingual corpora to generate compressions of transcribed speech. Adopting an integer linear programming (ILP) framework, Clarke & Lapata (2008) use hand-crafted syntactic constraints and an ngram language model, trained on uncompressed sentences, to find best compressions. The model of Filippova & Strube (2008) also uses ILP but the problem is formulated over dependencies and not ngrams. Conditional probabilities and word counts collected from a large treebank are combined in an ad hoc manner to assess grammatical importance and informativeness of dependencies. Similarly, Woodsend & Lapata (2010) formulate an ILP problem to gener- ate news story highlights using precomputed scores. 1482 Again, an ad hoc combination of the scores learned independently of the task is used in the objective function. Contributions of this paper. Our work is motivated by the obvious need for a large parallel corpus of sentences and compressions on which extractive systems can be trained. Furthermore, we want the compressions in the corpus to be structurally very close to the input. Ideally, in every pair, the compression should correspond to a subtree of the input. To this end, our contributions are three-fold: • • We describe an automatic procedure of constructing a parallel corpus ocf p 250,000 ese onften cocen-compression pairs such that the dependency tree of the compression is a subtree of the source tree. An evaluation with human raters demonstrates high quality of the parallel data in terms of readability and informativeness. We successfully apply the acquired data to train a neo svueclc supervised compression system ow thraicinh produces readable and informative compressions without employing a separate reranker. In particular, we start with the unsupervised method of Filippova & Strube (2008) and replace the ad hoc edge weighting with a linear function over a rich feature representation. The parameter vector is learned from our corpus specifically for the compression task using structured prediction (Collins, 2002). The new system significantly outperforms the baseline and hence provides further evidence for the utility of the parallel data. • We demonstrate that sparse lexical features are very eusmeofunls tfroart ese tnhtaetn spcea compression, aunreds st aharet a large parallel corpus is a requirement for applying them successfully. The compression framework we adopt and the unsupervised baseline are introduced in Section 2, the training algorithm for learning edge weights from parallel data is described in Section 3. In Section 4 we explain how to obtain the data and present an evaluation of its quality. In Section 5 we compare the baseline with our system and report the results of an experiment with humans as well as the results of an automatic evaluation. 2 Framework and baseline We adopt the unsupervised compression framework of Filippova & Strube (2008) as our baseline and extend it to a supervised structured prediction problem. In the experiments reported by Filippova & Strube (2008), the system was evaluated on the Edinburgh corpora. It achieved an F-score (Riezler et al., 2003) higher than reported by other systems on the same data under an aggressive compression rate and thus presents a competitive baseline. Tree pruning as optimization. In this framework, compressions are obtained by deleting edges of the source dependency structure so that (1) the retained edges form a valid syntactic tree, and (2) their total edge weight is maximized. The objective function is defined over set X = {xe, e ∈ E} of binary variables, corresponding {tox t,hee s∈et EE} o off t bhiesource edges, subject to the structural and length constraints, ×× = f(X) Xxe w(e) (1) Xe∈E Here, w(e) denotes the weight of edge e. This constrained optimization problem is solved under the tree structure and length constraints using ILP. If xe is resolved to 1, the respective edge is retained, otherwise it is deleted. The tree structure constraints enforce at most one parent for every node and structure connectivity (i.e., no disconnected subtrees). Given that length(node(e)) denotes the length of the node to which edge e points and α is the maximum permitted length for the compression, the length constraint is simply Xxe Xe∈E length(node(e)) ≤ α (2) Word limit is used in the original paper, whereas we use character length which is more appropriate for system comparisons (Napoles et al., 2011). If uniform weights are used in Eq. (1), the optimal solution would correspond to a subtree covering as many edges as possible while keeping the compression length under given limit. The solution to the surface realization problem (Belz et al., 2011) is standard: the words in the compression subtree are put in the same order they are found in the source. 1483 Due to space limitations, we refer the reader to (Filippova & Strube, 2008) for a detailed description on the method. Essential for the present discussion is that source dependency trees are transformed to dependency graphs in that (1) auxiliary, determiner, preposition, negation and possessive nodes are collapsed with their heads; (2) prepositions replace labels on the edges to their arguments; (3) the dummy root node is connected with every inflected verb. Figures 1(a)-1(b) illustrate most of the transformations. The transformations are deterministic and reversible, they can be implemented in a single top-down tree traversal2. The set E of edges in Eq. (1) is thus the set of edges of the transformed dependency graph, like in Fig. 1(b). A benefit of the transformations is that function words and negation appear in the compression if and only if their head words are present. Hence no separate constraints are required to en- × sure that negation or a determiner is preserved. The dummy root node makes constraint formulation easier and also allows for the generation of compressions from any finite clause of the source. The described pruning optimization framework is used both for the unsupervised baseline and for our supervised system. The difference between the baseline and our system is in how edge weights, w(e)’s in Eq. (1), are instantiated. Baseline edge weights. The precomputed edge weights reflect syntactic importance as well as informativeness of the nodes they point to. Given edge e from head node h to node n, the edge weight is the product of the syntactic and the informativeness weights, w(e) = wsynt(e) winfo(e) (3) The syntactic weight is defined as wsynt(e) = P(label(e) |lemma(h)) (4) For example, verb kill may have multiple arguments realized with dependency labels subj, dobj, in, etc. However, these argument labels are not equally likely, e.g., P(subj|kill) > P(in|kill) . When forced to prune an edge, t|hkiel system iwno|kuillld) prefer to keep 2Some of the transformations are comparable to what is implemented in the Stanford parser (de Marneffe et al., 2006). pspos prepnspoub j ro tdetamodc ompnsuabujxpas prep doebtjamodprep oabmjod ro tBritan’sMinsrto ryotfsoufbDjef nrsoeotsaysBritsha(bm)ocadcTosrmoaplndsifeor smubwejdagsrkapihledroadisnidaemoidnablast outheinr amiondAfghanistan ro tBritshamodrao stoldiersubjwas kiledinin a blastin Afghanistan Britain ’s Ministry of Defense says a British soldier was killed in a roadside blast in southern Afghanistan (a) Source dependency tree (c) Tree of extracted headline A British soldier was killed in a blast in Afghanistan detamodsubajuxpassrootpreppodbejtpreppobj A British soldier was killed in a blast in Afghanistan (d) Tree of extracted headline with transformations undone Figure 1: Source, transformed and extracted trees given headline British soldier killed in Afghanistan the subject edge over the preposition-in edge since it contributes more weight to the objective function. The informativeness score is inspired by Wood- send & Lapata (2012) and is defined as winfo(e) =PPhaeartdicl iene((lleemmmmaa((nn)))) (5) This weight tells us how likely it is that a word from an article appears in the headline. For exam- ple, given two edges one of which points to verb say and another one to verb kill, the latter would be preferred over the former because kill is more “headliny” than say. When collecting counts for the syntactic and informativeness scores, we used 9M news articles crawled from the Internet, much more than Filippova & Strube (2008). As a result our estimates are probably more accurate than theirs. Although both wsynt and winfo have a meaningful interpretation, there is no guarantee that product is the best way to combine the two when assigning edge weights. Also, it is unclear how to integrate other signals, such as distance to the root, node length or information about the siblings, which pre1484 sumably all play a role in determining the overall edge importance. 3 Learning edge weights Our supervised system differs from the unsupervised baseline in that instead of relying on precomputed scores, we define edge weight w(e) in Eq. (1) with a linear function over a feature representation, w(e) = w · f(e) (6) Here f(e) is a vector of binary variables for every feature from the set of all possible but very infrequent features in the training set. f(e) has 1for every feature extracted for edge e and zero otherwise. Table 1 gives an overview of the feature types we use (edge e points from head h to node n). Note that syntactic, structural and semantic features are closed-class. For all the structural features but char length, seven is used as maximum possible value; all possible character lengths are bucketed into six classes. All the features are local for a given edge, contextual information is included about – syntacticlabel(e); for e* to h, label(e*); pos(h); pos(n) structural depth(n); #children(n); #children(h); char length(n); #words in(n) semantic NE tag(h); NE tag(n); is negated(n) lexical lemma(n); lemma(h)-label(e); for e* to n’s siblings, lemma(h)-label(e*) Table 1: Types of features extracted for edge e from h to n the head and the target nodes, and the siblings as well as the children of the latter. The negation feature is only applicable to verb nodes which contain a negative particle, like not, after the tree transformations. Lexical features which combine lemmas and syntactic labels are inspired by the unsupervised baseline and are very sparse. In what follows, our assumption is that we have a compression corpus at our disposal where for every input sentence there is a correct “oracle” compression such that its transformed parse tree matches a subtree of the transformed input graph. Given such a corpus, we can apply structured prediction methods to learn the parameter vector w. In our study we employ an averaged variant of online structured perceptron (Collins, 2002). In the context of sentence fusion, a similar dependency structure pruning framework and a similar learning approach was adopted by Elsner & Santhanam (201 1). At every iteration, for every input graph, we find the optimal solution with ILP under the current parameter vector w. The maximum permitted compression length is set to be the same as the length of the oracle compression. Since the oracle compression is a subtree of the input graph, it represents a feasible solution for ILP. The parameter vector is updated if there is a mismatch between the predicted and the oracle sets of edges for all the features with a non-zero net count. More formally, given an input graph with the set of edges E, oracle compression C ⊂ E and compression Ct ⊆ E predicted at itera- tCion ⊂ ⊂t , t ahen parameter update ⊆vec Etor p raetd ti + 1d aist given by wt+1 = wt+ e∈XC\Ct f(e) −X f(e) X e∈XCt\C (7) w is averaged over all the wt’s so that features whose weight fluctuated a lot during training are penalized (Freund & Shapire, 1999). 1485 Of course, training a model with a large number of features, such as a lexicalized model, is only possible if there is a large compression corpus where the dependency tree of the compression is a subtree of the source sentence. In the next section we introduce our method of getting a sufficient amount of such data. 4 Acquiring parallel data automatically In this section we explain how we obtained a parallel corpus of sentences and compressions. The underlying idea is to harvest news articles from the Internet where the headline appears to be similar to the first sentence and use it to find an extractive compression of the sentence. Collecting headline-sentence pairs. Using a news crawler, we collected a corpus of news articles in English from the Internet. Similarly to previous work (Dolan et al., 2004; Wubben et al., 2009; Bejan & Harabagiu, 2010, inter alia), the Google News service3 was used to identify news. From every article, the headline and the first sentence, which are known to be semantically similar (Dorr et al., 2003), were extracted. Predictably, very few headlines are extractive compressions of the first sentence, therefore simply looking for pairs where the headline is a subsequence of the words from the first sentence would not solve the problem of getting a large amount of parallel data. Importantly, headlines are syntactically quite different from “normal” sentences. For example, they may have no main verb, omit determiners and appear incomplete, making it hard for a supervised deletion-based system to learn useful rules. Moreover, we observed poor parsing accuracy for headlines which would make syntactic annotations for headlines hardly useful. Thus, instead oftaking the headline as it is, we use it to find a proper extractive compression of the sen3http : / /news .google . com, Jan-Dec 2012. tence by matching lemmas of content words (nouns, verbs, adjectives, adverbs) and coreference IDs of entities from the headline with those of the sentence. The exact procedure is as follows (H, S and T stand for headline, sentence and transformed graph of the sentence): PREPROCESSING H and S are preprocessed in a standard way: tokenized, lemmatized, PoS and NE tagged. Additionally, S is parsed with a dependency parser (Nivre, 2006) and transformed as described in Section 2 to obtain T. Finally, pronominal anaphora is resolved in S. Recall that S is the first sentence, so the antecedent must be located in a preceding, higher-level clause. FILTERING To restrict the corpus to grammatical and informative headlines, we implemented a cascade of filters. Pair (H, S) is discarded if any of the questions in Table 2 is answered positively. Is H a question? Is H or S too short? (less than four word tokens) Is H about as long as S? (min ratio: 1.5) Does H lack a verb? Does H begin with a verb? Is there a noun, verb, adj, adv lemma from H not found in S? Are the noun, verb, adj, adv lemmas from H found in S in a different order? Table 2: Filters applied to candidate pair (H , S) MATCHING Given the content words of H, a subset of nodes in T is selected based on lemma or coreference identity of the main (head) word in the nodes. For example, the main word of a collapsed node in T, which covers two words was killed, is killed; was is its child attached with label aux in the untransformed parse tree. This node is marked if H contains word killed or killing because of the lemma identity. In some cases there are multiple possible matches. For example, given S Barack Obama said he will attend G20 and H mentioning Obama, both Barack Obama and he nodes are marked in T. Once all the nodes in T which match content words and entities from H are identified, a minimum subtree covering these nodes is found such that every word or entity from H occurs as many times in T as in 1486 H. So if H mentions Obama only once, then either Barack Obama or he must be covered by the subtree but not both. This minimum subtree corresponds to an extractive headline, H*, which we generate by ordering the surface forms of all the words in the subtree nodes by their offsets in S. Finally, the character length of H* is compared with the length of H. If H* is much longer than H, the pair (H S) is discarded (max ratio 1.5). As an illustration to the procedure, consider the example from Figure 1 with the extracted headline and its tree presented in Figure 1(c). Given the headline British soldier killed in Afghanistan, the extracted headline would be A British soldier was killed in a blast in Afghanistan. The lemmas british, soldier, kill, afghanistan from the headline match the nodes British, a soldier, was killed, in Afghanistan in the transformed graph. The node in a blast is added because it is on the path from was killed to in Afghanistan. Of course, it is possible to determinis- , tically undo the transformations in order to obtain a standard dependency tree. In this case the extracted headline would still correspond to a subtree of the input (compare Fig. 1(d) with Fig. 1(a)). Also note that a similar procedure can be implemented for constituency parses. The resulting corpus consists of 250K tuples (S T H H*), Appendix provides more examples of source sentences, original headlines and extracted headlines. We did not attempt to tune the values for minimum/maximum length and ratio lower thresholds may have produced comparable results. , , , – Evaluating data quality. The described procedure produces a comparatively large compression corpus but how good are automatically constructed compressions? To answer this question, we randomly selected 50 tuples from the corpus and set up an experiment with human raters to validate and assess data quality in terms of readability4 and informativeness5 which are standard measures of compression quality (Clarke & Lapata, 2006). Raters were asked to read a sentence and a compression (original H or extracted H* headline) and then rate the compression on two five-point scales. Three rat- ings were collected for every item. Table 3 gives 4Also called grammaticality and fluency. 5Also called importance and representativeness. average ratings with standard deviation. AVG read AVG info ORIG. HEADLINE4.36 (0.75)3.86 (0.79) EXTR. HEADLINE 4.26 (1.01) 3.70 (1.04) Table 3: Results for two kinds of headlines In terms of readability and informativeness the extracted headlines are comparable with humanwritten ones: at 95% confidence there is no statistically significant difference between the two. Encouraged by the results of the validation experiment we proceeded to our next question: Can a supervised compression system be successfully trained on this corpus? 5 System evaluation and discussion From the corpus of 250K tuples we used 100K to get pairs of extracted headlines and sentences for training (on the development set we did not observe much improvement from using more training data), 250 for development and the rest for testing. We ran the learning algorithm for 20 iterations, checking the performance on the development set. Features which applied to less than 20 edges were pruned, the size of the feature set is about 28K. 5.1 Evaluation with humans 50 pairs of original headlines and sentences (different from the data validation set in Sec. 4) were randomly selected for an evaluation with humans from the test data. As in the data quality validation experiment, we asked raters to assess the readability and informativeness of proposed compressions for the unsupervised system, our system and humanwritten headlines. The latter provide us with upper bounds on the evaluation criteria. Three ratings per item per parameter were collected. To get comparable results, the unsupervised and our systems used the same compression rate: for both, the requested maximum length was set to the length of the headline. Table 4 summarizes the results. The results indicate that the trained model significantly outperforms the unsupervised system, getting particularly good marks for readability. The differ- ence in readability between our system and original headlines is not statistically significant. Note that 1487 O U RNISRGU. SPY H.S ETYAESMDTLEIMNE4A43V. 376G0 6† read32A4.V. 571G20 † i‡nfo Table 4: Results for the systems and original headline: † and ‡ stand for significantly better than Unsupervised and Our system at 95% confidence, respectively the unsupervised baseline is also capable of generating readable compressions but does a much poorer job in selecting most important information. Our trained model successfully learned to optimize both scores. We refer the reader to Appendix for input and compression examples. Note that the ratings for the human-written headlines in this experiment are slightly different from the ratings in the data validation experiment because a different data sample was used. 5.2 Automatic evaluation Our automatic evaluation had the goal of explicitly addressing two relevant questions related to our claims about (1) the benefits of having a large parallel corpus and (2) employing a supervised approach with a rich feature representation. 1. Our primary motivation for collecting parallel data has been that having access to sparse lexical features, which considerably increase the feature space, would benefit compression systems. But is it really the case for sentence compression? Can a comparable performance be achieved with a closed, moderately sized set of dense, non-lexical features? If yes, then a large compression corpus is probably not needed. Furthermore, to demonstrate that a large corpus is not only sufficient but also necessary to learn weights for thousands of features, we need to compare the performance of the system when trained on the full data set and a small portion of it. 2. The syntactic and informativeness scores in Eq. (3) were calculated over millions of news articles and do provide us with meaninful statistics (see Sec. 2). Is there any benefit in replacing those scores with weights learned for their feature counterparts? Recall that one of our feature types in Table 1 is the concatenation of lemma(h) (parent lemma) and label(e) which relies on the same information as wsynt = P(label(e) |lemma(h)). The feature counterpart bofe winfo dmemfinae(dh i))n. Eq. (5) aislemma(n)–the lemma of the node to which edge points. How would the supervised system perform against the unsupervised one, if it only extracted features of these two types? To answer these questions, we sampled 1,000 tuples from the unused test data and measured F1 score (Riezler et al., 2003) by comparing the trees of the generated compression and the “correct”, extracted headline. The systems we compared are the unsupervised baseline (UNSUP. SYSTEM) and the supervised model trained on three kinds of feature sets: (1) SYNT-INFO FEATURES, corresponding to the supervised training of the unsupervised baseline model (i.e., lemma(h)-label(e) and lemma(n)); (2) NON-LEX FEATURES, corresponding to a dense, non-lexical feature representation (i.e., all the feature types from Table 1 excluding the three involving lemmas); (3) ALL FEATURES (same as OUR SYSTEM). Additionally, we trained the system on 10% of the data–10K as opposed to 100K tuples, ALL FEATURES ( 10K)–for 20 iterations ignoring features which applied to less than three edges6. As before, the same compression rate was used for all the systems. The results are summarized in Table 5. SANUYL ONSLTUF-LEPI.NASXTFYUOFSRE TAE STAMU(1R0ESK)F1578249s1c.0o364re#f12a7tN,u4835r.1A29e30s. Table 5: Results for the unsupervised baseline and the supervised system trained on three kinds of feature sets Clearly, having more features, lexicalized and unlexicalized, is important: there is a significant im6Recall from the beginning of the section that for the full (100K) training set the threshold was set to 20 with no tuning. For the 10K training set, we tried values of two, three, five and varied the number of iterations. The result we report is the highest we could get for 10K. 1488 provement in going beyond the closed set of 330 non-lexical features to all, from 79.6 to 84.3 points. Moreover, successful training requires a large corpus since the performance of the system degrades if only 10K training instances are used. Note that this number already exceeds all the existing compression corpora taken together. Hence, sparse lexical features are useful for compression and a large parallel corpus is a requirement for successful supervised training. Concerning our second question, learning feature weights from the data produces significantly better results than the hand-crafted way of making use of the same information, even if a much larger data set is used to collect statistics. We observed a dramatic increase from 52.3 to 75.0 points. Thus, we may conclude that training with dense and sparse features directly from data definitely improves the performance of the dependency pruning system. 5.3 Discussion It is important to note that the data we used is challenging: first sentences in news articles tend to be long, in fact longer than other news sentences, which implies less reliable syntactic analysis and noisier input to the syntax-based systems. In the test set we used for the evaluation with humans, the mean sentence length is 165 characters. The average compression rate in characters is 0.46 0. 16 which is quite iaogng rraetsesiv ine7 c. hRareaccatlel rtsha ist we u6se ±d 0th.1e6 very same framework for the unsupervised baseline and our system as well as the same compression rate. All the preprocessing errors affect both systems equally and the comparison of the two is fair. Predictably, wrong syntactic parses significantly increase chances of an ungrammatical compression, and parser errors seem to be a major source of readability deficiencies. A property of the described compression framework is that a desired compression length is expected to be provided by the user. This can be seen both as a strength and as a weakness, depending on the application. In a scenario where mobile devices with a limited screen size are used, or in a summarization scenario where a total summary length is ± provided (see the DUC/TAC guidelines8), being able 7We follow the standard terminology where smaller values imply shorter compressions. 8http : / /www .nist .gov/t ac / to specify a length is definitely an advantage. However, one can also think of other applications where the user does not have a strict length constraint but wants the text to be somewhat shorter. In this case, a reranker which compares compressions generated for a range of possible lengths can be employed to find a single compression (e.g., mean edge weight in the solution or a language model-based score). 6 Conclusions We have addressed a major problem for supervised extractive compression models the lack of a large parallel corpus. To this end, we presented a method to automatically build such a corpus from web documents available on the Internet. An evaluation with humans demonstrates that the quality of the corpus is high the compressions are grammatical and informative. We also significantly improved a competitive unsupervised method achieving high readability and informativeness scores by incorpo– – rating thousands of features and learning the feature weights from our corpus. This result further confirms the practical utility of the automatically obtained data. We have shown that employing lexical features is important for sentence compression, and that our supervised module can successfully learn their weights from the corpus. To our knowledge, we are the first to empirically demonstrate that sparse features are useful for compression and that a large parallel corpus is a requirement for a successful learning of their weights. We believe that other supervised deletion-based systems can benefit from our work. Acknowledgements: The authors are thankful to the EMNLP reviewers for their feedback and suggestions. Appendix The appendix presents examples of source sentences (S), original headlines (H), extracted headlines (H*), unsupervised baseline (U) and our system (O) compressions. 1489 References Bejan, C. & S. Harabagiu (2010). Unsupervised event coreference resolution with rich linguistic features. In Proc. of ACL-10, pp. 1412–1422. Belz, A., M. White, D. Espinosa, E. Kow, D. Hogan & A. Stent (201 1). The first surface realization shared task: Overview and evaluation results. In Proc. of ENLG-11, pp. 217–226. Berg-Kirkpatrick, T., D. Gillick & D. Klein (201 1). Jointly learning to extract and compress. In Proc. of ACL-11. Clarke, J. & M. Lapata (2006). Models for sentence compression: A comparison across domains, training requirements and evaluation measures. In Proc. of COLING-ACL-06, pp. 377–385. Clarke, J. & M. Lapata (2008). Global inference for sentence compression: An integer linear programming approach. Journal of Artificial Intelligence Research, 3 1:399–429. Cohn, T. & M. Lapata (2009). Sentence compres- sion as tree transduction. Journal of Artificial Intelligence Research, 34:637–674. Collins, M. (2002). Discriminative training methods for Hidden Markov Models: Theory and experiments with perceptron algorithms. In Proc. of EMNLP-02, pp. 1–8. de Marneffe, M.-C., B. MacCartney & C. D. Manning (2006). Generating typed dependency parses from phrase structure parses. In Proc. of LREC06, pp. 449–454. Dolan, B., C. Quirk & C. Brokett (2004). Unsupervised construction oflarge paraphrase corpora: Exploiting massively parallel news sources. In Proceedings of the 20th International Conference on Computational Linguistics, Geneva, Switzerland, 23–27 August 2004, pp. 350–356. Dorr, B., D. Zajic & R. Schwartz (2003). Hedge trimmer: A parse-and-trim approach to headline generation. In Proceedings of the Text Summarization Workshop at HLT-NAACL-03, Edmonton, Alberta, Canada, 2003, pp. 1–8. SCountry star Sara Evans has married former University of Alabama quarterback Jay Barker. H H* U O Country star Sara Evans marries Country star Sara Evans has married Sara Evans has married Jay Barker Sara Evans has married Jay Barker H H* U O Intel to build car batteries Intel would be building car batteries would be building the company said Intel would be building car batteries SIntel would be building car batteries, expanding its business beyond its core strength, the company said in a statement SA New Orleans Saints team spokesman says tight end Jeremy Shockey was taken to a hospital but is doing fine. H H* U O Spokesman: Shockey taken to hospital, doing fine spokesman says Jeremy Shockey was taken to a hospital but is doing fine A New Orleans Saints team spokesman says Jeremy Shockey was taken tight end Jeremy Shockey was taken to a hospital but is doing fine SPresident Obama declared a major disaster exists in the State of Florida and ordered Federal aid to supplement H H* U O State and beginning President President President President local recovery efforts in the area struck by severe storms, flooding, tornadoes, and straight-line winds on May 17, 2009, and continuing. Obama declares major disaster exists in the State of Florida Obama declared a major disaster exists in the State of Florida Obama declared a major disaster exists and ordered Federal aid Obama declared a major disaster exists in the State of Florida H H* U O mounting loan defaults. Regulators shut down small Florida bank Regulators shut down a small Florida bank shut down bringing the number of failures Regulators shut down a small Florida bank SRegulators Friday shut down a small Florida bank, bringing to 119 the number of US bank failures this year amid SThree men were arrested Wednesday night and Dayton police said their arrests are in connection to a west Dayton H H* U O bank robbery. 3 men arrested in connection with Bank robbery Three men were arrested are in connection to a bank robbery were arrested and Dayton police said their arrests are Three men were arrested and police said their arrests are SThe government and the social partners will resume the talks on the introduction of the so-called crisis tax, H H* U O which will be levied on all salaries, pensions and incomes over HRK 3,000. Government, social partners to resume talks on introduction of “crisis” tax. The government and the social partners will resume the talks on the introduction of the crisis tax The government will resume the talks on the introduction of the crisis tax which will be levied The government and the social partners will resume the talks on the introduction of the crisis tax SEngland star David Beckham may have the chance to return to AC Milan after the Italian club’s coach said H H* U O he was open to his move on Sunday. Beckham has chance of returning to Milan David Beckham may have the chance to return to AC Milan David Beckham may have the chance to return said star was David Beckham may have the chance to return to AC Milan SEastern Health and its insurance company have accepted liability for some patients involved in the breast cancer H H* U O testing scandal, according to a statement released Friday afternoon. Eastern Health accepts liability for some patients Eastern Health have accepted liability for some patients Health have accepted liability according to a statement Eastern Health have accepted liability for some patients SFrontier Communications Corp., a provider of phone, TV and Internet services, said Thursday H H* U it has started a cash tender offer to purchase up to $700 million of its notes. Frontier Communications starts tender offer for up to $700 million of notes Frontier Communications has started a tender offer to purchase $700 million of its notes Frontier Communications said Thursday a provider has started a tender offer OFrontier Communications has started a tender offe1r4 to9 p0urchase $700 million of its notes Elsner, M. & D. Santhanam (201 1). Learning to fuse disparate sentences. In Proceedings of the Workshop on Monolingual Text-to-text Generation, Prtland, OR, June 24 2011, pp. 54–63. Filippova, K. & M. Strube (2008). Dependency tree based sentence compression. In Proc. of INLG08, pp. 25–32. Freund, Y. & R. E. Shapire (1999). Large margin classification using the perceptron algorithm. Machine Learning, 37:277–296. Galanis, D. & I. Androutsopoulos (2010). An extractive supervised two-stage method for sentence compression. In Proc. of NAACL-HLT-10, pp. 885–893. Galanis, D. & I. Androutsopoulos (201 1). A new sentence compression dataset and its use in an abstractive generate-and-rank sentence compressor. In Proc. of UCNLG+Eval-11, pp. 1–1 1. Galley, M. & K. R. McKeown (2007). Lexicalized Markov grammars for sentence compression. In Proc. of NAACL-HLT-07, pp. 180–187. Gillick, D. & B. Favre (2009). A scalable global model for summarization. In ILP for NLP-09, pp. 10–18. Grefenstette, G. (1998). Producing intelligent telegraphic text reduction to provide an audio scanning service for the blind. In Working Notes of the Workshop on Intelligent Text Summarization, Palo Alto, Cal., 23 March 1998, pp. 111–1 17. Hori, C. & S. Furui (2004). Speech summarization: An approach through word extraction and a method for evaluation. IEEE Transactions on Information and Systems, E87-D(1): 15–25. Jing, H. & K. McKeown (2000). Cut and paste based text summarization. In Proc. of NAACL-00, pp. 178–185. Knight, K. & D. Marcu (2000). Statistics-based summarization – step one: Sentence compression. In Proc. of AAAI-00, pp. 703–71 1. Mani, I. (2001). Automatic Summarization. Amsterdam, Philadelphia: John Benjamins. 1491 McDonald, R. (2006). Discriminative sentence compression with soft syntactic evidence. In Proc. of EACL-06, pp. 297–304. Napoles, C., C. Callison-Burch, J. Ganitkevitch & B. Van Durme (201 1). Paraphrastic sentence compression with a character-based metric: Tightening without deletion. In Proceedings of the Workshop on Monolingual Text-to-text Generation, Prtland, OR, June 24 2011, pp. 84–90. Nivre, J. (2006). Springer. Inductive Dependency Parsing. Nomoto, T. (2008). A generic sentence trimmer with CRFs. In Proc. of ACL-HLT-08, pp. 299–307. Nomoto, T. (2009). A comparison of model free versus model intensive approaches to sentence compression. In Proc. of EMNLP-09, pp. 391–399. Riezler, S., T. H. King, R. Crouch & A. Zaenen (2003). Statistical sentence condensation using ambiguity packing and stochastic disambiguation methods for Lexical-Functional Grammar. In Proc. of HLT-NAACL-03, pp. 118–125. Turner, J. & E. Charniak (2005). Supervised and unsupervised learning for sentence compression. In Proc. of ACL-05, pp. 290–297. Woodsend, K. & M. Lapata (2010). Automatic generation of story highlights. In Proc. of ACL-10, pp. 565–574. Woodsend, K. & M. Lapata (2012). Multiple aspect summarization using Integer Linear Programming. In Proc. of EMNLP-12, pp. 233–243. Wubben, S., A. van den Bosch, E. Krahmer & E. Marsi (2009). Clustering and matching headlines for automatic paraphrase acquisition. In Proc. of ENLG-09, pp. 122–125. Zajic, D., B. J. Dorr, J. Lin & R. Schwartz (2007). Multi-candidate reduction: Sentence compression as a tool for document summarization tasks. Information Processing & Management, Special Issue on Text Summarization, 43(6): 1549–1570.
same-paper 2 0.80996567 121 emnlp-2013-Learning Topics and Positions from Debatepedia
Author: Swapna Gottipati ; Minghui Qiu ; Yanchuan Sim ; Jing Jiang ; Noah A. Smith
Abstract: We explore Debatepedia, a communityauthored encyclopedia of sociopolitical debates, as evidence for inferring a lowdimensional, human-interpretable representation in the domain of issues and positions. We introduce a generative model positing latent topics and cross-cutting positions that gives special treatment to person mentions and opinion words. We evaluate the resulting representation’s usefulness in attaching opinionated documents to arguments and its consistency with human judgments about positions.
3 0.61691928 129 emnlp-2013-Measuring Ideological Proportions in Political Speeches
Author: Yanchuan Sim ; Brice D. L. Acree ; Justin H. Gross ; Noah A. Smith
Abstract: We seek to measure political candidates’ ideological positioning from their speeches. To accomplish this, we infer ideological cues from a corpus of political writings annotated with known ideologies. We then represent the speeches of U.S. Presidential candidates as sequences of cues and lags (filler distinguished only by its length in words). We apply a domain-informed Bayesian HMM to infer the proportions of ideologies each candidate uses in each campaign. The results are validated against a set of preregistered, domain expertauthored hypotheses.
4 0.59327155 51 emnlp-2013-Connecting Language and Knowledge Bases with Embedding Models for Relation Extraction
Author: Jason Weston ; Antoine Bordes ; Oksana Yakhnenko ; Nicolas Usunier
Abstract: This paper proposes a novel approach for relation extraction from free text which is trained to jointly use information from the text and from existing knowledge. Our model is based on scoring functions that operate by learning low-dimensional embeddings of words, entities and relationships from a knowledge base. We empirically show on New York Times articles aligned with Freebase relations that our approach is able to efficiently use the extra information provided by a large subset of Freebase data (4M entities, 23k relationships) to improve over methods that rely on text features alone.
5 0.59140557 48 emnlp-2013-Collective Personal Profile Summarization with Social Networks
Author: Zhongqing Wang ; Shoushan LI ; Fang Kong ; Guodong Zhou
Abstract: Personal profile information on social media like LinkedIn.com and Facebook.com is at the core of many interesting applications, such as talent recommendation and contextual advertising. However, personal profiles usually lack organization confronted with the large amount of available information. Therefore, it is always a challenge for people to find desired information from them. In this paper, we address the task of personal profile summarization by leveraging both personal profile textual information and social networks. Here, using social networks is motivated by the intuition that, people with similar academic, business or social connections (e.g. co-major, co-university, and cocorporation) tend to have similar experience and summaries. To achieve the learning process, we propose a collective factor graph (CoFG) model to incorporate all these resources of knowledge to summarize personal profiles with local textual attribute functions and social connection factors. Extensive evaluation on a large-scale dataset from LinkedIn.com demonstrates the effectiveness of the proposed approach. 1
6 0.58956134 26 emnlp-2013-Assembling the Kazakh Language Corpus
7 0.58813918 64 emnlp-2013-Discriminative Improvements to Distributional Sentence Similarity
8 0.58638746 53 emnlp-2013-Cross-Lingual Discriminative Learning of Sequence Models with Posterior Regularization
9 0.58485001 110 emnlp-2013-Joint Bootstrapping of Corpus Annotations and Entity Types
10 0.58484864 69 emnlp-2013-Efficient Collective Entity Linking with Stacking
11 0.58457625 143 emnlp-2013-Open Domain Targeted Sentiment
12 0.58429903 152 emnlp-2013-Predicting the Presence of Discourse Connectives
13 0.58347869 179 emnlp-2013-Summarizing Complex Events: a Cross-Modal Solution of Storylines Extraction and Reconstruction
14 0.58311498 132 emnlp-2013-Mining Scientific Terms and their Definitions: A Study of the ACL Anthology
15 0.58187002 47 emnlp-2013-Collective Opinion Target Extraction in Chinese Microblogs
16 0.58164573 56 emnlp-2013-Deep Learning for Chinese Word Segmentation and POS Tagging
17 0.58112043 82 emnlp-2013-Exploring Representations from Unlabeled Data with Co-training for Chinese Word Segmentation
18 0.58075088 193 emnlp-2013-Unsupervised Induction of Cross-Lingual Semantic Relations
19 0.58042765 80 emnlp-2013-Exploiting Zero Pronouns to Improve Chinese Coreference Resolution
20 0.57986885 79 emnlp-2013-Exploiting Multiple Sources for Open-Domain Hypernym Discovery