acl acl2013 acl2013-138 knowledge-graph by maker-knowledge-mining
Source: pdf
Author: Gae-won You ; Young-rok Cha ; Jinhan Kim ; Seung-won Hwang
Abstract: This paper studies named entity translation and proposes “selective temporality” as a new feature, as using temporal features may be harmful for translating “atemporal” entities. Our key contribution is building an automatic classifier to distinguish temporal and atemporal entities then align them in separate procedures to boost translation accuracy by 6. 1%.
Reference: text
sentIndex sentText sentNum sentScore
1 Enriching Entity Translation Discovery using Selective Temporality Gae-won You, Young-rok Cha, Jinhan Kim, and Seung-won Hwang Pohang University of Science and Technology, Republic of Korea {gwyou l ine 0 9 3 0 wl sgks 0 8 swhwang}@po stech edu , , , . [sent-1, score-0.024]
2 Abstract This paper studies named entity translation and proposes “selective temporality” as a new feature, as using temporal features may be harmful for translating “atemporal” entities. [sent-2, score-0.498]
3 Our key contribution is building an automatic classifier to distinguish temporal and atemporal entities then align them in separate procedures to boost translation accuracy by 6. [sent-3, score-0.916]
4 1 Introduction Named entity translation discovery aims at mapping entity names for people, locations, etc. [sent-5, score-0.45]
5 in source language into their corresponding names in target language. [sent-6, score-0.048]
6 As many new named entities appear every day in newspapers and web sites, their translations are non-trivial yet essential. [sent-7, score-0.227]
7 Early efforts of named entity translation have focused on using phonetic feature (called PH) to estimate a phonetic similarity between two names (Knight and Graehl, 1998; Li et al. [sent-8, score-0.457]
8 In contrast, some approaches have focused on using context feature (called CX) which compares surrounding words of entities (Fung and Yee, 1998; Diab and Finch, 2000; Laroche and Langlais, 2010). [sent-10, score-0.152]
9 Recently, holistic approaches combining such similarities have been studied (Shao and Ng, 2004; You et al. [sent-11, score-0.06]
10 (Shao and Ng, 2004) rank translation candidates using PH and CX independently and return results with the highest average rank. [sent-14, score-0.085]
11 , 2010) compute initial translation scores using PH and iteratively update the scores using relationshipfeature (called R). [sent-16, score-0.115]
12 More recent approaches consider temporal feature (called T) of entities in two corpora (Klementiev and Roth, 2006; Tao et al. [sent-19, score-0.407]
13 , 2006; Sproat et cyequFnr21350 0 0 05105205Wek305405CEhn5ig0nsilehs cyFnreu q84260 0 510 520 W5e k30 540 5CEhn5ig0nleis he (a) Temporal entity: “Usain Bolt” (b) Atemporal entity: “Hillary Clinton” Figure 1: Illustration on temporality al. [sent-20, score-0.192]
14 T is computed using frequency vectors for entities and combined with PH (Klementiev and Roth, 2006; Tao et al. [sent-23, score-0.152]
15 , 2006) extend Tao’s approach by iteratively updating overall similarities using R. [sent-26, score-0.026]
16 However, T used in previous approaches is a good feature only if temporal behaviors are “symmetric” across corpora. [sent-29, score-0.222]
17 In contrast, Figure 1 illustrates asymmetry, by showing the frequencies of “Usain Bolt,” a Jamaican sprinter, and “Hillary Clinton,” an American politician, in comparable news articles during the year 2008. [sent-30, score-0.103]
18 The former is mostly mentioned in the context of some temporal events, e. [sent-31, score-0.222]
19 In such case, as Hillary Clinton is a famous female leader, she may be associated with other Chinese female leaders in Chinese corpus, while such association is rarely observed in English corpus, which causes asymmetry. [sent-34, score-0.097]
20 That is, Hillary Clinton is “atemporal,” as Figure 1(b) shows, such that using such dissimilarity against deciding this pair as a correct translation would be harmful. [sent-35, score-0.119]
21 In clear contrast, for Usain Bolt, similarity oftemporal dis201 Proce dingSsof oifa, th Beu 5l1gsarti Aan,An u aglu Mste 4e-ti9n2g 0 o1f3 t. [sent-36, score-0.055]
22 c A2s0s1o3ci Aatsiosonc fioartio Cno fmorpu Ctoamtiopnuatalt Lioin gauli Lsitnicgsu,i psatgices 201–205, tributions in Figure 1(a) is a good feature for concluding this pair as a correct one. [sent-38, score-0.034]
23 To overcome such problems, we propose a new notion of “selective temporality” (called this feature ST to distinguish from T) to automatically distinguish temporal and atemporal entities. [sent-39, score-0.699]
24 Toward this goal, we design a classifier to distinguish temporal entities from atemporal entities, based on which we align temporal projections of entity graphs for the temporal ones and the entire entity graphs for the atemporal ones. [sent-40, score-2.112]
25 We also propose a method to identify the optimal window size for temporal entities. [sent-41, score-0.409]
26 We validate this “selective” use of temporal features boosts the accuracy by 6. [sent-42, score-0.222]
27 2 Preliminaries Our approach follows a graph alignment framework proposed in (You et al. [sent-44, score-0.054]
28 1 Step 1: Graph Construction We first build a graph G = (V, E) from each language corpus, where V is a set of entities (nodes) and E is a set of co-occurrence relationships (unweighted edges) between entities. [sent-48, score-0.206]
29 We consider en- tities occurring more than η times as nodes and entity pairs co-occurring more than σ times as edges. [sent-49, score-0.146]
30 To identify entities, we use a CRF-based named entity tagger (Finkel et al. [sent-50, score-0.218]
31 2 Step 2: Initialization Given two graphs Ge = (Ve, Ee) and Gc = (Vc, Ec), we initialize |Ve |-by-|Vc| initial similarity mat)r,ix w Re0 i using eP |HV |a-bndy C|VX| fionirt every pair (e, c) where e ∈ Ve and c ∈ Vc. [sent-54, score-0.182]
32 , 2010) between English entity and a romanized representation of Chinese entity called Pinyin. [sent-56, score-0.32]
33 For CX, the context similarity is computed based on entity context which is defined as a set of words near to the entity (we ignore some words such as stop words and other entities). [sent-57, score-0.347]
34 We compute similarity of the most frequent 20 words for each entity using a variant of Jaccard index. [sent-58, score-0.231]
35 To integrate two similarity scores, we adopt an average as a composite function. [sent-59, score-0.055]
36 ×× We finally compute initial similarity scores for all pairs (e, c) where e ∈ Ve and c ∈ Vc, and build tahlel pianiitrisal ( similarity em ea ∈trix V Ra0n. [sent-60, score-0.14]
37 However, as we cannot assure the correctly matched neighbors (u, v), a chicken-and-egg dilemma, we take advantage of the current similarity Rt to estimate the next similarity Rt+1. [sent-64, score-0.276]
38 Algorithm 1 describes the process of matching the neighbors where N(i) and N(j) are the sets of neighbor nodes of i∈ Ve and j ∈ Vc, respectively, anneidg hHb oisr a priority queue sorting ∈th Ve matched pairs in non-increasing order of similarities. [sent-65, score-0.166]
39 To guarantee that the neighbors are correctly matched, we use only the matches such that R(tu,v) ≥ θ. [sent-66, score-0.061]
40 poro pv( )are matched yet then 6: M ← M ∪ {(u, v) } 7: endM Mif 8: end while 9: return M 2. [sent-70, score-0.105]
41 We then compute the reinforced matrix Rs∞ obtained from the window starting at the timestamp s. [sent-74, score-0.219]
42 To keep the best match scores among all windows, we update R using the best similarity among ∀s, Rs∞. [sent-75, score-0.055]
43 we then extract the candidate itrtayn asmlaotinogn pairs Mours by running step 4. [sent-76, score-0.033]
44 As there can exist atemporal entities in Mours, we classify them (Section 3. [sent-77, score-0.537]
45 Specifically, we build two entire graphs and compute R∞. [sent-79, score-0.123]
46 We then distinguish temporal entities from atemporal ones using our proposed metric for each matched pair (i, j) ∈ Mours and, if the pair is atemporal, R(i,j) i(si updated as the atemporal similarity R(∞i,j) . [sent-80, score-1.418]
47 From the final matrix R, we extract the matched pairs by running step 4 with R once again. [sent-81, score-0.168]
48 1 Projecting Graph for Temporal Entities We first project graphs temporally to improve translation quality for temporal entities. [sent-83, score-0.4]
49 As the optimal projection would differ across entities, we generate many projected graphs by shifting time window over all periods, and then identify the best window for each entity. [sent-84, score-0.415]
50 The rest of this section describes how we set the right window size w. [sent-85, score-0.135]
51 Though each entity may have its own optimal w, we find optimizing for each entity may negatively influence on considering relationships with entities of different window sizes. [sent-86, score-0.604]
52 Thus, we instead find the optimal window size wˆ to maximize the global “symmetry” of the given two graphs. [sent-87, score-0.16]
53 We now define “symmetry” with respect to the truth translation pair M. [sent-88, score-0.174]
54 We note it is infeasible to assume we have M during translation, and will later relax to consider how M can be approximated. [sent-89, score-0.034]
55 We first define the node symmetry Sn as follows: Sn(Ge,Gc;M) =∑(e,mc)∈axVe{×|VVec|I,(|Ve,c|c};M) where I(u, v; M) to be 1 if (u, v) ∈ M, 0 otherwwihseer. [sent-91, score-0.241]
56 e High nvo;dMe symmetry fle (aud,sv t)o ∈acc Mura,t 0e torthanesr-lation in R0 (Initialization step). [sent-92, score-0.265]
57 Similarly, we de- fine the edge symmetry Se as follows: Se(Ge, Gc; M) = ∑(e1,e2)∈Ee ∑(c1,c2)∈Ec I(e1,c1; M)I(e2, c2; M) ∑∑max{|Ee|,|Ec|} In contrast, high edge symmetry leads to accurate translation in R∞ (Reinforcement step). [sent-93, score-0.617]
58 We finally define the symmetry S as the weighted sum of Sn and Se with parameter α (empirically tuned to 0. [sent-94, score-0.241]
59 S(Ge, Gc; M) = αSn(Ge, Gc; M) + (1 − α)Se(Ge, Gc; M) However, as it is infeasible to assume we have the truth translation pair M, we approximate M using intermediate translation results Mours computed at step 4. [sent-96, score-0.326]
60 To insert only true positive pairs in Mours, we set threshold higher than the optimized value from the step 4. [sent-97, score-0.033]
61 We found out that symmetry from Mours closely estimates that from M: S(Ge, Gc; M) ≈ S(Ge, Gc; Mours) Specifically, observe from Table 1 that, given a manually built ground-truth set Mg ⊂ M as described in Section 4. [sent-98, score-0.278]
62 1, S(Ge, Gc; Mours) returns the best symmetry value in two weeks for person entities, which is expectedly the same as the result of S(Ge, Gc; Mg). [sent-99, score-0.319]
63 This suggests that we can use Mours for optimizing window size. [sent-100, score-0.135]
64 As a first step, we observe their characteristics: Temporal entities have peaks in the frequency distribution of both corpora and these peaks are aligned, while such distribution of atemporal entities are more uniform and less aligned. [sent-108, score-0.983]
65 203 Based on these observations, we identify the following criteria for temporal entities: (1) Their two distributions m in English corpus and n in Chinese corpus should have aligned peaks. [sent-109, score-0.249]
66 (2) Frequencies at the peaks are the higher the better. [sent-110, score-0.112]
67 For the first criterion, we first normalize the two vectors mˆ and since two corpora have different scales, i. [sent-111, score-0.033]
68 For the second criterion, we have a spectrum of option from taking the frequencies at all peaks in one extreme, to taking only the maximum frequency in another extreme. [sent-115, score-0.174]
69 A metric representing such a spectrum is p-norm, which represents sum when p = 1and maximum when p = ∞. [sent-116, score-0.034]
70 We empirically =tu 1ne a tnhde m right m buamlan wceh eton distinguish te emm-poral and atemporal entities, which turns out to be p = 2. [sent-117, score-0.455]
71 03 for the distributions in Figure 1(a) and (b), respectively, from which we can determine the translation of Figure 1(a) is temporal and the one of Figure 1(b) is atemporal. [sent-121, score-0.307]
72 1 Experimental Settings We obtained comparable corpora from English and Chinese Gigaword Corpora (LDC2009T13 and LDC2009T27) published by the Xinhua News Agency during the year 2008. [sent-123, score-0.108]
73 From them, we extracted person entities and built two graphs, Ge = (Ve, Ee) and Gc = (Vc, Ec) by setting η = 20 which was used in (Kim et al. [sent-124, score-0.193]
74 Next, we built a ground truth translation pair set Mg for person entities. [sent-126, score-0.215]
75 We first selected 500 person names randomly from English corpus. [sent-127, score-0.089]
76 Among them, only 201 person names were matched to our Chinese corpus. [sent-129, score-0.194]
77 We used all such pairs to identify the best parameters and compute the evaluation measures. [sent-130, score-0.057]
78 We performed a 5-fold cross validation by dividing ground truth into five groups. [sent-139, score-0.055]
79 2 Experimental Results Effect of window size We first validated the effectiveness of our approach for various window sizes (Table 2). [sent-143, score-0.307]
80 Observe that it shows the best performance in two weeks for MRR and F1 measures. [sent-144, score-0.037]
81 We can also observe the effect of selective temporality, which maximizes the symmetry between two graphs as shown in Table 1, i. [sent-154, score-0.478]
82 Figure 2: The translation examples where shaded cells indicate the correctly translated pairs. [sent-225, score-0.085]
83 All approaches found famous entities such as “Hu Jintao,” a former leader of China, but (PH+CX+R) failed to find translation of lesser known entities, such as “Kim Yong Nam. [sent-231, score-0.338]
84 ” Using temporal features help both (PH+CX+R+T) and (PH+CX+R+ST) identify the right translation, as Kim’s temporal occurrence is strong and symmetric in both corpora. [sent-232, score-0.471]
85 In contrast, (PH+CX+R+T) failed to find the translation of “Karzai”, the president of Afghanistan, as it only appears weakly and transiently during a short period time, for which only (PH+CX+R+ST) applying varying sizes of window per entity is effective. [sent-233, score-0.394]
86 5 Conclusion This paper validated that considering temporality selectively is helpful for improving the trans- lation quality. [sent-234, score-0.229]
87 We developed a classifier to dis- tinguish temporal/atemporal entities and our proposed method outperforms the state-of-the-art approach by 6. [sent-235, score-0.152]
88 A statistical word level translation model for comparable corpora. [sent-241, score-0.134]
89 Named entity transliteration and discovery from multilingual comparable corpora. [sent-265, score-0.277]
90 Revisiting context-based projection methods for term-translation spotting in comparable corpora. [sent-273, score-0.049]
wordName wordTfidf (topN-words)
[('atemporal', 0.385), ('ph', 0.33), ('cx', 0.263), ('mours', 0.247), ('symmetry', 0.241), ('temporal', 0.222), ('temporality', 0.192), ('gc', 0.185), ('ge', 0.181), ('entities', 0.152), ('vc', 0.148), ('entity', 0.146), ('window', 0.135), ('peaks', 0.112), ('kim', 0.11), ('selective', 0.107), ('matched', 0.105), ('tao', 0.105), ('ve', 0.105), ('graphs', 0.093), ('clinton', 0.09), ('hillary', 0.09), ('translation', 0.085), ('mrr', 0.084), ('usain', 0.082), ('hwang', 0.077), ('sproat', 0.077), ('jinhan', 0.073), ('chinese', 0.07), ('mg', 0.063), ('shao', 0.063), ('neighbors', 0.061), ('transliteration', 0.057), ('similarity', 0.055), ('virga', 0.055), ('truth', 0.055), ('graph', 0.054), ('ee', 0.052), ('klementiev', 0.052), ('bolt', 0.051), ('laroche', 0.049), ('ec', 0.049), ('comparable', 0.049), ('names', 0.048), ('distinguish', 0.046), ('named', 0.045), ('sn', 0.044), ('leader', 0.042), ('person', 0.041), ('st', 0.04), ('song', 0.04), ('phonetic', 0.039), ('observe', 0.037), ('validated', 0.037), ('weeks', 0.037), ('reinforcement', 0.035), ('infeasible', 0.034), ('holistic', 0.034), ('spectrum', 0.034), ('pair', 0.034), ('corpora', 0.033), ('female', 0.033), ('step', 0.033), ('rt', 0.033), ('se', 0.032), ('chengxiang', 0.031), ('famous', 0.031), ('translations', 0.03), ('finkel', 0.03), ('compute', 0.03), ('matrix', 0.03), ('jiang', 0.029), ('mu', 0.029), ('frequencies', 0.028), ('fung', 0.028), ('called', 0.028), ('failed', 0.028), ('initialization', 0.027), ('windows', 0.027), ('identify', 0.027), ('diab', 0.026), ('similarities', 0.026), ('year', 0.026), ('boost', 0.026), ('discovery', 0.025), ('agency', 0.025), ('optimal', 0.025), ('edge', 0.025), ('reinforced', 0.024), ('afghanistan', 0.024), ('mke', 0.024), ('nipa', 0.024), ('pohang', 0.024), ('stech', 0.024), ('swhwang', 0.024), ('fle', 0.024), ('audrey', 0.024), ('mrc', 0.024), ('eton', 0.024), ('hor', 0.024)]
simIndex simValue paperId paperTitle
same-paper 1 1.0000002 138 acl-2013-Enriching Entity Translation Discovery using Selective Temporality
Author: Gae-won You ; Young-rok Cha ; Jinhan Kim ; Seung-won Hwang
Abstract: This paper studies named entity translation and proposes “selective temporality” as a new feature, as using temporal features may be harmful for translating “atemporal” entities. Our key contribution is building an automatic classifier to distinguish temporal and atemporal entities then align them in separate procedures to boost translation accuracy by 6. 1%.
2 0.19174285 71 acl-2013-Bootstrapping Entity Translation on Weakly Comparable Corpora
Author: Taesung Lee ; Seung-won Hwang
Abstract: This paper studies the problem of mining named entity translations from comparable corpora with some “asymmetry”. Unlike the previous approaches relying on the “symmetry” found in parallel corpora, the proposed method is tolerant to asymmetry often found in comparable corpora, by distinguishing different semantics of relations of entity pairs to selectively propagate seed entity translations on weakly comparable corpora. Our experimental results on English-Chinese corpora show that our selective propagation approach outperforms the previous approaches in named entity translation in terms of the mean reciprocal rank by up to 0.16 for organization names, and 0.14 in a low com- parability case.
3 0.15335751 339 acl-2013-Temporal Signals Help Label Temporal Relations
Author: Leon Derczynski ; Robert Gaizauskas
Abstract: Automatically determining the temporal order of events and times in a text is difficult, though humans can readily perform this task. Sometimes events and times are related through use of an explicit co-ordination which gives information about the temporal relation: expressions like “before ” and “as soon as”. We investigate the r oˆle that these co-ordinating temporal signals have in determining the type of temporal relations in discourse. Using machine learning, we improve upon prior approaches to the problem, achieving over 80% accuracy at labelling the types of temporal relation between events and times that are related by temporal signals.
4 0.12664086 212 acl-2013-Language-Independent Discriminative Parsing of Temporal Expressions
Author: Gabor Angeli ; Jakob Uszkoreit
Abstract: Temporal resolution systems are traditionally tuned to a particular language, requiring significant human effort to translate them to new languages. We present a language independent semantic parser for learning the interpretation of temporal phrases given only a corpus of utterances and the times they reference. We make use of a latent parse that encodes a language-flexible representation of time, and extract rich features over both the parse and associated temporal semantics. The parameters of the model are learned using a weakly supervised bootstrapping approach, without the need for manually tuned parameters or any other language expertise. We achieve state-of-the-art accuracy on all languages in the TempEval2 temporal normalization task, reporting a 4% improvement in both English and Spanish accuracy, and to our knowledge the first results for four other languages.
5 0.10254849 164 acl-2013-FudanNLP: A Toolkit for Chinese Natural Language Processing
Author: Xipeng Qiu ; Qi Zhang ; Xuanjing Huang
Abstract: The growing need for Chinese natural language processing (NLP) is largely in a range of research and commercial applications. However, most of the currently Chinese NLP tools or components still have a wide range of issues need to be further improved and developed. FudanNLP is an open source toolkit for Chinese natural language processing (NLP) , which uses statistics-based and rule-based methods to deal with Chinese NLP tasks, such as word segmentation, part-ofspeech tagging, named entity recognition, dependency parsing, time phrase recognition, anaphora resolution and so on.
6 0.096148342 153 acl-2013-Extracting Events with Informal Temporal References in Personal Histories in Online Communities
7 0.096085757 160 acl-2013-Fine-grained Semantic Typing of Emerging Entities
8 0.087732382 179 acl-2013-HYENA-live: Fine-Grained Online Entity Type Classification from Natural-language Text
9 0.086837523 352 acl-2013-Towards Accurate Distant Supervision for Relational Facts Extraction
10 0.086632788 25 acl-2013-A Tightly-coupled Unsupervised Clustering and Bilingual Alignment Model for Transliteration
11 0.085370757 139 acl-2013-Entity Linking for Tweets
12 0.084635824 169 acl-2013-Generating Synthetic Comparable Questions for News Articles
13 0.084217593 255 acl-2013-Name-aware Machine Translation
14 0.078930475 296 acl-2013-Recognizing Identical Events with Graph Kernels
15 0.078443095 174 acl-2013-Graph Propagation for Paraphrasing Out-of-Vocabulary Words in Statistical Machine Translation
16 0.07127168 93 acl-2013-Context Vector Disambiguation for Bilingual Lexicon Extraction from Comparable Corpora
17 0.069480896 301 acl-2013-Resolving Entity Morphs in Censored Data
18 0.065982789 172 acl-2013-Graph-based Local Coherence Modeling
19 0.065075129 307 acl-2013-Scalable Decipherment for Machine Translation via Hash Sampling
20 0.063259825 223 acl-2013-Learning a Phrase-based Translation Model from Monolingual Data with Application to Domain Adaptation
topicId topicWeight
[(0, 0.167), (1, -0.002), (2, 0.03), (3, -0.032), (4, 0.116), (5, 0.093), (6, -0.054), (7, 0.034), (8, 0.028), (9, -0.007), (10, 0.004), (11, -0.078), (12, -0.006), (13, -0.012), (14, 0.039), (15, -0.019), (16, -0.013), (17, -0.009), (18, -0.116), (19, 0.001), (20, 0.025), (21, -0.084), (22, 0.05), (23, 0.128), (24, -0.085), (25, 0.004), (26, -0.001), (27, -0.001), (28, -0.142), (29, 0.033), (30, 0.023), (31, -0.095), (32, 0.117), (33, -0.115), (34, 0.066), (35, -0.064), (36, -0.104), (37, -0.055), (38, -0.005), (39, -0.005), (40, -0.043), (41, -0.032), (42, 0.047), (43, 0.057), (44, -0.018), (45, -0.038), (46, -0.045), (47, -0.016), (48, 0.026), (49, -0.1)]
simIndex simValue paperId paperTitle
same-paper 1 0.94540519 138 acl-2013-Enriching Entity Translation Discovery using Selective Temporality
Author: Gae-won You ; Young-rok Cha ; Jinhan Kim ; Seung-won Hwang
Abstract: This paper studies named entity translation and proposes “selective temporality” as a new feature, as using temporal features may be harmful for translating “atemporal” entities. Our key contribution is building an automatic classifier to distinguish temporal and atemporal entities then align them in separate procedures to boost translation accuracy by 6. 1%.
2 0.74782044 71 acl-2013-Bootstrapping Entity Translation on Weakly Comparable Corpora
Author: Taesung Lee ; Seung-won Hwang
Abstract: This paper studies the problem of mining named entity translations from comparable corpora with some “asymmetry”. Unlike the previous approaches relying on the “symmetry” found in parallel corpora, the proposed method is tolerant to asymmetry often found in comparable corpora, by distinguishing different semantics of relations of entity pairs to selectively propagate seed entity translations on weakly comparable corpora. Our experimental results on English-Chinese corpora show that our selective propagation approach outperforms the previous approaches in named entity translation in terms of the mean reciprocal rank by up to 0.16 for organization names, and 0.14 in a low com- parability case.
3 0.69591361 339 acl-2013-Temporal Signals Help Label Temporal Relations
Author: Leon Derczynski ; Robert Gaizauskas
Abstract: Automatically determining the temporal order of events and times in a text is difficult, though humans can readily perform this task. Sometimes events and times are related through use of an explicit co-ordination which gives information about the temporal relation: expressions like “before ” and “as soon as”. We investigate the r oˆle that these co-ordinating temporal signals have in determining the type of temporal relations in discourse. Using machine learning, we improve upon prior approaches to the problem, achieving over 80% accuracy at labelling the types of temporal relation between events and times that are related by temporal signals.
4 0.65252674 212 acl-2013-Language-Independent Discriminative Parsing of Temporal Expressions
Author: Gabor Angeli ; Jakob Uszkoreit
Abstract: Temporal resolution systems are traditionally tuned to a particular language, requiring significant human effort to translate them to new languages. We present a language independent semantic parser for learning the interpretation of temporal phrases given only a corpus of utterances and the times they reference. We make use of a latent parse that encodes a language-flexible representation of time, and extract rich features over both the parse and associated temporal semantics. The parameters of the model are learned using a weakly supervised bootstrapping approach, without the need for manually tuned parameters or any other language expertise. We achieve state-of-the-art accuracy on all languages in the TempEval2 temporal normalization task, reporting a 4% improvement in both English and Spanish accuracy, and to our knowledge the first results for four other languages.
5 0.58056813 160 acl-2013-Fine-grained Semantic Typing of Emerging Entities
Author: Ndapandula Nakashole ; Tomasz Tylenda ; Gerhard Weikum
Abstract: Methods for information extraction (IE) and knowledge base (KB) construction have been intensively studied. However, a largely under-explored case is tapping into highly dynamic sources like news streams and social media, where new entities are continuously emerging. In this paper, we present a method for discovering and semantically typing newly emerging out-ofKB entities, thus improving the freshness and recall of ontology-based IE and improving the precision and semantic rigor of open IE. Our method is based on a probabilistic model that feeds weights into integer linear programs that leverage type signatures of relational phrases and type correlation or disjointness constraints. Our experimental evaluation, based on crowdsourced user studies, show our method performing significantly better than prior work.
6 0.54430151 301 acl-2013-Resolving Entity Morphs in Censored Data
7 0.54357541 352 acl-2013-Towards Accurate Distant Supervision for Relational Facts Extraction
8 0.54110134 139 acl-2013-Entity Linking for Tweets
9 0.53162646 153 acl-2013-Extracting Events with Informal Temporal References in Personal Histories in Online Communities
10 0.49881467 179 acl-2013-HYENA-live: Fine-Grained Online Entity Type Classification from Natural-language Text
11 0.49655199 242 acl-2013-Mining Equivalent Relations from Linked Data
12 0.4878521 219 acl-2013-Learning Entity Representation for Entity Disambiguation
13 0.47039738 256 acl-2013-Named Entity Recognition using Cross-lingual Resources: Arabic as an Example
14 0.45759755 164 acl-2013-FudanNLP: A Toolkit for Chinese Natural Language Processing
15 0.45333597 236 acl-2013-Mapping Source to Target Strings without Alignment by Analogical Learning: A Case Study with Transliteration
16 0.45098191 296 acl-2013-Recognizing Identical Events with Graph Kernels
17 0.44568866 255 acl-2013-Name-aware Machine Translation
18 0.44308278 169 acl-2013-Generating Synthetic Comparable Questions for News Articles
19 0.41242489 178 acl-2013-HEADY: News headline abstraction through event pattern clustering
20 0.41113937 93 acl-2013-Context Vector Disambiguation for Bilingual Lexicon Extraction from Comparable Corpora
topicId topicWeight
[(0, 0.042), (6, 0.049), (11, 0.115), (15, 0.021), (24, 0.052), (26, 0.033), (28, 0.01), (35, 0.072), (42, 0.026), (48, 0.034), (67, 0.282), (70, 0.074), (88, 0.024), (90, 0.015), (95, 0.08)]
simIndex simValue paperId paperTitle
1 0.86994821 100 acl-2013-Crowdsourcing Interaction Logs to Understand Text Reuse from the Web
Author: Martin Potthast ; Matthias Hagen ; Michael Volske ; Benno Stein
Abstract: unkown-abstract
same-paper 2 0.81926501 138 acl-2013-Enriching Entity Translation Discovery using Selective Temporality
Author: Gae-won You ; Young-rok Cha ; Jinhan Kim ; Seung-won Hwang
Abstract: This paper studies named entity translation and proposes “selective temporality” as a new feature, as using temporal features may be harmful for translating “atemporal” entities. Our key contribution is building an automatic classifier to distinguish temporal and atemporal entities then align them in separate procedures to boost translation accuracy by 6. 1%.
3 0.7208153 388 acl-2013-Word Alignment Modeling with Context Dependent Deep Neural Network
Author: Nan Yang ; Shujie Liu ; Mu Li ; Ming Zhou ; Nenghai Yu
Abstract: In this paper, we explore a novel bilingual word alignment approach based on DNN (Deep Neural Network), which has been proven to be very effective in various machine learning tasks (Collobert et al., 2011). We describe in detail how we adapt and extend the CD-DNNHMM (Dahl et al., 2012) method introduced in speech recognition to the HMMbased word alignment model, in which bilingual word embedding is discriminatively learnt to capture lexical translation information, and surrounding words are leveraged to model context information in bilingual sentences. While being capable to model the rich bilingual correspondence, our method generates a very compact model with much fewer parameters. Experiments on a large scale EnglishChinese word alignment task show that the proposed method outperforms the HMM and IBM model 4 baselines by 2 points in F-score.
4 0.69959271 110 acl-2013-Deepfix: Statistical Post-editing of Statistical Machine Translation Using Deep Syntactic Analysis
Author: Rudolf Rosa ; David Marecek ; Ales Tamchyna
Abstract: Deepfix is a statistical post-editing system for improving the quality of statistical machine translation outputs. It attempts to correct errors in verb-noun valency using deep syntactic analysis and a simple probabilistic model of valency. On the English-to-Czech translation pair, we show that statistical post-editing of statistical machine translation leads to an improvement of the translation quality when helped by deep linguistic knowledge.
5 0.683025 309 acl-2013-Scaling Semi-supervised Naive Bayes with Feature Marginals
Author: Michael Lucas ; Doug Downey
Abstract: Semi-supervised learning (SSL) methods augment standard machine learning (ML) techniques to leverage unlabeled data. SSL techniques are often effective in text classification, where labeled data is scarce but large unlabeled corpora are readily available. However, existing SSL techniques typically require multiple passes over the entirety of the unlabeled data, meaning the techniques are not applicable to large corpora being produced today. In this paper, we show that improving marginal word frequency estimates using unlabeled data can enable semi-supervised text classification that scales to massive unlabeled data sets. We present a novel learning algorithm, which optimizes a Naive Bayes model to accord with statistics calculated from the unlabeled corpus. In experiments with text topic classification and sentiment analysis, we show that our method is both more scalable and more accurate than SSL techniques from previous work.
6 0.67396981 252 acl-2013-Multigraph Clustering for Unsupervised Coreference Resolution
7 0.56607562 216 acl-2013-Large tagset labeling using Feed Forward Neural Networks. Case study on Romanian Language
8 0.55680186 169 acl-2013-Generating Synthetic Comparable Questions for News Articles
9 0.55578315 275 acl-2013-Parsing with Compositional Vector Grammars
10 0.55508834 254 acl-2013-Multimodal DBN for Predicting High-Quality Answers in cQA portals
11 0.55136895 38 acl-2013-Additive Neural Networks for Statistical Machine Translation
12 0.55121613 242 acl-2013-Mining Equivalent Relations from Linked Data
13 0.54794061 245 acl-2013-Modeling Human Inference Process for Textual Entailment Recognition
14 0.54436815 155 acl-2013-Fast and Accurate Shift-Reduce Constituent Parsing
15 0.54242265 272 acl-2013-Paraphrase-Driven Learning for Open Question Answering
16 0.54165846 61 acl-2013-Automatic Interpretation of the English Possessive
17 0.53986812 333 acl-2013-Summarization Through Submodularity and Dispersion
18 0.53852153 156 acl-2013-Fast and Adaptive Online Training of Feature-Rich Translation Models
19 0.53838611 154 acl-2013-Extracting bilingual terminologies from comparable corpora
20 0.53586948 215 acl-2013-Large-scale Semantic Parsing via Schema Matching and Lexicon Extension