acl acl2013 acl2013-212 knowledge-graph by maker-knowledge-mining
Source: pdf
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.
Reference: text
sentIndex sentText sentNum sentScore
1 edu i Abstract Temporal resolution systems are traditionally tuned to a particular language, requiring significant human effort to translate them to new languages. [sent-2, score-0.078]
2 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. [sent-3, score-0.637]
3 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. [sent-4, score-0.735]
4 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. [sent-5, score-0.083]
5 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. [sent-6, score-0.481]
6 1 Introduction Temporal resolution is the task of mapping from a textual phrase describing a potentially complex time, date, or duration to a normalized (grounded) temporal representation. [sent-7, score-0.607]
7 For example, possibly complex phrases such as the week before last1 are often more useful in their grounded form e. [sent-8, score-0.383]
8 Many approaches to this problem make use of rule-based methods, combining regularexpression matching and hand-written interpretation functions. [sent-11, score-0.092]
9 In contrast, we would like to learn the interpretation of a temporal expression probabilistically. [sent-12, score-0.54]
10 In addition, we would like to use a representation of time which is broadly applicable to multiple languages, without the need for language-specific rules or manually tuned parameters. [sent-17, score-0.165]
11 Our system requires annotated data consisting only of an input phrase and an associated grounded time, relative to some reference time; the language-flexible parse is entirely latent. [sent-18, score-0.36]
12 Training data of this weakly-supervised form is generally easier to collect than the alternative of manually creating and tuning potentially complex interpretation rules. [sent-19, score-0.092]
13 A large number oflanguages conceptualize time as lying on a one dimensional line. [sent-20, score-0.124]
14 Although the surface forms of temporal expressions differ, the basic operations many languages use can be mapped to operations on this time line (see Section 3). [sent-21, score-0.812]
15 Furthermore, many common languages share temporal units (hours, weekdays, etc. [sent-22, score-0.481]
16 By structuring a latent parse to reflect these semantics, we can define a single model which performs well on multiple languages. [sent-24, score-0.191]
17 A discriminative parsing model allows us to define sparse features over not only lexical cues but also the temporal value of our prediction. [sent-25, score-0.577]
18 We briefly describe our temporal representation and grammar, followed by a description of the 14th – learning algorithm; we conclude with experimental results on the six languages of the TempEval-2 A task. [sent-28, score-0.519]
19 (2012), both in the bootstrapping training methodology and the temporal grammar. [sent-32, score-0.493]
20 The latent parse parallels the formal semantics in previous work. [sent-35, score-0.148]
21 For example, Zettlemoyer and Collins (2007) learn a mapping from textual queries to a logical form. [sent-38, score-0.062]
22 Importantly, the logical form of these parses contain all of the predicates and entities used in the parse unlike the label provided in our case, where a grounded time can correspond to any of a number of latent parses. [sent-39, score-0.584]
23 (201 1) relax supervision to require only annotated answers rather than full logical forms. [sent-42, score-0.062]
24 Related work on interpreting temporal expressions has focused on constructing hand-crafted interpretation rules (Mani and Wilson, 2000; Saquete et al. [sent-43, score-0.632]
25 Recent probabilistic approaches to temporal resolution include UzZaman and Allen (2010), who employ a parser to produce deep logical forms, in conjunction with a CRF classifier. [sent-47, score-0.614]
26 In a similar vein, Kolomiyets and Moens (2010) em– ploy a maximum entropy classifier to detect the location and temporal type of expressions; the grounding is then done via deterministic rules. [sent-48, score-0.651]
27 , 2010) was the only system to perform bilingual interpretation for English and Spanish. [sent-52, score-0.092]
28 3 Temporal Representation We define a compositional representation of time, similar to Angeli et al. [sent-54, score-0.116]
29 (2012), but with a greater focus on efficiency and simplicity. [sent-55, score-0.048]
30 The representation makes use of a notion of temporal types and their associated semantic values; a grammar is constructed over these types, and is grounded by appealing to the associated values. [sent-56, score-0.801]
31 A summary of the temporal type system is provided in Section 3. [sent-57, score-0.504]
32 1 Temporal Types Temporal expressions are represented either as a Range, Sequence, or Duration. [sent-62, score-0.092]
33 The root of a parse tree should be one of these types. [sent-63, score-0.135]
34 In addition, phrases can be tagged as a Function; or, as a special Nil type corresponding to segments without a direct temporal interpretation. [sent-64, score-0.554]
35 Range [and Instant] A period between two dates (or times), as per an interval-based theory of time (Allen, 1981). [sent-67, score-0.17]
36 This includes entities such as Today, 19 8 7 , or Now. [sent-68, score-0.044]
37 Sequence A sequence of Ranges, occurring at regular but not necessarily constant intervals. [sent-69, score-0.131]
38 For example, November 2 7 th would define a Sequence whose year is unspecified, month 27th; is November, and day is the spanning the entire range of the lower order fields (in this case, a day). [sent-72, score-0.448]
39 Note that a Sequence implicitly selects a possibly infinite number of possible Ranges. [sent-74, score-0.045]
40 To select a particular grounded time for a Sequence, we appeal to a notion of a reference time (Reichenbach, 1947). [sent-75, score-0.437]
41 For the TempEval-2 corpus, we approximate this as the publication time of the article. [sent-76, score-0.143]
42 While this is conflating Reichenbach’s reference time with speech time, and comes at the expense of certain mistakes (see Section 5. [sent-77, score-0.186]
43 To a first approximation, grounding a sequence given a reference time corresponds to filling in the unspecified fields of the sequence with the fullyspecified fields of the reference time. [sent-79, score-0.849]
44 This pro84 mon day Sequence: y—ear wNeoevk 2w7tehe–kd 2a8yth h0ou0r m0i0n s0e0c mon —mon —day Reference Time: 2y0ea13r wAeuegk we0e6ktdhay h0ou3r m2i5n s0e0c 32 Tue Figure 1: An illustration of grounding a Sequence. [sent-80, score-0.446]
45 with a reference time 2 0 1 3-0 8 -0 6 day 2y0ea13r wNeoevk 2w7tehe–kd 2a8yth h0ou0r m0i0n s0e0c — — When grounding the Sequence Novembe r 2 7 th 0 3 :2 5 : 0 0, we complete the missing fields in the Sequence (the year) with the corresponding field in the reference time (2013). [sent-81, score-0.692]
46 cess has a number of special cases not enumerated here,2 but the complexity remains constant time. [sent-82, score-0.134]
47 This includes entities like Week, Month, and 7 days . [sent-84, score-0.145]
48 A special case of the Duration type is defined to represent approximate durations, such as a few years or some days. [sent-85, score-0.16]
49 This captures semantic entities such as those implied in last x, the third x [of y], or x days ago. [sent-87, score-0.183]
50 Nil A special Nil type denotes terms which are not directly contributing to the semantic meaning of the expression. [sent-89, score-0.106]
51 This is intended for words such as a or the, which serve as cues without bearing temporal content themselves. [sent-90, score-0.448]
52 Number Lastly, a special Number type is defined for tagging numeric expressions. [sent-91, score-0.195]
53 2 Temporal Grammar Our approach assumes that natural language descriptions of time are compositional in nature; that is, each word attached to a temporal phrase is compositionally modifying the meaning of the phrase. [sent-93, score-0.572]
54 We define a grammar jointly over temporal types and values. [sent-94, score-0.644]
55 The types serve to constrain the parse and allow for coarse features; the values encode specific semantics, and allow for finer features. [sent-95, score-0.176]
56 At the root of a parse tree, we recursively apply 2Some of these special cases are caused by variable days of the month, daylight savings time, etc. [sent-96, score-0.287]
57 , the next Monday in August uttered in the last week of August should ground to August of next year (rather than the reference time’s year). [sent-99, score-0.451]
58 the functions in the tree to obtain a final temporal value. [sent-100, score-0.519]
59 Formally, we define our grammar as G = (Σ, S, V, T, R). [sent-103, score-0.157]
60 VFo tor beaec thh v ∈ V of we dees,fin ase an (infinite) sSeetc Tv corresponding vto ∈ th Ve possible i annst (ainncfiensi eof) type v. [sent-108, score-0.056]
61 The structure of the tree is bound by the first part over types v – these types are used to populate the chart, and allow for efficient inference. [sent-118, score-0.112]
62 Each preterminal consists of a type and a value; neither which are lexically informed. [sent-122, score-0.136]
63 That is, the word week and preterminal (Week, Duration) are not tied in any way. [sent-123, score-0.301]
64 A total of 62 preterminals are defined corresponding to instances of Ranges, Sequences, and Durations; these are summarized in Table 1. [sent-124, score-0.065]
65 In addition, 10 functions are defined for manipulating temporal expressions (see Table 2). [sent-125, score-0.577]
66 The majority of these mirror generic operations on intervals on a timeline, or manipulations of a sequence. [sent-126, score-0.204]
67 Note that the Sequence type contains more elements than enumerated here; however, only a few of each characteristic type are shown here for brevity. [sent-128, score-0.196]
68 The name and a brief description of the function are given; the functions are most easily interpreted as operations on either an interval or sequence. [sent-130, score-0.18]
69 All operations on Ranges can equivalently be applied to Sequences. [sent-131, score-0.075]
70 moved (3 weeks ago) or their size changed (the first two days of the month), or a new interval can be started from one of the endpoints (the last 2 days). [sent-132, score-0.209]
71 Additionally, a sequence can be modified by shifting its origin (last Friday), or taking the element of the sequence within some bound (fourth Sunday in November). [sent-133, score-0.262]
72 Combination rules in the grammar mirror typechecked curried function application. [sent-134, score-0.199]
73 For instance, the function moveLe ft 1applied to week (as in last week) yields a grammar rule: nth (EveryWeek -1 ,Seq. [sent-135, score-0.406]
74 ) In more generality, we create grammar rules for applying a function on either the left or the right, for all possible type signatures of f: f(x, y) ? [sent-139, score-0.203]
75 Additionally, a grammar rule is created for intersecting two Ranges or Sequences, for multiplying a duration by a number, and for absorbing a Nil span. [sent-142, score-0.392]
76 Each of these can be though of as an implicit function application (in the last case, the identity function). [sent-143, score-0.071]
77 3 Differences From Previous Work While the grammar formalism is strongly inspired by Angeli et al. [sent-145, score-0.114]
78 Sequence Grounding The most timeconsuming and conceptually nuanced aspect of temporal inference in Angeli et al. [sent-147, score-0.448]
79 In particular, there are two modes of expressing dates which resist intersection: a day-of-month-based mode and a week-based mode. [sent-149, score-0.045]
80 Properly grounding a sequence which defines both a day of the month and a day of the week (or week of the year) requires backing off to an expensive search problem. [sent-150, score-1.107]
81 To illustrate, consider the example: Friday the Although both a Friday and a of the month are easily found, the intersection of the two requires iterating through elements of one until it 13th. [sent-151, score-0.173]
82 At training time, a number of candidate parses are generated for each phrase. [sent-153, score-0.079]
83 When considering that these parses can become both complex and pragmatically unreasonable, this can result in a noticeable efficiency hit; e. [sent-154, score-0.246]
84 , during training a sentence could have a [likely incorrect] candidate interpretation of: nineteen ninety-six Friday the now. [sent-156, score-0.092]
85 13ths from 30th Sequence Pragmatics For the sake of simplicity the pragmatic distribution over possible groundings of a sequence is replaced with the single most likely offset, as learned empirically from the English TempEval-2 corpus by Angeli et al. [sent-158, score-0.166]
86 86 More precisely, there is a single nonterminal (Nil), rather than a nonterminal symbol characterizing the phrase it is subsuming (Nil-the, Nil-a, etc. [sent-161, score-0.09]
87 We describe the parser below, followed by the features implemented. [sent-165, score-0.064]
88 1 Parser Inference A discriminative k-best parser was used to allow for arbitrary features in the parse tree. [sent-167, score-0.208]
89 A rule-based number recognizer was used for each language to recognize and ground numeric expressions, including information on whether the number was an ordinal (e. [sent-169, score-0.089]
90 Numeric expressions are treated as if the numeric value replaced the expression. [sent-173, score-0.181]
91 Each rule of the parse derivation was assigned a score according to a log-linear factor. [sent-174, score-0.154]
92 However, we can approximate the algorithm in O(n3k log k) time with cube pruning (Chiang, 2007). [sent-176, score-0.143]
93 With features which are not context-free, we are not guaranteed an optimal beam with this approach; however, empirically the approximation yields a significant efficiency improvement without noticeable loss in performance. [sent-177, score-0.17]
94 Training We adopt an EM-style bootstrapping approach similar to Angeli et al. [sent-178, score-0.045]
95 (2012), in order to handle the task of parsing the temporal expression without annotations for the latent parses. [sent-179, score-0.538]
96 Each training instance is a tuple consisting of the words in the temporal phrase, the annotated grounded time τ∗, and the reference time. [sent-180, score-0.796]
97 Given an input sentence, our parser will output k possible parses; when grounded to the reference time these correspond to k candidate times: τ1 . [sent-181, score-0.412]
98 This corresponds to an approximate E step in the EM algorithm, where the distribution over latent parses is approximated by a beam of size k. [sent-185, score-0.248]
99 Although for long sentences the number of parses is far greater than the beam size, as the parameters improve, increasingly longersentences will have correct derivations in the beam. [sent-186, score-0.147]
100 To approximate the M step, we define a multiclass hinge loss l(θ) over the beam, and optimize using Stochastic Gradient Descent with AdaGrad (Duchi et al. [sent-188, score-0.097]
wordName wordTfidf (topN-words)
[('temporal', 0.448), ('angeli', 0.26), ('week', 0.221), ('nil', 0.212), ('grounded', 0.162), ('friday', 0.153), ('grounding', 0.147), ('month', 0.137), ('august', 0.134), ('sequence', 0.131), ('day', 0.125), ('november', 0.123), ('duration', 0.119), ('grammar', 0.114), ('days', 0.101), ('parse', 0.101), ('reference', 0.097), ('year', 0.095), ('expressions', 0.092), ('interpretation', 0.092), ('numeric', 0.089), ('time', 0.089), ('mon', 0.087), ('enumerated', 0.084), ('allen', 0.081), ('everyweek', 0.08), ('preterminal', 0.08), ('tvi', 0.08), ('vjvk', 0.08), ('wneoevk', 0.08), ('parses', 0.079), ('operations', 0.075), ('intersecting', 0.071), ('reichenbach', 0.071), ('ranges', 0.07), ('beam', 0.068), ('preterminals', 0.065), ('pragmatically', 0.065), ('parser', 0.064), ('zettlemoyer', 0.063), ('logical', 0.062), ('unspecified', 0.061), ('durations', 0.061), ('vi', 0.057), ('type', 0.056), ('approximate', 0.054), ('noticeable', 0.054), ('kd', 0.054), ('rule', 0.053), ('mirror', 0.052), ('special', 0.05), ('march', 0.049), ('fields', 0.048), ('efficiency', 0.048), ('latent', 0.047), ('bootstrapping', 0.045), ('infinite', 0.045), ('nonterminal', 0.045), ('dates', 0.045), ('entities', 0.044), ('define', 0.043), ('discriminative', 0.043), ('parsing', 0.043), ('intervals', 0.042), ('resolution', 0.04), ('types', 0.039), ('nonetheless', 0.039), ('collins', 0.039), ('lastly', 0.038), ('last', 0.038), ('tuned', 0.038), ('representation', 0.038), ('functions', 0.037), ('synchronous', 0.037), ('coarse', 0.036), ('intersection', 0.036), ('period', 0.036), ('manipulations', 0.035), ('calendar', 0.035), ('gertz', 0.035), ('heideltime', 0.035), ('kolomiyets', 0.035), ('sutime', 0.035), ('daylight', 0.035), ('absorbing', 0.035), ('endpoints', 0.035), ('foremost', 0.035), ('groundings', 0.035), ('grover', 0.035), ('llorens', 0.035), ('oflanguages', 0.035), ('saquete', 0.035), ('tipsem', 0.035), ('tvj', 0.035), ('compositional', 0.035), ('interval', 0.035), ('tree', 0.034), ('utterances', 0.033), ('languages', 0.033), ('function', 0.033)]
simIndex simValue paperId paperTitle
same-paper 1 1.0000002 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.
2 0.31396756 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.
3 0.25354409 153 acl-2013-Extracting Events with Informal Temporal References in Personal Histories in Online Communities
Author: Miaomiao Wen ; Zeyu Zheng ; Hyeju Jang ; Guang Xiang ; Carolyn Penstein Rose
Abstract: We present a system for extracting the dates of illness events (year and month of the event occurrence) from posting histories in the context of an online medical support community. A temporal tagger retrieves and normalizes dates mentioned informally in social media to actual month and year referents. Building on this, an event date extraction system learns to integrate the likelihood of candidate dates extracted from time-rich sentences with temporal constraints extracted from eventrelated sentences. Our integrated model achieves 89.7% of the maximum performance given the performance of the temporal expression retrieval step.
4 0.13004139 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.
5 0.1280158 36 acl-2013-Adapting Discriminative Reranking to Grounded Language Learning
Author: Joohyun Kim ; Raymond Mooney
Abstract: We adapt discriminative reranking to improve the performance of grounded language acquisition, specifically the task of learning to follow navigation instructions from observation. Unlike conventional reranking used in syntactic and semantic parsing, gold-standard reference trees are not naturally available in a grounded setting. Instead, we show how the weak supervision of response feedback (e.g. successful task completion) can be used as an alternative, experimentally demonstrating that its performance is comparable to training on gold-standard parse trees.
6 0.12664086 138 acl-2013-Enriching Entity Translation Discovery using Selective Temporality
7 0.095986888 311 acl-2013-Semantic Neighborhoods as Hypergraphs
8 0.094113976 228 acl-2013-Leveraging Domain-Independent Information in Semantic Parsing
9 0.091774911 358 acl-2013-Transition-based Dependency Parsing with Selectional Branching
10 0.077508479 313 acl-2013-Semantic Parsing with Combinatory Categorial Grammars
11 0.075899936 226 acl-2013-Learning to Prune: Context-Sensitive Pruning for Syntactic MT
12 0.073562711 296 acl-2013-Recognizing Identical Events with Graph Kernels
13 0.073482446 132 acl-2013-Easy-First POS Tagging and Dependency Parsing with Beam Search
14 0.069542319 301 acl-2013-Resolving Entity Morphs in Censored Data
15 0.067299157 348 acl-2013-The effect of non-tightness on Bayesian estimation of PCFGs
16 0.067013793 343 acl-2013-The Effect of Higher-Order Dependency Features in Discriminative Phrase-Structure Parsing
17 0.065694995 129 acl-2013-Domain-Independent Abstract Generation for Focused Meeting Summarization
18 0.065399073 46 acl-2013-An Infinite Hierarchical Bayesian Model of Phrasal Translation
19 0.065207228 275 acl-2013-Parsing with Compositional Vector Grammars
20 0.063079864 314 acl-2013-Semantic Roles for String to Tree Machine Translation
topicId topicWeight
[(0, 0.197), (1, -0.025), (2, -0.084), (3, -0.039), (4, -0.04), (5, 0.121), (6, 0.091), (7, 0.035), (8, 0.006), (9, 0.035), (10, -0.036), (11, -0.005), (12, 0.015), (13, -0.053), (14, 0.015), (15, -0.081), (16, 0.01), (17, 0.021), (18, -0.018), (19, -0.05), (20, 0.036), (21, -0.187), (22, 0.09), (23, 0.11), (24, -0.078), (25, -0.015), (26, -0.111), (27, -0.003), (28, -0.226), (29, -0.005), (30, 0.048), (31, 0.004), (32, 0.16), (33, -0.141), (34, 0.202), (35, -0.135), (36, -0.064), (37, -0.059), (38, 0.004), (39, 0.04), (40, -0.173), (41, -0.015), (42, 0.052), (43, 0.008), (44, 0.06), (45, 0.113), (46, 0.079), (47, -0.016), (48, 0.006), (49, -0.086)]
simIndex simValue paperId paperTitle
same-paper 1 0.96158653 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.
2 0.86490458 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.
3 0.73426658 153 acl-2013-Extracting Events with Informal Temporal References in Personal Histories in Online Communities
Author: Miaomiao Wen ; Zeyu Zheng ; Hyeju Jang ; Guang Xiang ; Carolyn Penstein Rose
Abstract: We present a system for extracting the dates of illness events (year and month of the event occurrence) from posting histories in the context of an online medical support community. A temporal tagger retrieves and normalizes dates mentioned informally in social media to actual month and year referents. Building on this, an event date extraction system learns to integrate the likelihood of candidate dates extracted from time-rich sentences with temporal constraints extracted from eventrelated sentences. Our integrated model achieves 89.7% of the maximum performance given the performance of the temporal expression retrieval step.
4 0.56230915 311 acl-2013-Semantic Neighborhoods as Hypergraphs
Author: Chris Quirk ; Pallavi Choudhury
Abstract: Ambiguity preserving representations such as lattices are very useful in a number of NLP tasks, including paraphrase generation, paraphrase recognition, and machine translation evaluation. Lattices compactly represent lexical variation, but word order variation leads to a combinatorial explosion of states. We advocate hypergraphs as compact representations for sets of utterances describing the same event or object. We present a method to construct hypergraphs from sets of utterances, and evaluate this method on a simple recognition task. Given a set of utterances that describe a single object or event, we construct such a hypergraph, and demonstrate that it can recognize novel descriptions of the same event with high accuracy.
5 0.55542666 36 acl-2013-Adapting Discriminative Reranking to Grounded Language Learning
Author: Joohyun Kim ; Raymond Mooney
Abstract: We adapt discriminative reranking to improve the performance of grounded language acquisition, specifically the task of learning to follow navigation instructions from observation. Unlike conventional reranking used in syntactic and semantic parsing, gold-standard reference trees are not naturally available in a grounded setting. Instead, we show how the weak supervision of response feedback (e.g. successful task completion) can be used as an alternative, experimentally demonstrating that its performance is comparable to training on gold-standard parse trees.
6 0.55408883 138 acl-2013-Enriching Entity Translation Discovery using Selective Temporality
7 0.53130627 161 acl-2013-Fluid Construction Grammar for Historical and Evolutionary Linguistics
8 0.47279418 175 acl-2013-Grounded Language Learning from Video Described with Sentences
9 0.45903841 296 acl-2013-Recognizing Identical Events with Graph Kernels
10 0.44923472 313 acl-2013-Semantic Parsing with Combinatory Categorial Grammars
11 0.41396016 176 acl-2013-Grounded Unsupervised Semantic Parsing
12 0.4065493 165 acl-2013-General binarization for parsing and translation
13 0.40203428 224 acl-2013-Learning to Extract International Relations from Political Context
14 0.3946549 348 acl-2013-The effect of non-tightness on Bayesian estimation of PCFGs
15 0.39347327 364 acl-2013-Typesetting for Improved Readability using Lexical and Syntactic Information
16 0.38916567 61 acl-2013-Automatic Interpretation of the English Possessive
17 0.38359627 228 acl-2013-Leveraging Domain-Independent Information in Semantic Parsing
18 0.38130277 163 acl-2013-From Natural Language Specifications to Program Input Parsers
19 0.38098297 301 acl-2013-Resolving Entity Morphs in Censored Data
20 0.36205342 260 acl-2013-Nonconvex Global Optimization for Latent-Variable Models
topicId topicWeight
[(0, 0.042), (6, 0.053), (7, 0.011), (11, 0.084), (15, 0.017), (24, 0.047), (26, 0.066), (35, 0.09), (42, 0.053), (48, 0.034), (54, 0.188), (64, 0.017), (70, 0.077), (88, 0.017), (90, 0.034), (95, 0.079)]
simIndex simValue paperId paperTitle
same-paper 1 0.86388808 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.
2 0.8269577 69 acl-2013-Bilingual Lexical Cohesion Trigger Model for Document-Level Machine Translation
Author: Guosheng Ben ; Deyi Xiong ; Zhiyang Teng ; Yajuan Lu ; Qun Liu
Abstract: In this paper, we propose a bilingual lexical cohesion trigger model to capture lexical cohesion for document-level machine translation. We integrate the model into hierarchical phrase-based machine translation and achieve an absolute improvement of 0.85 BLEU points on average over the baseline on NIST Chinese-English test sets.
3 0.76985079 251 acl-2013-Mr. MIRA: Open-Source Large-Margin Structured Learning on MapReduce
Author: Vladimir Eidelman ; Ke Wu ; Ferhan Ture ; Philip Resnik ; Jimmy Lin
Abstract: We present an open-source framework for large-scale online structured learning. Developed with the flexibility to handle cost-augmented inference problems such as statistical machine translation (SMT), our large-margin learner can be used with any decoder. Integration with MapReduce using Hadoop streaming allows efficient scaling with increasing size of training data. Although designed with a focus on SMT, the decoder-agnostic design of our learner allows easy future extension to other structured learning problems such as sequence labeling and parsing.
4 0.71572948 169 acl-2013-Generating Synthetic Comparable Questions for News Articles
Author: Oleg Rokhlenko ; Idan Szpektor
Abstract: We introduce the novel task of automatically generating questions that are relevant to a text but do not appear in it. One motivating example of its application is for increasing user engagement around news articles by suggesting relevant comparable questions, such as “is Beyonce a better singer than Madonna?”, for the user to answer. We present the first algorithm for the task, which consists of: (a) offline construction of a comparable question template database; (b) ranking of relevant templates to a given article; and (c) instantiation of templates only with entities in the article whose comparison under the template’s relation makes sense. We tested the suggestions generated by our algorithm via a Mechanical Turk experiment, which showed a significant improvement over the strongest baseline of more than 45% in all metrics.
5 0.71212542 155 acl-2013-Fast and Accurate Shift-Reduce Constituent Parsing
Author: Muhua Zhu ; Yue Zhang ; Wenliang Chen ; Min Zhang ; Jingbo Zhu
Abstract: Shift-reduce dependency parsers give comparable accuracies to their chartbased counterparts, yet the best shiftreduce constituent parsers still lag behind the state-of-the-art. One important reason is the existence of unary nodes in phrase structure trees, which leads to different numbers of shift-reduce actions between different outputs for the same input. This turns out to have a large empirical impact on the framework of global training and beam search. We propose a simple yet effective extension to the shift-reduce process, which eliminates size differences between action sequences in beam-search. Our parser gives comparable accuracies to the state-of-the-art chart parsers. With linear run-time complexity, our parser is over an order of magnitude faster than the fastest chart parser.
6 0.71083051 132 acl-2013-Easy-First POS Tagging and Dependency Parsing with Beam Search
7 0.7064743 272 acl-2013-Paraphrase-Driven Learning for Open Question Answering
8 0.70610511 346 acl-2013-The Impact of Topic Bias on Quality Flaw Prediction in Wikipedia
9 0.70567423 159 acl-2013-Filling Knowledge Base Gaps for Distant Supervision of Relation Extraction
10 0.70540625 318 acl-2013-Sentiment Relevance
11 0.70411092 134 acl-2013-Embedding Semantic Similarity in Tree Kernels for Domain Adaptation of Relation Extraction
12 0.70368183 333 acl-2013-Summarization Through Submodularity and Dispersion
13 0.70298427 80 acl-2013-Chinese Parsing Exploiting Characters
14 0.70287585 343 acl-2013-The Effect of Higher-Order Dependency Features in Discriminative Phrase-Structure Parsing
15 0.69988966 18 acl-2013-A Sentence Compression Based Framework to Query-Focused Multi-Document Summarization
16 0.69866848 215 acl-2013-Large-scale Semantic Parsing via Schema Matching and Lexicon Extension
18 0.6973325 265 acl-2013-Outsourcing FrameNet to the Crowd
19 0.69720423 46 acl-2013-An Infinite Hierarchical Bayesian Model of Phrasal Translation
20 0.69688207 83 acl-2013-Collective Annotation of Linguistic Resources: Basic Principles and a Formal Model