emnlp emnlp2011 emnlp2011-15 knowledge-graph by maker-knowledge-mining

15 emnlp-2011-A novel dependency-to-string model for statistical machine translation

Source: pdf

Author: Jun Xie ; Haitao Mi ; Qun Liu

Abstract: Dependency structure, as a first step towards semantics, is believed to be helpful to improve translation quality. However, previous works on dependency structure based models typically resort to insertion operations to complete translations, which make it difficult to specify ordering information in translation rules. In our model of this paper, we handle this problem by directly specifying the ordering information in head-dependents rules which represent the source side as head-dependents relations and the target side as strings. The head-dependents rules require only substitution operation, thus our model requires no heuristics or separate ordering models of the previous works to control the word order of translations. Large-scale experiments show that our model performs well on long distance reordering, and outperforms the state- of-the-art constituency-to-string model (+1.47 BLEU on average) and hierarchical phrasebased model (+0.46 BLEU on average) on two Chinese-English NIST test sets without resort to phrases or parse forest. For the first time, a source dependency structure based model catches up with and surpasses the state-of-theart translation models.

Reference: text

Summary: the most important sentenses genereted by tfidf model

sentIndex sentText sentNum sentScore

1 However, previous works on dependency structure based models typically resort to insertion operations to complete translations, which make it difficult to specify ordering information in translation rules. [sent-6, score-0.654]

2 In our model of this paper, we handle this problem by directly specifying the ordering information in head-dependents rules which represent the source side as head-dependents relations and the target side as strings. [sent-7, score-0.671]

3 The head-dependents rules require only substitution operation, thus our model requires no heuristics or separate ordering models of the previous works to control the word order of translations. [sent-8, score-0.307]

4 For the first time, a source dependency structure based model catches up with and surpasses the state-of-theart translation models. [sent-12, score-0.54]

5 Those attractive characteristics make it pos216 sible to improve translation quality by using dependency structures. [sent-15, score-0.347]

6 Some researchers pay more attention to use dependency structure on the target side. [sent-16, score-0.397]

7 , 2008) presents a string-to-dependency model, which restricts the target side of each hierarchical rule to be a well-formed dependency tree fragment, and em- ploys a dependency language model to make the output more grammatically. [sent-18, score-0.911]

8 Using dependency structure on the source side is also a promising way, as tree-based systems run much faster (linear time vs. [sent-22, score-0.494]

9 Conventional dependency structure based models (Lin, 2004; Quirk et al. [sent-25, score-0.305]

10 , 2007) typically employ both substitution and insertion operation to complete translations, which make it difficult to specify ordering information directly in the translation rules. [sent-27, score-0.36]

11 In this paper, we handle this problem by directly specifying the ordering information in headdependents rules that represent the source side as head-dependents relations and the target side as string. [sent-31, score-0.834]

12 For the first time, a source dependency tree based model catches up with and surpasses the stateof-the-art translation models. [sent-41, score-0.474]

13 1 Dependency Sturcture A dependency structure for a sentence is a directed acyclic graph with words as nodes and modification relations as edges. [sent-43, score-0.541]

14 Figure 1 (a) shows an example dependency structure of a Chinese sentence. [sent-45, score-0.305]

15 For convenience, we use the lexicon dependency grammar (Hellwig, 2006) which adopts a bracket representation to express a projective dependency structure. [sent-47, score-0.525]

16 The dependency structure of Figure 1 (a) can be expressed as: ((2010年) (FIFA) 世界杯) (在(南非)) (成功) 举行 where the lexicon in brackets represents the depen- dents, while the lexicon out the brackets is the head. [sent-48, score-0.305]

17 To construct the dependency structure of a sentence, the most important thing is to establish dependency relations and distinguish the head from the dependent. [sent-49, score-0.811]

18 ssuucccceessssffuull yly Figure 1: Examples of dependency structure (a), headdependents relation (b), head-dependents rule (r1 of Figure 2) and head rule (d). [sent-97, score-1.163]

19 Hudson, 1990) for identifying a syntactic relation between a head and a dependent between a headdependent pair: 1. [sent-101, score-0.312]

20 2 Head-Dependents Relation A head-dependents relation is composed of a head and all its dependents as shown in Figure 1(b). [sent-105, score-0.309]

21 Since all the head-dependent pairs satisfy criteria 1 and 2, we can deduce that a head-dependents relation L holds the property that the head deter- mines the syntactic and semantic categories of L, and can often replace L. [sent-106, score-0.347]

22 Therefore, we can recursively replace the bottom level head-dependent relations of a dependency structure with their heads until the root. [sent-107, score-0.411]

23 This implies an representation of the generation of a dependency structure on the basis of head-dependents relation. [sent-108, score-0.305]

24 The former is an example of head-dependent rules that represent the source side as head-dependents relations and act as both translation rules and reordering rules. [sent-111, score-0.639]

25 The latter is an example of head rules which are used for translating words. [sent-112, score-0.352]

26 Formally, a dependency-to-string grammar is defined as a tuple ⟨Σ, N, ∆, R⟩, where Σ is a set of source language terminals, RN⟩, ,i sw a esreet oΣf categories for the terminals in Σ , ∆ is a set of target language terminals, and R is a set of translation rules. [sent-113, score-0.331]

27 A rule r in R is a tuple ⟨t, s, ϕ⟩, where: - t is a node labeled by terminal from Σ; or a head-dependents relation of the source dependency structures, with each node labeled by a terminal from Σ or a variable from a set X = {x1, x2 , . [sent-114, score-0.75]

28 For example, the head-dependents rule shown in Figure 1 (c) can be formalized as: t = ((x1:世界杯) (x2:在) (x3:AD) 举行) s = x1 was held x3 ϕ = {x1:世界杯 ↔ x2 , x2:在 ↔ x1 x2, x3:AD ↔ x3} where the underline indicates a leaf node, and xi:letters indicates a pair of variable and its constraint. [sent-118, score-0.379]

29 A derivation is informally defined as a sequence of steps converting a source dependency structure into a target language string, with each step apply- ing one translation rule. [sent-119, score-0.629]

30 / NNRR (f) 22001100 F FIIFFAA WWoorrldld CCuupp wwaass hh eeldld s suucccceessssffuull yly i nn ? [sent-190, score-0.371]

31 South Africa (g) 22001100 F FIIFFAA WWoorrldld CCuupp wwaass hh eeldld s suucccceessssffuull yly i nn [ SSoouutthh A Affrricicaa]] Figure 2: An example derivation of dependency-to-string translation. [sent-195, score-0.411]

32 CH 2010年 FIFA 世界杯在南非成功举行 EN 2010 FIFA World Cup was held successfully in South Africa The Chinese sentence (a) is first parsed into a dependency structure (b), which is converted into an English string in five steps. [sent-198, score-0.475]

33 First, at the root node, we apply head-dependents rule r1 shown in Figure 1(c) to translate the top level head-dependents relation and result in three unfinished substructures and target string in (c). [sent-199, score-0.38]

34 Second, we use head rule r2 translating “成功” into “successfully” and reach situation (d). [sent-201, score-0.42]

35 Third, we apply headdependents rule r3 translating the head-dependents relation rooted at “世界杯” and yield (e). [sent-202, score-0.52]

36 Finally, we apply head rule r5 translating the residual node “南非” and obtain the final translation in (g). [sent-204, score-0.628]

37 4 Rule Acquisition The rule acquisition begins with a word-aligned cor- pus: a set of triples ⟨T, S, A⟩, where T is a source dependency structure, ,SS ,isA a target si Tde i sentence, and A is an alignment relation between T and S. [sent-205, score-0.688]

38 We extract from each triple ⟨T, S, A⟩ head rules that are ceoxntrsaicstte frnot mw eitahc thh ter wpleor ⟨dT alignments da rnudl hse tahda-t dependents rules that satisfy the intuition that syntactically close items tend to stay close across languages. [sent-206, score-0.456]

39 We accomplish the rule acquisition through three steps: tree annotation, head-dependents fragments identification and rule induction. [sent-207, score-0.435]

40 1 Tree Annotation Given a triple ⟨T, S, A⟩ as shown in Figure 3, we fGirivste annn ao ttraiptel eea ⟨cTh, nSo,Ade⟩ n so fs hTo wwnith in ntw Foig aurtterib 3,ut ewse: head span and dependency span, which are defined as follows. [sent-209, score-0.531]

41 Given a node n, its head span hsp(n) is a set of index of the target words aligned to n. [sent-211, score-0.484]

42 A head span hsp(n) is consistent if it satisfies the following property: ∀n′̸=nhsp(n′) ∩ hsp(n) = ∅. [sent-214, score-0.292]

43 Each node is annotated with two spans, the former is head span and the latter dependency span. [sent-216, score-0.631]

44 The nodes in acceptable head set are displayed in gray, and the nodes in acceptable dependent set are denoted by boxes. [sent-217, score-1.015]

45 Given a head span hsp(n), its closure cloz(hsp(n)) is the smallest contiguous head span that is a superset of hsp(n). [sent-221, score-0.584]

46 Given a subtree rooted at n, the dependency span dsp(n) of n is defined as: T′ ∪ hsp(n′)). [sent-225, score-0.481]

47 hsp(n′)n i′∪s∈ cTo′nsistent If the head spans of all the nodes of T′ is not consisdsp(n) = cloz( tent, dsp(n) = ∅. [sent-226, score-0.377]

48 The extraction of head rules from each node can be readily achieved with the same criteria as (Och and Ney, 2004). [sent-229, score-0.407]

49 2 Head-Dependents Fragments Identification We then identify the head-dependents fragments that are suitable for rule induction from the annotated dependency structure. [sent-232, score-0.458]

50 To facilitate the identification process, we first define two sets of dependency structure related to head spans and dependency spans. [sent-233, score-0.804]

51 A acceptable head set ahs(T) of a dependency structure T is a set of nodes, each of which has a consistent head span. [sent-235, score-0.978]

52 For example, the elements of the acceptable head set of the dependency structure in Figure 3 are displayed in gray. [sent-236, score-0.773]

53 A acceptable dependent set adt(T) of a dependency structure T is a set of nodes, each of which satisfies: dep(n) ∅. [sent-238, score-0.618]

54 For example, the elements of the acceptable dependent set of the dependency structure in Figure 3 are denoted by boxes. [sent-239, score-0.618]

55 We say a head-dependents fragments is acceptable if it satisfies the following properties: = 1. [sent-241, score-0.312]

56 An acceptable head-dependents fragment holds the property that the head span ofthe root and the dependency spans ofthe sinks do not overlap with each other, which enables us to determine the reordering in the target side. [sent-244, score-1.234]

57 The identification of acceptable head-dependents fragments can be achieved by a single preorder transversal of the annotated dependency structure. [sent-245, score-0.597]

58 For each accessed internal node n, we check whether the head-dependents fragment f rooted at n is acceptable. [sent-246, score-0.367]

59 Typically, each acceptable head-dependents fragment has three types of nodes: internal nodes, internal nodes of the dependency structure; leaf nodes, leaf nodes of the dependency structure; head node, a special internal node acting as the head of the related head-dependents relation. [sent-248, score-2.023]

60 x1 held successfully x2 Figure 4: A lexicalized head-dependents rule (b) induced from the only acceptable head-dependents fragment (a) of Figure 3. [sent-272, score-0.756]

61 3 Rule Induction From each acceptable head-dependents fragment, we induce a set of lexicalized and unlexicalized head-dependents rules. [sent-274, score-0.448]

62 1 Lexicalized Rule We induce a lexicalized head-dependents rule from an acceptable head-dependents fragment by the following procedure: 1. [sent-277, score-0.647]

63 extract the head-dependents relation and mark the internal nodes as substitution sites. [sent-278, score-0.334]

64 place the nodes in order according to the head span of the root and the dependency spans of the sinks, then replace the internal nodes with variables and the other nodes with the target words covered by their head spans. [sent-280, score-1.354]

65 Figure 4 shows an acceptable head-dependents fragment and a lexicalized head-dependents rule induced from it. [sent-282, score-0.647]

66 To alleviate this problem, we generalize the lexicalized headdependents rules and induce rules with unlexicalized nodes. [sent-286, score-0.617]

67 As we know, the modification relation of a headdependents relation is determined by the edges. [sent-287, score-0.334]

68 POS) and obtain new head-dependents relations with unlexicalized nodes holding the same modification relation. [sent-290, score-0.313]

69 Here we call the lexicalized and unlexicalized head-dependents relations as instances of the modification relation. [sent-291, score-0.304]

70 Based on this observation, from each lexicalized head-dependent rule, we generate new headdependents rules with unlexicalized nodes according to the following principles: 1. [sent-295, score-0.567]

71 change the aligned part of the target string into a new variable when turning a head node or a leaf node into its category; 2. [sent-296, score-0.66]

72 keep the target side unchanged when turning a internal node into its category. [sent-297, score-0.373]

73 3 Unaligned Words We handle the unaligned words of the target side by extending the head spans of the lexicalized head and leaf nodes on both left and right directions. [sent-350, score-1.041]

74 Let d be a derivation that convert a source dependency structure T into a target string e. [sent-358, score-0.582]

76 It finds the best derivation d∗ that convert the input dependency structure into a target string among all possible derivations D: d∗ = argmaxd∈DP(D) (2) Given a source dependency structure T, the decoder transverses T in post-order. [sent-362, score-0.887]

77 For each accessed internal node n, it enumerates all instances of the related modification relation of the head-dependents relation rooted at n, and checks the rule set for matched translation rules. [sent-363, score-0.71]

78 For example, suppose that we can not find any translation rule about to “(2010年) (FIFA) 界杯”, we will construct a pseudo translation rule “(x1:2010年) (x2:FIFA) x3:世界杯 → x1 x2 x3”. [sent-365, score-0.556]

79 A larger translation is generated by substituting the variables in the target side of a translation rule with the translations of the corresponding dependents. [sent-366, score-0.653]

80 We parse the source sentences with Stanford Parser (Klein and Manning, 2003) into projective dependency structures, whose nodes are annotated by POS tags and edges by typed dependencies. [sent-378, score-0.44]

81 Figure 6 shows two translations of our test sets MT04 and MT05, which are selected because each holds a long distance dependency commonly used in Chinese. [sent-413, score-0.463]

82 1373(2001) and help developing countries strength counter-terrorism capability building for the efforts, Hiero-re: China appreciates Anti - Terrorism Committee to promote countries implement resolution No . [sent-437, score-0.388]

83 In the second example, the Chinese input holds a long distance dependency “中国赞赏 . [sent-481, score-0.393]

84 Cons2str and hiero-re fail to capture this long distance dependency and provide monotonic translations which do not reflect the meaning of the source sentence. [sent-486, score-0.506]

85 In contrast, dep2str successfully captures this long distance dependency and translates it into “China appreciates efforts of 224 . [sent-487, score-0.607]

86 And the incapability of cons2str and hiero-re in handling long distance reordering of these sentences does not lie in the representation of translation rules but the compromises in rule extraction or decoding so as to bal- ance the speed or grammar size and performance. [sent-499, score-0.616]

87 Both strategies limit the ability of the models in processing long distance reordering of sentences with long and complex modification relations. [sent-504, score-0.304]

88 Both models require projection of the source dependency structure to the target side via word alignment, and thus can not handle non-isomorphism between languages. [sent-513, score-0.586]

89 , 2007) presents a dependency treelet string correspondence model which directly map a dependency structure to a target string. [sent-515, score-0.749]

90 (Ding and Palmer, 2005) presents a translation model based on Synchronous Dependency Insertion Grammar(SDIG), which handles some of the non-isomorphism but requires both source and target dependency structures. [sent-516, score-0.523]

91 Most important, all these works do not specify the ordering information directly in translation rules, and resort to either heuristics (Lin, 2004; Xiong et al. [sent-517, score-0.302]

92 , 2008) exploits target dependency structures as dependency language models to ensure the grammaticality of the target string. [sent-522, score-0.706]

93 , 225 2008) extends the hierarchical phrase-based model and present a string-to-dependency model, which employs string-to-dependency rules whose source side are string and the target as well-formed dependency structures. [sent-524, score-0.749]

94 In contrast, our model exploits source dependency structures, as a tree-based system, it run much faster (linear time vs. [sent-525, score-0.323]

95 9 Conclusions and future work In this paper, we present a novel dependency-tostring model, which employs head-dependents rules that represent the source side as head-dependents relations and the target side as string. [sent-528, score-0.55]

96 The headdependents rules specify the ordering information directly and require only substitution operation. [sent-529, score-0.47]

97 Large scale experiments show that our model exhibits good performance in long distance reordering and outperforms the state-ofthe-art constituency-to-string model and hierarchical phrase-based model without resort to phrases and parse forest. [sent-531, score-0.328]

98 In our future works, we will exploit the semantic information encoded in the dependency structures which is expected to further improve the translations, and replace 1-best dependency structures with dependency forests so as to alleviate the influence caused by parse errors. [sent-533, score-0.914]

99 A new string-to-dependency machine translation al- gorithm with a target dependency language model. [sent-622, score-0.439]

100 A dependency treelet string correspondence model for statistical machine translation. [sent-631, score-0.352]

similar papers computed by tfidf model

tfidf for this paper:

wordName wordTfidf (topN-words)

[('hsp', 0.325), ('acceptable', 0.263), ('dependency', 0.239), ('head', 0.205), ('rule', 0.17), ('headdependents', 0.163), ('fifa', 0.16), ('appreciates', 0.139), ('ad', 0.137), ('ordering', 0.121), ('nodes', 0.117), ('barnier', 0.116), ('xiong', 0.11), ('translation', 0.108), ('lexicalized', 0.108), ('fragment', 0.106), ('side', 0.105), ('rules', 0.102), ('leaf', 0.102), ('node', 0.1), ('africa', 0.093), ('dsp', 0.093), ('wwaass', 0.093), ('yly', 0.093), ('target', 0.092), ('powell', 0.091), ('span', 0.087), ('rooted', 0.085), ('countries', 0.084), ('substitution', 0.084), ('source', 0.084), ('quirk', 0.082), ('ding', 0.078), ('unlexicalized', 0.077), ('reordering', 0.076), ('internal', 0.076), ('resort', 0.073), ('committee', 0.073), ('subtree', 0.07), ('translations', 0.07), ('cloz', 0.07), ('eeldld', 0.07), ('fiiffaa', 0.07), ('sinks', 0.07), ('suucccceessssffuull', 0.07), ('china', 0.069), ('efforts', 0.067), ('hierarchical', 0.066), ('structure', 0.066), ('alleviate', 0.065), ('bleu', 0.064), ('relations', 0.062), ('string', 0.061), ('held', 0.06), ('said', 0.06), ('anti', 0.06), ('nnrr', 0.06), ('och', 0.059), ('long', 0.058), ('modification', 0.057), ('relation', 0.057), ('spans', 0.055), ('distance', 0.055), ('cubic', 0.054), ('unaligned', 0.052), ('ld', 0.052), ('treelet', 0.052), ('cup', 0.05), ('nr', 0.05), ('dependent', 0.05), ('fragments', 0.049), ('successfully', 0.049), ('chiang', 0.048), ('underline', 0.047), ('dependents', 0.047), ('vv', 0.047), ('grammar', 0.047), ('huang', 0.047), ('insertion', 0.047), ('nist', 0.047), ('palmer', 0.047), ('ccuupp', 0.046), ('talks', 0.046), ('terrorism', 0.046), ('transversal', 0.046), ('wwoorrldld', 0.046), ('acquisition', 0.046), ('translating', 0.045), ('chinese', 0.045), ('south', 0.045), ('hh', 0.045), ('structures', 0.044), ('replace', 0.044), ('dd', 0.043), ('surpasses', 0.043), ('holds', 0.041), ('promote', 0.041), ('derivation', 0.04), ('developing', 0.04), ('aug', 0.04)]

similar papers list:

simIndex simValue paperId paperTitle

same-paper 1 1.0 15 emnlp-2011-A novel dependency-to-string model for statistical machine translation

Author: Jun Xie ; Haitao Mi ; Qun Liu

2 0.2403426 123 emnlp-2011-Soft Dependency Constraints for Reordering in Hierarchical Phrase-Based Translation

Author: Yang Gao ; Philipp Koehn ; Alexandra Birch

Abstract: Long-distance reordering remains one of the biggest challenges facing machine translation. We derive soft constraints from the source dependency parsing to directly address the reordering problem for the hierarchical phrasebased model. Our approach significantly improves Chinese–English machine translation on a large-scale task by 0.84 BLEU points on average. Moreover, when we switch the tuning function from BLEU to the LRscore which promotes reordering, we observe total improvements of 1.21 BLEU, 1.30 LRscore and 3.36 TER over the baseline. On average our approach improves reordering precision and recall by 6.9 and 0.3 absolute points, respectively, and is found to be especially effective for long-distance reodering.

3 0.22174098 20 emnlp-2011-Augmenting String-to-Tree Translation Models with Fuzzy Use of Source-side Syntax

Author: Jiajun Zhang ; Feifei Zhai ; Chengqing Zong

Abstract: Due to its explicit modeling of the grammaticality of the output via target-side syntax, the string-to-tree model has been shown to be one of the most successful syntax-based translation models. However, a major limitation of this model is that it does not utilize any useful syntactic information on the source side. In this paper, we analyze the difficulties of incorporating source syntax in a string-totree model. We then propose a new way to use the source syntax in a fuzzy manner, both in source syntactic annotation and in rule matching. We further explore three algorithms in rule matching: 0-1 matching, likelihood matching, and deep similarity matching. Our method not only guarantees grammatical output with an explicit target tree, but also enables the system to choose the proper translation rules via fuzzy use of the source syntax. Our extensive experiments have shown significant improvements over the state-of-the-art string-to-tree system. 1

4 0.21592066 108 emnlp-2011-Quasi-Synchronous Phrase Dependency Grammars for Machine Translation

Author: Kevin Gimpel ; Noah A. Smith

Abstract: We present a quasi-synchronous dependency grammar (Smith and Eisner, 2006) for machine translation in which the leaves of the tree are phrases rather than words as in previous work (Gimpel and Smith, 2009). This formulation allows us to combine structural components of phrase-based and syntax-based MT in a single model. We describe a method of extracting phrase dependencies from parallel text using a target-side dependency parser. For decoding, we describe a coarse-to-fine approach based on lattice dependency parsing of phrase lattices. We demonstrate performance improvements for Chinese-English and UrduEnglish translation over a phrase-based baseline. We also investigate the use of unsupervised dependency parsers, reporting encouraging preliminary results.

5 0.17263833 58 emnlp-2011-Fast Generation of Translation Forest for Large-Scale SMT Discriminative Training

Author: Xinyan Xiao ; Yang Liu ; Qun Liu ; Shouxun Lin

Abstract: Although discriminative training guarantees to improve statistical machine translation by incorporating a large amount of overlapping features, it is hard to scale up to large data due to decoding complexity. We propose a new algorithm to generate translation forest of training data in linear time with the help of word alignment. Our algorithm also alleviates the oracle selection problem by ensuring that a forest always contains derivations that exactly yield the reference translation. With millions of features trained on 519K sentences in 0.03 second per sentence, our system achieves significant improvement by 0.84 BLEU over the baseline system on the NIST Chinese-English test sets.

6 0.14789766 10 emnlp-2011-A Probabilistic Forest-to-String Model for Language Generation from Typed Lambda Calculus Expressions

7 0.14337239 50 emnlp-2011-Evaluating Dependency Parsing: Robust and Heuristics-Free Cross-Annotation Evaluation

8 0.12793621 125 emnlp-2011-Statistical Machine Translation with Local Language Models

9 0.11876018 13 emnlp-2011-A Word Reordering Model for Improved Machine Translation

10 0.11569548 136 emnlp-2011-Training a Parser for Machine Translation Reordering

11 0.11181772 75 emnlp-2011-Joint Models for Chinese POS Tagging and Dependency Parsing

12 0.10830866 74 emnlp-2011-Inducing Sentence Structure from Parallel Corpora for Reordering

13 0.10615042 83 emnlp-2011-Learning Sentential Paraphrases from Bilingual Parallel Corpora for Text-to-Text Generation

14 0.10300668 4 emnlp-2011-A Fast, Accurate, Non-Projective, Semantically-Enriched Parser

15 0.10167141 65 emnlp-2011-Heuristic Search for Non-Bottom-Up Tree Structure Prediction

16 0.098850608 16 emnlp-2011-Accurate Parsing with Compact Tree-Substitution Grammars: Double-DOP

17 0.097477555 2 emnlp-2011-A Cascaded Classification Approach to Semantic Head Recognition

18 0.097261265 22 emnlp-2011-Better Evaluation Metrics Lead to Better Machine Translation

19 0.093858831 44 emnlp-2011-Domain Adaptation via Pseudo In-Domain Data Selection

20 0.090369016 85 emnlp-2011-Learning to Simplify Sentences with Quasi-Synchronous Grammar and Integer Programming

similar papers computed by lsi model

lsi for this paper:

topicId topicWeight

[(0, 0.335), (1, 0.232), (2, 0.048), (3, -0.012), (4, -0.036), (5, -0.086), (6, 0.079), (7, 0.025), (8, 0.149), (9, -0.016), (10, 0.06), (11, 0.168), (12, 0.036), (13, 0.009), (14, 0.077), (15, -0.076), (16, -0.022), (17, 0.059), (18, -0.024), (19, 0.067), (20, 0.09), (21, 0.059), (22, -0.17), (23, 0.126), (24, -0.151), (25, 0.054), (26, 0.028), (27, 0.113), (28, -0.119), (29, -0.028), (30, 0.057), (31, 0.012), (32, -0.046), (33, -0.083), (34, 0.089), (35, -0.02), (36, -0.011), (37, 0.057), (38, 0.035), (39, 0.013), (40, -0.106), (41, 0.042), (42, -0.039), (43, -0.138), (44, -0.057), (45, -0.081), (46, 0.065), (47, 0.053), (48, -0.055), (49, 0.055)]

similar papers list:

simIndex simValue paperId paperTitle

same-paper 1 0.96901488 15 emnlp-2011-A novel dependency-to-string model for statistical machine translation

Author: Jun Xie ; Haitao Mi ; Qun Liu

2 0.78762788 20 emnlp-2011-Augmenting String-to-Tree Translation Models with Fuzzy Use of Source-side Syntax

Author: Jiajun Zhang ; Feifei Zhai ; Chengqing Zong

3 0.77596575 123 emnlp-2011-Soft Dependency Constraints for Reordering in Hierarchical Phrase-Based Translation

Author: Yang Gao ; Philipp Koehn ; Alexandra Birch

4 0.69689262 108 emnlp-2011-Quasi-Synchronous Phrase Dependency Grammars for Machine Translation

Author: Kevin Gimpel ; Noah A. Smith

5 0.66179556 47 emnlp-2011-Efficient retrieval of tree translation examples for Syntax-Based Machine Translation

Author: Fabien Cromieres ; Sadao Kurohashi

Abstract: We propose an algorithm allowing to efficiently retrieve example treelets in a parsed tree database in order to allow on-the-fly extraction of syntactic translation rules. We also propose improvements of this algorithm allowing several kinds of flexible matchings.

6 0.64582598 58 emnlp-2011-Fast Generation of Translation Forest for Large-Scale SMT Discriminative Training

7 0.59111422 85 emnlp-2011-Learning to Simplify Sentences with Quasi-Synchronous Grammar and Integer Programming

8 0.51049942 10 emnlp-2011-A Probabilistic Forest-to-String Model for Language Generation from Typed Lambda Calculus Expressions

9 0.49219957 50 emnlp-2011-Evaluating Dependency Parsing: Robust and Heuristics-Free Cross-Annotation Evaluation

10 0.47540653 2 emnlp-2011-A Cascaded Classification Approach to Semantic Head Recognition

11 0.45556787 65 emnlp-2011-Heuristic Search for Non-Bottom-Up Tree Structure Prediction

12 0.44017151 103 emnlp-2011-Parser Evaluation over Local and Non-Local Deep Dependencies in a Large Corpus

13 0.43272799 13 emnlp-2011-A Word Reordering Model for Improved Machine Translation

14 0.41886389 74 emnlp-2011-Inducing Sentence Structure from Parallel Corpora for Reordering

15 0.41154709 83 emnlp-2011-Learning Sentential Paraphrases from Bilingual Parallel Corpora for Text-to-Text Generation

16 0.41084343 16 emnlp-2011-Accurate Parsing with Compact Tree-Substitution Grammars: Double-DOP

17 0.39870274 4 emnlp-2011-A Fast, Accurate, Non-Projective, Semantically-Enriched Parser

18 0.38351986 125 emnlp-2011-Statistical Machine Translation with Local Language Models

19 0.36218917 44 emnlp-2011-Domain Adaptation via Pseudo In-Domain Data Selection

20 0.35844174 75 emnlp-2011-Joint Models for Chinese POS Tagging and Dependency Parsing

similar papers computed by lda model

lda for this paper:

topicId topicWeight

[(23, 0.102), (35, 0.014), (36, 0.029), (37, 0.039), (45, 0.066), (53, 0.034), (54, 0.04), (57, 0.017), (62, 0.03), (64, 0.029), (65, 0.293), (66, 0.045), (69, 0.02), (79, 0.057), (82, 0.022), (90, 0.034), (96, 0.034), (98, 0.033)]

similar papers list:

simIndex simValue paperId paperTitle

same-paper 1 0.79927105 15 emnlp-2011-A novel dependency-to-string model for statistical machine translation

Author: Jun Xie ; Haitao Mi ; Qun Liu

2 0.76625592 118 emnlp-2011-SMT Helps Bitext Dependency Parsing

Author: Wenliang Chen ; Jun'ichi Kazama ; Min Zhang ; Yoshimasa Tsuruoka ; Yujie Zhang ; Yiou Wang ; Kentaro Torisawa ; Haizhou Li

Abstract: We propose a method to improve the accuracy of parsing bilingual texts (bitexts) with the help of statistical machine translation (SMT) systems. Previous bitext parsing methods use human-annotated bilingual treebanks that are hard to obtain. Instead, our approach uses an auto-generated bilingual treebank to produce bilingual constraints. However, because the auto-generated bilingual treebank contains errors, the bilingual constraints are noisy. To overcome this problem, we use large-scale unannotated data to verify the constraints and design a set of effective bilingual features for parsing models based on the verified results. The experimental results show that our new parsers significantly outperform state-of-theart baselines. Moreover, our approach is still able to provide improvement when we use a larger monolingual treebank that results in a much stronger baseline. Especially notable is that our approach can be used in a purely monolingual setting with the help of SMT.

3 0.50326645 123 emnlp-2011-Soft Dependency Constraints for Reordering in Hierarchical Phrase-Based Translation

Author: Yang Gao ; Philipp Koehn ; Alexandra Birch

4 0.50060171 20 emnlp-2011-Augmenting String-to-Tree Translation Models with Fuzzy Use of Source-side Syntax

Author: Jiajun Zhang ; Feifei Zhai ; Chengqing Zong

5 0.48663497 25 emnlp-2011-Cache-based Document-level Statistical Machine Translation

Author: Zhengxian Gong ; Min Zhang ; Guodong Zhou

Abstract: Statistical machine translation systems are usually trained on a large amount of bilingual sentence pairs and translate one sentence at a time, ignoring document-level information. In this paper, we propose a cache-based approach to document-level translation. Since caches mainly depend on relevant data to supervise subsequent decisions, it is critical to fill the caches with highly-relevant data of a reasonable size. In this paper, we present three kinds of caches to store relevant document-level information: 1) a dynamic cache, which stores bilingual phrase pairs from the best translation hypotheses of previous sentences in the test document; 2) a static cache, which stores relevant bilingual phrase pairs extracted from similar bilingual document pairs (i.e. source documents similar to the test document and their corresponding target documents) in the training parallel corpus; 3) a topic cache, which stores the target-side topic words related with the test document in the source-side. In particular, three new features are designed to explore various kinds of document-level information in above three kinds of caches. Evaluation shows the effectiveness of our cache-based approach to document-level translation with the performance improvement of 0.8 1 in BLUE score over Moses. Especially, detailed analysis and discussion are presented to give new insights to document-level translation. 1

6 0.48068544 108 emnlp-2011-Quasi-Synchronous Phrase Dependency Grammars for Machine Translation

7 0.47746515 53 emnlp-2011-Experimental Support for a Categorical Compositional Distributional Model of Meaning

8 0.47590166 1 emnlp-2011-A Bayesian Mixture Model for PoS Induction Using Multiple Features

9 0.46977642 136 emnlp-2011-Training a Parser for Machine Translation Reordering

10 0.46951461 68 emnlp-2011-Hypotheses Selection Criteria in a Reranking Framework for Spoken Language Understanding

11 0.46850881 35 emnlp-2011-Correcting Semantic Collocation Errors with L1-induced Paraphrases

12 0.46797809 111 emnlp-2011-Reducing Grounded Learning Tasks To Grammatical Inference

13 0.4656156 66 emnlp-2011-Hierarchical Phrase-based Translation Representations

14 0.46551368 54 emnlp-2011-Exploiting Parse Structures for Native Language Identification

15 0.46507418 8 emnlp-2011-A Model of Discourse Predictions in Human Sentence Processing

16 0.46496186 85 emnlp-2011-Learning to Simplify Sentences with Quasi-Synchronous Grammar and Integer Programming

17 0.46482247 22 emnlp-2011-Better Evaluation Metrics Lead to Better Machine Translation

18 0.46319801 83 emnlp-2011-Learning Sentential Paraphrases from Bilingual Parallel Corpora for Text-to-Text Generation

19 0.46285146 97 emnlp-2011-Multiword Expression Identification with Tree Substitution Grammars: A Parsing tour de force with French

20 0.46119875 59 emnlp-2011-Fast and Robust Joint Models for Biomedical Event Extraction