emnlp emnlp2010 emnlp2010-68 knowledge-graph by maker-knowledge-mining
Source: pdf
Author: Tao Zhuang ; Chengqing Zong
Abstract: We show that jointly performing semantic role labeling (SRL) on bitext can improve SRL results on both sides. In our approach, we use monolingual SRL systems to produce argument candidates for predicates in bitext at first. Then, we simultaneously generate SRL results for two sides of bitext using our joint inference model. Our model prefers the bilingual SRL result that is not only reasonable on each side of bitext, but also has more consistent argument structures between two sides. To evaluate the consistency between two argument structures, we also formulate a log-linear model to compute the probability of aligning two arguments. We have experimented with our model on Chinese-English parallel PropBank data. Using our joint inference model, F1 scores of SRL results on Chinese and English text achieve 79.53% and 77.87% respectively, which are 1.52 and 1.74 points higher than the results of baseline monolingual SRL combination systems respectively.
Reference: text
sentIndex sentText sentNum sentScore
1 cn a Abstract We show that jointly performing semantic role labeling (SRL) on bitext can improve SRL results on both sides. [sent-4, score-0.237]
2 In our approach, we use monolingual SRL systems to produce argument candidates for predicates in bitext at first. [sent-5, score-0.676]
3 Then, we simultaneously generate SRL results for two sides of bitext using our joint inference model. [sent-6, score-0.33]
4 Our model prefers the bilingual SRL result that is not only reasonable on each side of bitext, but also has more consistent argument structures between two sides. [sent-7, score-0.551]
5 To evaluate the consistency between two argument structures, we also formulate a log-linear model to compute the probability of aligning two arguments. [sent-8, score-0.411]
6 A conventional way to perform SRL on bitext is performing SRL on each side of bitext separately, as has been done by Fung et al. [sent-17, score-0.36]
7 However, it is very difficult to obtain good SRL results on both sides of bitext 304 in this way. [sent-19, score-0.219]
8 On the other hand, the semantic equivalence between two sides of bitext means that they should have consistent predicate-argument structures. [sent-28, score-0.27]
9 This bilingual argument structure consistency can guide us to find better SRL results. [sent-29, score-0.479]
10 For example, in Figure 1(a), the argument structure consistency can guide us to choose a correct SRL result on Chinese side. [sent-30, score-0.411]
11 Consistency between two argument structures is reflected by sound argument alignments between them, as shown in Figure 1(b). [sent-31, score-0.729]
12 In this paper, we show that the bilingual argument structure consistency can be leveraged to substantially improve SRL results on both sides of bitext. [sent-34, score-0.586]
13 Using automatic word alignment on bitext, we first identify a pair of predicates that align with each other. [sent-36, score-0.234]
14 And we use monolingual SRL systems to produce argument candidates for each predicate. [sent-37, score-0.503]
15 Then, our model jointly generate SRL results for both predicates from their argument candidates, using integer linear programming (ILP) technique. [sent-38, score-0.408]
16 But it is to more difficult to get correct SRL result on Chinese side, because the AM-TMP argument embeds into a discontinuous A1 argument. [sent-47, score-0.378]
17 The argument structure consistency can guide us to choose the correct Chinese SRL result. [sent-50, score-0.411]
18 These three components correspond to three interrelated factors: the quality ofthe SRL result on source side, the quality of the SRL result on target side, and the argument structure consistency between the SRL results on both sides. [sent-53, score-0.498]
19 To evaluate the consistency between the two argument structures in our joint inference model, we formulate a log-linear model to compute the probability of aligning two arguments. [sent-54, score-0.522]
20 (2007) did pioneering work on studying argument alignment on Chinese-English parallel PropBank. [sent-76, score-0.502]
21 Then, given the SRL result on both sides, they automatically induced the argument alignment between two sides. [sent-78, score-0.46]
22 The major difference between our work and all existing research is that our model performs SRL inference on two sides of bitext simultaneously. [sent-79, score-0.278]
23 In our model, we jointly consider three interrelated factors: SRL result on the source side, SRL result on the target side, and the argument alignment between them. [sent-80, score-0.547]
24 1 Monolingual SRL System As shown in Figure 2, we need to use a monolingual SRL system to generate candidates for our joint inference model. [sent-82, score-0.267]
25 In this system, the whole SRL process is comprised of three stages: pruning, argument identification, and argument classification. [sent-84, score-0.694]
26 In the argument identification stage, a number of argument locations are identified in a sentence. [sent-86, score-0.721]
27 In the argument classification stage, each location identified in the previous stage is assigned a semantic role label. [sent-87, score-0.448]
28 Maximum entropy classifier is employed for both the argument identification and classification tasks. [sent-88, score-0.347]
29 We use the classification probability of the argument classification stage as an argument’s probability. [sent-96, score-0.347]
30 As illustrated in Figure 3, in an individual system’s output, each argument has three attributes: its location in sentence loc, represented by the number of its first word and last word; its semantic role label l; and its probability p. [sent-97, score-0.472]
31 So each argument outputted by a system is a triple (loc, l,p). [sent-98, score-0.347]
32 For example, the A0 argument in Figure 3 is ((0, 2) , A0 , 0. [sent-99, score-0.347]
33 3 Generating and Merging Candidates To generate candidates for joint inference, we need to have multiple SRL results on each side of bitext. [sent-118, score-0.245]
34 4 Joint Inference Model Our model can be conceptually decomposed to three components: the source side, the target side, and the argument alignment. [sent-126, score-0.41]
35 1 Source Side Component The source side component aims to improve the SRL result on source side. [sent-131, score-0.2]
36 , lLss}, rino wehlaibchel l1s ∼ l6s ostuarncde faonrg tuhaeg key argument label}s, A0 ∼ A5 respectively. [sent-136, score-0.347]
37 Then the source side sub-objective Os in equation (1) is the sum of arguments’ probabilities on source side: XNs = XLs Os XX(pisj − Ts)xij (2) Xi=1 Xj=1 where Ts is a bias to prevent including too many candidates in solution (Surdeanu et al. [sent-144, score-0.257]
38 ∀1 ≤ j ≤ 6 :XNsxij≤ 1 (4) Xi=1 For any source side location locis, let Ci denote the index set of the locations that overlap with it. [sent-157, score-0.195]
39 2 Target Side Component In principle, the target side component of our joint inference model is the same with the source side component. [sent-168, score-0.446]
40 And l1t ∼ l6t stand for the key argument labels A0 ∼ A5 respectively. [sent-177, score-0.347]
41 2 Argument Alignment The argument alignment component is the core of our joint inference model. [sent-182, score-0.571]
42 It gives preference to the bilingual SRL results that have more consistent argument structures. [sent-183, score-0.415]
43 For a source side argument argis = (locis, ls) and a target side argument argkt = (loctk, lt), let zik be the following indicator variable: zik = [argis aligns with argkt] . [sent-184, score-1.788]
44 We use piak to represent the probability that argis and argkt align with each other, i. [sent-185, score-0.445]
45 We call piak the argument alignment probability between argis and argkt. [sent-188, score-0.737]
46 1 Argument Alignment Probability Model We use a log-linear model to compute the argument alignment probability piak between argis and argkt. [sent-191, score-0.737]
47 Our loglinear model defines a distribution on zik given the tuple tup = (argis , argkt, wa, s, t) : P(zik|tup) ∝ exp(wTφ(tup)) where φ(tup) is the feature vector. [sent-193, score-0.265]
48 With this model, piak can be computed as piak = P(zik = 1|tup). [sent-194, score-0.216]
49 On this data set, we get the argument alignment matrix in Table 1. [sent-199, score-0.46]
50 Each entry in Table 1 is the number of times for which one type of Chinese argument aligns with one type of English argument. [sent-201, score-0.419]
51 , and NUL , 308 means that the argument on the other side cannot be aligned with any argument on this side. [sent-203, score-0.874]
52 For example, the number 46 in the A0 row and NUL column means that Chinese A0 argument cannot be aligned with any argument on English side for 46 times in our manually aligned corpus. [sent-204, score-0.918]
53 Word alignment feature: If there are many wordto-word alignments between argis and argkt, then it is very probable that argis and argkt would align with each other. [sent-206, score-0.654]
54 We adopt the method used in (Pad o´ and Lapata, 2009) to measure the word-to-word alignments between argis and argkt. [sent-207, score-0.204]
55 Head word alignment feature: The head word of an argument is usually more representative than other words. [sent-210, score-0.46]
56 So we use whether the head words of argis and argkt align with each other as a binary feature. [sent-211, score-0.337]
57 Semantic role labels of two arguments: From Ta- ble 1, we can see that semantic role labels of two arguments are a good indicator of whether they should align with each other. [sent-213, score-0.272]
58 For example, a Chinese A0 argument aligns with an English A0 argument most of the times, and never aligns with an English AM* argument in Table 1. [sent-214, score-1.185]
59 Therefore, the semantic role labels of argis and argkt are used as a feature. [sent-215, score-0.378]
60 Predicate verb pair: Different predicate pairs have different argument alignment patterns. [sent-216, score-0.522]
61 The argument alignment matrix for all instances of the Chinese-English predicate pair (zengzhang, grow) in our manually aligned corpus is shown in Table 2. [sent-219, score-0.566]
62 From Table 2 we can see that all A0 arguments of zengzhang align with A1 arguments of grow. [sent-221, score-0.218]
63 This is very different from the results in Table 1, where a Chinese A0 argument tends to align with an English A0 argument. [sent-222, score-0.407]
64 Oa reflects the consistency between argument structures on two sides of bitext. [sent-230, score-0.518]
65 Larger Oa means better argument alignment between two sides, thus indicates more consistency between argument structures on two sides. [sent-231, score-0.871]
66 For all candidates on both source and target sides, only those that are chosen to be arguments on each side can be aligned. [sent-234, score-0.317]
67 Each argument can not be aligned with more than 3 arguments. [sent-237, score-0.391]
68 Each argument on source side must be aligned with at least one argument on target side, and vice versa. [sent-240, score-0.937]
69 The Conformity with bilingual SRL result constraint is necessary to validly integrate the bilingual SRL result with the argument alignment. [sent-241, score-0.561]
70 This constraint means that if argis and argkt align with each other, i. [sent-242, score-0.415]
71 , zik = 1, then loPcis must be assigned a label on source side, i. [sent-244, score-0.261]
72 So this constraint can be represented PjL=t1 aPs: ∀1 ≤ i ≤ Ns,1 ≤ k ≤ Nt:XLsxij≥ zik ∀1 ≤ k ≤ Nt,1 ≤ i ≤ Ns:XLtykj≥ zik (11) Xj=1 Xj=1 (12) 309 The One-to-many alignment limit constraint comes from our observation on manually aligned corpus. [sent-249, score-0.723]
73 We have found that no argument aligns with more than 3 arguments in our manually aligned corpus. [sent-250, score-0.524]
74 This constraint can be represented as: XNtzik ≤ 3 ∀1 ≤ k ≤ Nt:XNszik≤ 3 ∀1 ≤ i ≤ Ns : (13) kX= X1 (14) Xi= X1 The Complete argument alignment constraint comes from the semantic equivalence between two sides of bitext. [sent-251, score-0.774]
75 This can Pb e represented as: PNk=t1 XNt XLs ∀1 ≤ i≤ Ns : Xzik ≥ Xxij Xk=1 Xj=1 (15) Similarly, each target side argument must be aligned to at least one source side argument. [sent-257, score-0.726]
76 3 (16) jX= X1 Complete Argument Alignment as a Soft Constraint Although the hard Complement argument alignment constraint is ideally reasonable, in real situations this constraint does not always hold. [sent-259, score-0.616]
77 The manual argument alignment result shown in Table 1 indicates that in some cases an argument cannot be aligned with any argument on the other side (see the NUL row and column in Table 1). [sent-260, score-1.334]
78 Therefore, it would be reasonable to change the hard Complement argument alignment constraint to a soft one. [sent-261, score-0.564]
79 To do so, we need to remove the hard Complement argument alignment constraint and add penalty for violation of this constraint. [sent-262, score-0.538]
80 If an argument does not align with any argument on the other side, we say it aligns with NUL. [sent-263, score-0.826]
81 the nuPmber of target side argumentsP that align with NUL. [sent-267, score-0.227]
82 For each argument that aligns with NUL, we add a penalty λ3 to the argument alignment subobjective Oa. [sent-268, score-0.879]
83 Therefore, the sub-objective Oa in equation (10) is changed to: Oa=XNsXNt(piak− Ta)zik Xi= X1 kX= X1 XXNs XNt −λ3(Xzi,NUL+XzNUL,k) Xi=1 Xk=1 (17) From the definition of zi,NUL, it is obvious that, for any 1 ≤ i ≤ Ns, zi,NUL and zik(1 ≤ k ≤ Nt) have the following relationship: If zik ≥ 1, i. [sent-269, score-0.205]
84 , argis aligns with some argumentPs on target side, then zi,NUL = 0; Otherwise, zi,NUL = 1. [sent-271, score-0.272]
85 The first version treats Complement argument alignment as a hard constraint. [sent-274, score-0.46]
86 The sencond version treats Complement argument alignment as a soft constraint. [sent-278, score-0.486]
87 We will refer to the source side combination model as SrcCmb. [sent-283, score-0.197]
88 Similarly, we will refer to the target side combination model as TrgCmb. [sent-286, score-0.196]
89 Although our joint inference model needs no training, we still need to train a log-linear argument alignment probability model, which is used in the joint inference model. [sent-308, score-0.682]
90 ing set for the argument alignment probability model consists of 60 files (chtb 0121 . [sent-314, score-0.51]
91 A bilingual predicate pair is defined to be a pair of predicates in bitext which align with each other in automatic word alignment. [sent-321, score-0.363]
92 The first part of a discontinuous argument is labeled as it is, and the second part is labeled with a prefix “C-”. [sent-342, score-0.378]
93 4 Effects of Different Constraints The One-to-many limit and Complete argument alignment constraints in subsection 4. [sent-379, score-0.54]
94 In Table 6, ‘c2’ refers to the One-to-many limit constraint, ‘c3’ refers to the Complete argument alignment constraint, and ‘-’ means removing. [sent-384, score-0.46]
95 1837 Table 7: Comparison between monolingual combination model and our joint inference model on test set. [sent-418, score-0.239]
96 Our joint inference model incorporates not only linguistic constraints on source and target sides of bitext, but also the bilingual argument structure consistency requirement on bitext. [sent-430, score-0.801]
97 Experiments on Chinese-English parallel PropBank show that our joint inference model is very effective for bilingual SRL. [sent-431, score-0.221]
98 Compared to stateof-the-art monolingual SRL combination baselines, our joint inference model substantially improves SRL results on both sides of bitext. [sent-432, score-0.346]
99 In fact, the solution of our joint inference model contains not only the SRL results on bitext, but also the optimal argument alignment between two sides of bitext. [sent-433, score-0.678]
100 This makes our model especially suitable for application in machine translation, which needs to obtain the argument alignment. [sent-434, score-0.347]
wordName wordTfidf (topN-words)
[('srl', 0.666), ('argument', 0.347), ('chtb', 0.217), ('zik', 0.205), ('argis', 0.169), ('side', 0.136), ('alignment', 0.113), ('bitext', 0.112), ('argkt', 0.108), ('piak', 0.108), ('sides', 0.107), ('monolingual', 0.099), ('srccmb', 0.096), ('trgcmb', 0.096), ('chinese', 0.092), ('locis', 0.084), ('constraint', 0.078), ('loctk', 0.072), ('aligns', 0.072), ('oa', 0.069), ('bilingual', 0.068), ('ns', 0.068), ('consistency', 0.064), ('predicate', 0.062), ('xij', 0.062), ('arguments', 0.061), ('predicates', 0.061), ('nul', 0.06), ('tup', 0.06), ('align', 0.06), ('inference', 0.059), ('candidates', 0.057), ('nt', 0.056), ('propbank', 0.056), ('loc', 0.056), ('joint', 0.052), ('semantic', 0.051), ('files', 0.05), ('role', 0.05), ('xls', 0.048), ('xlt', 0.048), ('ykj', 0.048), ('jx', 0.048), ('surdeanu', 0.048), ('arquez', 0.048), ('kx', 0.048), ('aligned', 0.044), ('punyakanok', 0.043), ('ilp', 0.043), ('duplication', 0.043), ('pradhan', 0.043), ('parallel', 0.042), ('constraints', 0.041), ('subsection', 0.039), ('pjl', 0.036), ('xnt', 0.036), ('zengzhang', 0.036), ('alignments', 0.035), ('zhuang', 0.034), ('equations', 0.034), ('xj', 0.033), ('fung', 0.032), ('english', 0.032), ('source', 0.032), ('complement', 0.031), ('discontinuous', 0.031), ('target', 0.031), ('llu', 0.03), ('proposition', 0.03), ('ot', 0.03), ('os', 0.03), ('combination', 0.029), ('powell', 0.028), ('ontonotes', 0.028), ('pad', 0.027), ('locations', 0.027), ('soft', 0.026), ('xi', 0.026), ('chineseenglish', 0.026), ('vasin', 0.026), ('burkett', 0.026), ('outputs', 0.025), ('label', 0.024), ('labeling', 0.024), ('chengqing', 0.024), ('conformity', 0.024), ('coreferential', 0.024), ('interrelated', 0.024), ('koomen', 0.024), ('ljs', 0.024), ('ljt', 0.024), ('pred', 0.024), ('ptkj', 0.024), ('xlsxij', 0.024), ('xltykj', 0.024), ('xns', 0.024), ('xnsxnt', 0.024), ('xnszik', 0.024), ('xntzik', 0.024), ('xykj', 0.024)]
simIndex simValue paperId paperTitle
same-paper 1 0.99999976 68 emnlp-2010-Joint Inference for Bilingual Semantic Role Labeling
Author: Tao Zhuang ; Chengqing Zong
Abstract: We show that jointly performing semantic role labeling (SRL) on bitext can improve SRL results on both sides. In our approach, we use monolingual SRL systems to produce argument candidates for predicates in bitext at first. Then, we simultaneously generate SRL results for two sides of bitext using our joint inference model. Our model prefers the bilingual SRL result that is not only reasonable on each side of bitext, but also has more consistent argument structures between two sides. To evaluate the consistency between two argument structures, we also formulate a log-linear model to compute the probability of aligning two arguments. We have experimented with our model on Chinese-English parallel PropBank data. Using our joint inference model, F1 scores of SRL results on Chinese and English text achieve 79.53% and 77.87% respectively, which are 1.52 and 1.74 points higher than the results of baseline monolingual SRL combination systems respectively.
2 0.34761459 95 emnlp-2010-SRL-Based Verb Selection for ESL
Author: Xiaohua Liu ; Bo Han ; Kuan Li ; Stephan Hyeonjun Stiller ; Ming Zhou
Abstract: In this paper we develop an approach to tackle the problem of verb selection for learners of English as a second language (ESL) by using features from the output of Semantic Role Labeling (SRL). Unlike existing approaches to verb selection that use local features such as n-grams, our approach exploits semantic features which explicitly model the usage context of the verb. The verb choice highly depends on its usage context which is not consistently captured by local features. We then combine these semantic features with other local features under the generalized perceptron learning framework. Experiments on both indomain and out-of-domain corpora show that our approach outperforms the baseline and achieves state-of-the-art performance. 1
3 0.23153965 121 emnlp-2010-What a Parser Can Learn from a Semantic Role Labeler and Vice Versa
Author: Stephen Boxwell ; Dennis Mehay ; Chris Brew
Abstract: In many NLP systems, there is a unidirectional flow of information in which a parser supplies input to a semantic role labeler. In this paper, we build a system that allows information to flow in both directions. We make use of semantic role predictions in choosing a single-best parse. This process relies on an averaged perceptron model to distinguish likely semantic roles from erroneous ones. Our system penalizes parses that give rise to low-scoring semantic roles. To explore the consequences of this we perform two experiments. First, we use a baseline generative model to produce n-best parses, which are then re-ordered by our semantic model. Second, we use a modified version of our semantic role labeler to predict semantic roles at parse time. The performance of this modified labeler is weaker than that of our best full SRL, because it is restricted to features that can be computed directly from the parser’s packed chart. For both experiments, the resulting semantic predictions are then used to select parses. Finally, we feed the selected parses produced by each experiment to the full version of our semantic role labeler. We find that SRL performance can be improved over this baseline by selecting parses with likely semantic roles.
4 0.098060004 36 emnlp-2010-Discriminative Word Alignment with a Function Word Reordering Model
Author: Hendra Setiawan ; Chris Dyer ; Philip Resnik
Abstract: We address the modeling, parameter estimation and search challenges that arise from the introduction of reordering models that capture non-local reordering in alignment modeling. In particular, we introduce several reordering models that utilize (pairs of) function words as contexts for alignment reordering. To address the parameter estimation challenge, we propose to estimate these reordering models from a relatively small amount of manuallyaligned corpora. To address the search challenge, we devise an iterative local search algorithm that stochastically explores reordering possibilities. By capturing non-local reordering phenomena, our proposed alignment model bears a closer resemblance to stateof-the-art translation model. Empirical results show significant improvements in alignment quality as well as in translation performance over baselines in a large-scale ChineseEnglish translation task.
5 0.077351123 57 emnlp-2010-Hierarchical Phrase-Based Translation Grammars Extracted from Alignment Posterior Probabilities
Author: Adria de Gispert ; Juan Pino ; William Byrne
Abstract: We report on investigations into hierarchical phrase-based translation grammars based on rules extracted from posterior distributions over alignments of the parallel text. Rather than restrict rule extraction to a single alignment, such as Viterbi, we instead extract rules based on posterior distributions provided by the HMM word-to-word alignmentmodel. We define translation grammars progressively by adding classes of rules to a basic phrase-based system. We assess these grammars in terms of their expressive power, measured by their ability to align the parallel text from which their rules are extracted, and the quality of the translations they yield. In Chinese-to-English translation, we find that rule extraction from posteriors gives translation improvements. We also find that grammars with rules with only one nonterminal, when extracted from posteri- ors, can outperform more complex grammars extracted from Viterbi alignments. Finally, we show that the best way to exploit source-totarget and target-to-source alignment models is to build two separate systems and combine their output translation lattices.
6 0.073508792 67 emnlp-2010-It Depends on the Translation: Unsupervised Dependency Parsing via Word Alignment
7 0.069543503 29 emnlp-2010-Combining Unsupervised and Supervised Alignments for MT: An Empirical Study
8 0.068638124 33 emnlp-2010-Cross Language Text Classification by Model Translation and Semi-Supervised Learning
9 0.055431157 98 emnlp-2010-Soft Syntactic Constraints for Hierarchical Phrase-Based Translation Using Latent Syntactic Distributions
10 0.055298448 15 emnlp-2010-A Unified Framework for Scope Learning via Simplified Shallow Semantic Parsing
11 0.05342247 79 emnlp-2010-Mining Name Translations from Entity Graph Mapping
12 0.049373295 63 emnlp-2010-Improving Translation via Targeted Paraphrasing
13 0.047967263 86 emnlp-2010-Non-Isomorphic Forest Pair Translation
14 0.047647983 21 emnlp-2010-Automatic Discovery of Manner Relations and its Applications
15 0.047035865 107 emnlp-2010-Towards Conversation Entailment: An Empirical Investigation
16 0.043823622 18 emnlp-2010-Assessing Phrase-Based Translation Models with Oracle Decoding
17 0.040306076 69 emnlp-2010-Joint Training and Decoding Using Virtual Nodes for Cascaded Segmentation and Tagging Tasks
18 0.039603297 3 emnlp-2010-A Fast Fertility Hidden Markov Model for Word Alignment Using MCMC
19 0.038632441 50 emnlp-2010-Facilitating Translation Using Source Language Paraphrase Lattices
20 0.038287755 118 emnlp-2010-Utilizing Extra-Sentential Context for Parsing
topicId topicWeight
[(0, 0.175), (1, -0.016), (2, 0.128), (3, 0.146), (4, 0.025), (5, -0.072), (6, 0.055), (7, -0.425), (8, 0.165), (9, 0.16), (10, -0.153), (11, 0.256), (12, -0.183), (13, -0.048), (14, 0.028), (15, 0.097), (16, 0.006), (17, -0.04), (18, 0.209), (19, 0.13), (20, 0.04), (21, 0.02), (22, 0.136), (23, -0.12), (24, 0.055), (25, 0.108), (26, -0.051), (27, -0.013), (28, -0.138), (29, -0.056), (30, 0.058), (31, 0.014), (32, -0.061), (33, 0.098), (34, -0.044), (35, 0.025), (36, -0.065), (37, 0.013), (38, 0.036), (39, -0.098), (40, 0.033), (41, -0.002), (42, -0.097), (43, -0.137), (44, 0.006), (45, -0.064), (46, 0.029), (47, -0.007), (48, 0.011), (49, 0.033)]
simIndex simValue paperId paperTitle
same-paper 1 0.96845496 68 emnlp-2010-Joint Inference for Bilingual Semantic Role Labeling
Author: Tao Zhuang ; Chengqing Zong
Abstract: We show that jointly performing semantic role labeling (SRL) on bitext can improve SRL results on both sides. In our approach, we use monolingual SRL systems to produce argument candidates for predicates in bitext at first. Then, we simultaneously generate SRL results for two sides of bitext using our joint inference model. Our model prefers the bilingual SRL result that is not only reasonable on each side of bitext, but also has more consistent argument structures between two sides. To evaluate the consistency between two argument structures, we also formulate a log-linear model to compute the probability of aligning two arguments. We have experimented with our model on Chinese-English parallel PropBank data. Using our joint inference model, F1 scores of SRL results on Chinese and English text achieve 79.53% and 77.87% respectively, which are 1.52 and 1.74 points higher than the results of baseline monolingual SRL combination systems respectively.
2 0.73007697 95 emnlp-2010-SRL-Based Verb Selection for ESL
Author: Xiaohua Liu ; Bo Han ; Kuan Li ; Stephan Hyeonjun Stiller ; Ming Zhou
Abstract: In this paper we develop an approach to tackle the problem of verb selection for learners of English as a second language (ESL) by using features from the output of Semantic Role Labeling (SRL). Unlike existing approaches to verb selection that use local features such as n-grams, our approach exploits semantic features which explicitly model the usage context of the verb. The verb choice highly depends on its usage context which is not consistently captured by local features. We then combine these semantic features with other local features under the generalized perceptron learning framework. Experiments on both indomain and out-of-domain corpora show that our approach outperforms the baseline and achieves state-of-the-art performance. 1
3 0.59790212 121 emnlp-2010-What a Parser Can Learn from a Semantic Role Labeler and Vice Versa
Author: Stephen Boxwell ; Dennis Mehay ; Chris Brew
Abstract: In many NLP systems, there is a unidirectional flow of information in which a parser supplies input to a semantic role labeler. In this paper, we build a system that allows information to flow in both directions. We make use of semantic role predictions in choosing a single-best parse. This process relies on an averaged perceptron model to distinguish likely semantic roles from erroneous ones. Our system penalizes parses that give rise to low-scoring semantic roles. To explore the consequences of this we perform two experiments. First, we use a baseline generative model to produce n-best parses, which are then re-ordered by our semantic model. Second, we use a modified version of our semantic role labeler to predict semantic roles at parse time. The performance of this modified labeler is weaker than that of our best full SRL, because it is restricted to features that can be computed directly from the parser’s packed chart. For both experiments, the resulting semantic predictions are then used to select parses. Finally, we feed the selected parses produced by each experiment to the full version of our semantic role labeler. We find that SRL performance can be improved over this baseline by selecting parses with likely semantic roles.
4 0.25488102 36 emnlp-2010-Discriminative Word Alignment with a Function Word Reordering Model
Author: Hendra Setiawan ; Chris Dyer ; Philip Resnik
Abstract: We address the modeling, parameter estimation and search challenges that arise from the introduction of reordering models that capture non-local reordering in alignment modeling. In particular, we introduce several reordering models that utilize (pairs of) function words as contexts for alignment reordering. To address the parameter estimation challenge, we propose to estimate these reordering models from a relatively small amount of manuallyaligned corpora. To address the search challenge, we devise an iterative local search algorithm that stochastically explores reordering possibilities. By capturing non-local reordering phenomena, our proposed alignment model bears a closer resemblance to stateof-the-art translation model. Empirical results show significant improvements in alignment quality as well as in translation performance over baselines in a large-scale ChineseEnglish translation task.
5 0.22090702 21 emnlp-2010-Automatic Discovery of Manner Relations and its Applications
Author: Eduardo Blanco ; Dan Moldovan
Abstract: This paper presents a method for the automatic discovery of MANNER relations from text. An extended definition of MANNER is proposed, including restrictions on the sorts of concepts that can be part of its domain and range. The connections with other relations and the lexico-syntactic patterns that encode MANNER are analyzed. A new feature set specialized on MANNER detection is depicted and justified. Experimental results show improvement over previous attempts to extract MANNER. Combinations of MANNER with other semantic relations are also discussed.
6 0.21056767 29 emnlp-2010-Combining Unsupervised and Supervised Alignments for MT: An Empirical Study
7 0.20788167 15 emnlp-2010-A Unified Framework for Scope Learning via Simplified Shallow Semantic Parsing
8 0.20167546 107 emnlp-2010-Towards Conversation Entailment: An Empirical Investigation
9 0.20167039 7 emnlp-2010-A Mixture Model with Sharing for Lexical Semantics
10 0.19907928 92 emnlp-2010-Predicting the Semantic Compositionality of Prefix Verbs
11 0.18934745 67 emnlp-2010-It Depends on the Translation: Unsupervised Dependency Parsing via Word Alignment
12 0.18830395 79 emnlp-2010-Mining Name Translations from Entity Graph Mapping
13 0.18300459 33 emnlp-2010-Cross Language Text Classification by Model Translation and Semi-Supervised Learning
14 0.16552317 57 emnlp-2010-Hierarchical Phrase-Based Translation Grammars Extracted from Alignment Posterior Probabilities
15 0.14975312 43 emnlp-2010-Enhancing Domain Portability of Chinese Segmentation Model Using Chi-Square Statistics and Bootstrapping
16 0.14637628 109 emnlp-2010-Translingual Document Representations from Discriminative Projections
17 0.14632617 105 emnlp-2010-Title Generation with Quasi-Synchronous Grammar
18 0.14244066 25 emnlp-2010-Better Punctuation Prediction with Dynamic Conditional Random Fields
19 0.13864511 63 emnlp-2010-Improving Translation via Targeted Paraphrasing
20 0.13763486 118 emnlp-2010-Utilizing Extra-Sentential Context for Parsing
topicId topicWeight
[(10, 0.012), (12, 0.054), (29, 0.091), (30, 0.013), (32, 0.014), (38, 0.355), (52, 0.056), (56, 0.08), (62, 0.028), (66, 0.092), (72, 0.034), (76, 0.045), (82, 0.021), (87, 0.011)]
simIndex simValue paperId paperTitle
same-paper 1 0.75056314 68 emnlp-2010-Joint Inference for Bilingual Semantic Role Labeling
Author: Tao Zhuang ; Chengqing Zong
Abstract: We show that jointly performing semantic role labeling (SRL) on bitext can improve SRL results on both sides. In our approach, we use monolingual SRL systems to produce argument candidates for predicates in bitext at first. Then, we simultaneously generate SRL results for two sides of bitext using our joint inference model. Our model prefers the bilingual SRL result that is not only reasonable on each side of bitext, but also has more consistent argument structures between two sides. To evaluate the consistency between two argument structures, we also formulate a log-linear model to compute the probability of aligning two arguments. We have experimented with our model on Chinese-English parallel PropBank data. Using our joint inference model, F1 scores of SRL results on Chinese and English text achieve 79.53% and 77.87% respectively, which are 1.52 and 1.74 points higher than the results of baseline monolingual SRL combination systems respectively.
2 0.63725597 103 emnlp-2010-Tense Sense Disambiguation: A New Syntactic Polysemy Task
Author: Roi Reichart ; Ari Rappoport
Abstract: Polysemy is a major characteristic of natural languages. Like words, syntactic forms can have several meanings. Understanding the correct meaning of a syntactic form is of great importance to many NLP applications. In this paper we address an important type of syntactic polysemy the multiple possible senses of tense syntactic forms. We make our discussion concrete by introducing the task of Tense Sense Disambiguation (TSD): given a concrete tense syntactic form present in a sentence, select its appropriate sense among a set of possible senses. Using English grammar textbooks, we compiled a syntactic sense dictionary comprising common tense syntactic forms and semantic senses for each. We annotated thousands of BNC sentences using the – defined senses. We describe a supervised TSD algorithm trained on these annotations, which outperforms a strong baseline for the task.
3 0.42092234 105 emnlp-2010-Title Generation with Quasi-Synchronous Grammar
Author: Kristian Woodsend ; Yansong Feng ; Mirella Lapata
Abstract: The task of selecting information and rendering it appropriately appears in multiple contexts in summarization. In this paper we present a model that simultaneously optimizes selection and rendering preferences. The model operates over a phrase-based representation of the source document which we obtain by merging PCFG parse trees and dependency graphs. Selection preferences for individual phrases are learned discriminatively, while a quasi-synchronous grammar (Smith and Eisner, 2006) captures rendering preferences such as paraphrases and compressions. Based on an integer linear programming formulation, the model learns to generate summaries that satisfy both types of preferences, while ensuring that length, topic coverage and grammar constraints are met. Experiments on headline and image caption generation show that our method obtains state-of-the-art performance using essentially the same model for both tasks without any major modifications.
4 0.42066371 18 emnlp-2010-Assessing Phrase-Based Translation Models with Oracle Decoding
Author: Guillaume Wisniewski ; Alexandre Allauzen ; Francois Yvon
Abstract: Extant Statistical Machine Translation (SMT) systems are very complex softwares, which embed multiple layers of heuristics and embark very large numbers of numerical parameters. As a result, it is difficult to analyze output translations and there is a real need for tools that could help developers to better understand the various causes of errors. In this study, we make a step in that direction and present an attempt to evaluate the quality of the phrase-based translation model. In order to identify those translation errors that stem from deficiencies in the phrase table (PT), we propose to compute the oracle BLEU-4 score, that is the best score that a system based on this PT can achieve on a reference corpus. By casting the computation of the oracle BLEU-1 as an Integer Linear Programming (ILP) problem, we show that it is possible to efficiently compute accurate lower-bounds of this score, and report measures performed on several standard benchmarks. Various other applications of these oracle decoding techniques are also reported and discussed. 1 Phrase-Based Machine Translation 1.1 Principle A Phrase-Based Translation System (PBTS) consists of a ruleset and a scoring function (Lopez, 2009). The ruleset, represented in the phrase table, is a set of phrase1pairs {(f, e) }, each pair expressing that the source phrase f can ,bee) r}e,w earicthten p (atirra enxslparteedss)i inngto t a target phrase e. Trarsaens flation hypotheses are generated by iteratively rewriting portions of the source sentence as prescribed by the ruleset, until each source word has been consumed by exactly one rule. The order of target words in an hypothesis is uniquely determined by the order in which the rewrite operation are performed. The search space ofthe translation model corresponds to the set of all possible sequences of 1Following the usage in statistical machine translation literature, use “phrase” to denote a subsequence of consecutive words. we 933 rules applications. The scoring function aims to rank all possible translation hypotheses in such a way that the best one has the highest score. A PBTS is learned from a parallel corpus in two independent steps. In a first step, the corpus is aligned at the word level, by using alignment tools such as Gi z a++ (Och and Ney, 2003) and some symmetrisation heuristics; phrases are then extracted by other heuristics (Koehn et al., 2003) and assigned numerical weights. In the second step, the parameters of the scoring function are estimated, typically through Minimum Error Rate training (Och, 2003). Translating a sentence amounts to finding the best scoring translation hypothesis in the search space. Because of the combinatorial nature of this problem, translation has to rely on heuristic search techniques such as greedy hill-climbing (Germann, 2003) or variants of best-first search like multi-stack decoding (Koehn, 2004). Moreover, to reduce the overall complexity of decoding, the search space is typically pruned using simple heuristics. For instance, the state-of-the-art phrase-based decoder Moses (Koehn et al., 2007) considers only a restricted number of translations for each source sequence2 and enforces a distortion limit3 over which phrases can be reordered. As a consequence, the best translation hypothesis returned by the decoder is not always the one with the highest score. 1.2 Typology of PBTS Errors Analyzing the errors of a SMT system is not an easy task, because of the number of models that are combined, the size of these models, and the high complexity of the various decision making processes. For a SMT system, three different kinds of errors can be distinguished (Germann et al., 2004; Auli et al., 2009): search errors, induction errors and model errors. The former corresponds to cases where the hypothesis with the best score is missed by the search procedure, either because of the use of an ap2the 3the option of Moses, defaulting to 20. dl option of Moses, whose default value is 7. tt l ProceMedITin,g Ms oasfs thaceh 2u0se1t0ts C,o UnSfAer,e n9c-e11 on O Ectmobpeir ic 2a0l1 M0.e ?tc ho2d0s10 in A Nsastouciraatlio Lnan fogru Cagoem Ppruotcaetisosninagl, L pinaggeusis 9t3ic3s–943, proximate search method or because of the restrictions of the search space. Induction errors correspond to cases where, given the model, the search space does not contain the reference. Finally, model errors correspond to cases where the hypothesis with the highest score is not the best translation according to the evaluation metric. Model errors encompass several types oferrors that occur during learning (Bottou and Bousquet, 2008)4. Approximation errors are errors caused by the use of a restricted and oversimplistic class of functions (here, finitestate transducers to model the generation of hypotheses and a linear scoring function to discriminate them) to model the translation process. Estimation errors correspond to the use of sub-optimal values for both the phrase pairs weights and the parameters of the scoring function. The reasons behind these errors are twofold: first, training only considers a finite sample of data; second, it relies on error prone alignments. As a result, some “good” phrases are extracted with a small weight, or, in the limit, are not extracted at all; and conversely that some “poor” phrases are inserted into the phrase table, sometimes with a really optimistic score. Sorting out and assessing the impact of these various causes of errors is of primary interest for SMT system developers: for lack of such diagnoses, it is difficult to figure out which components of the system require the most urgent attention. Diagnoses are however, given the tight intertwining among the various component of a system, very difficult to obtain: most evaluations are limited to the computation of global scores and usually do not imply any kind of failure analysis. 1.3 Contribution and organization To systematically assess the impact of the multiple heuristic decisions made during training and decoding, we propose, following (Dreyer et al., 2007; Auli et al., 2009), to work out oracle scores, that is to evaluate the best achievable performances of a PBTS. We aim at both studying the expressive power of PBTS and at providing tools for identifying and quantifying causes of failure. Under standard metrics such as BLEU (Papineni et al., 2002), oracle scores are difficult (if not impossible) to compute, but, by casting the computation of the oracle unigram recall and precision as an Integer Linear Programming (ILP) problem, we show that it is possible to efficiently compute accurate lower-bounds of the oracle BLEU-4 scores and report measurements performed on several standard benchmarks. The main contributions of this paper are twofold. We first introduce an ILP program able to efficiently find the best hypothesis a PBTS can achieve. This program can be easily extended to test various improvements to 4We omit here optimization errors. 934 phrase-base systems or to evaluate the impact of different parameter settings. Second, we present a number of complementary results illustrating the usage of our oracle decoder for identifying and analyzing PBTS errors. Our experimental results confirm the main conclusions of (Turchi et al., 2008), showing that extant PBTs have the potential to generate hypotheses having very high BLEU4 score and that their main bottleneck is their scoring function. The rest of this paper is organized as follows: in Section 2, we introduce and formalize the oracle decoding problem, and present a series of ILP problems of increasing complexity designed so as to deliver accurate lowerbounds of oracle score. This section closes with various extensions allowing to model supplementary constraints, most notably reordering constraints (Section 2.5). Our experiments are reported in Section 3, where we first introduce the training and test corpora, along with a description of our system building pipeline (Section 3. 1). We then discuss the baseline oracle BLEU scores (Section 3.2), analyze the non-reachable parts of the reference translations, and comment several complementary results which allow to identify causes of failures. Section 4 discuss our approach and findings with respect to the existing literature on error analysis and oracle decoding. We conclude and discuss further prospects in Section 5. 2 Oracle Decoder 2.1 The Oracle Decoding Problem Definition To get some insights on the errors of phrasebased systems and better understand their limits, we propose to consider the oracle decoding problem defined as follows: given a source sentence, its reference translation5 and a phrase table, what is the “best” translation hypothesis a system can generate? As usual, the quality of an hypothesis is evaluated by the similarity between the reference and the hypothesis. Note that in the oracle decoding problem, we are only assessing the ability of PBT systems to generate good candidate translations, irrespective of their ability to score them properly. We believe that studying this problem is interesting for various reasons. First, as described in Section 3.4, comparing the best hypothesis a system could have generated and the hypothesis it actually generates allows us to carry on both quantitative and qualitative failure analysis. The oracle decoding problem can also be used to assess the expressive power of phrase-based systems (Auli et al., 2009). Other applications include computing acceptable pseudo-references for discriminative training (Tillmann and Zhang, 2006; Liang et al., 2006; Arun and 5The oracle decoding problem can be extended to the case of multiple references. For the sake of simplicity, we only describe the case of a single reference. Koehn, 2007) or combining machine translation systems in a multi-source setting (Li and Khudanpur, 2009). We have also used oracle decoding to identify erroneous or difficult to translate references (Section 3.3). Evaluation Measure To fully define the oracle decoding problem, a measure of the similarity between a translation hypothesis and its reference translation has to be chosen. The most obvious choice is the BLEU-4 score (Papineni et al., 2002) used in most machine translation evaluations. However, using this metric in the oracle decoding problem raises several issues. First, BLEU-4 is a metric defined at the corpus level and is hard to interpret at the sentence level. More importantly, BLEU-4 is not decomposable6: as it relies on 4-grams statistics, the contribution of each phrase pair to the global score depends on the translation of the previous and following phrases and can not be evaluated in isolation. Because of its nondecomposability, maximizing BLEU-4 is hard; in particular, the phrase-level decomposability of the evaluation × metric is necessary in our approach. To circumvent this difficulty, we propose to evaluate the similarity between a translation hypothesis and a reference by the number of their common words. This amounts to evaluating translation quality in terms of unigram precision and recall, which are highly correlated with human judgements (Lavie et al., ). This measure is closely related to the BLEU-1 evaluation metric and the Meteor (Banerjee and Lavie, 2005) metric (when it is evaluated without considering near-matches and the distortion penalty). We also believe that hypotheses that maximize the unigram precision and recall at the sentence level yield corpus level BLEU-4 scores close the maximal achievable. Indeed, in the setting we will introduce in the next section, BLEU-1 and BLEU-4 are highly correlated: as all correct words of the hypothesis will be compelled to be at their correct position, any hypothesis with a high 1-gram precision is also bound to have a high 2-gram precision, etc. 2.2 Formalizing the Oracle Decoding Problem The oracle decoding problem has already been considered in the case of word-based models, in which all translation units are bound to contain only one word. The problem can then be solved by a bipartite graph matching algorithm (Leusch et al., 2008): given a n m binary matarligxo describing possible t 2r0an08sl)a:ti goinv elinn aks n b×emtw beeinna source words and target words7, this algorithm finds the subset of links maximizing the number of words of the reference that have been translated, while ensuring that each word 6Neither at the sentence (Chiang et al., 2008), nor at the phrase level. 7The (i, j) entry of the matrix is 1if the ith word of the source can be translated by the jth word of the reference, 0 otherwise. 935 is translated only once. Generalizing this approach to phrase-based systems amounts to solving the following problem: given a set of possible translation links between potential phrases of the source and of the target, find the subset of links so that the unigram precision and recall are the highest possible. The corresponding oracle hypothesis can then be easily generated by selecting the target phrases that are aligned with one source phrase, disregarding the others. In addition, to mimic the way OOVs are usually handled, we match identical OOV tokens appearing both in the source and target sentences. In this approach, the unigram precision is always one (every word generated in the oracle hypothesis matches exactly one word in the reference). As a consequence, to find the oracle hypothesis, we just have to maximize the recall, that is the number of words appearing both in the hypothesis and in the reference. Considering phrases instead of isolated words has a major impact on the computational complexity: in this new setting, the optimal segmentations in phrases of both the source and of the target have to be worked out in addition to links selection. Moreover, constraints have to be taken into account so as to enforce a proper segmentation of the source and target sentences. These constraints make it impossible to use the approach of (Leusch et al., 2008) and concur in making the oracle decoding problem for phrase-based models more complex than it is for word-based models: it can be proven, using arguments borrowed from (De Nero and Klein, 2008), that this problem is NP-hard even for the simple unigram precision measure. 2.3 An Integer Program for Oracle Decoding To solve the combinatorial problem introduced in the previous section, we propose to cast it into an Integer Linear Programming (ILP) problem, for which many generic solvers exist. ILP has already been used in SMT to find the optimal translation for word-based (Germann et al., 2001) and to study the complexity of learning phrase alignments (De Nero and Klein, 2008) models. Following the latter reference, we introduce the following variables: fi,j (resp. ek,l) is a binary indicator variable that is true when the phrase contains all spans from betweenword position i to j (resp. k to l) of the source (resp. target) sentence. We also introduce a binary variable, denoted ai,j,k,l, to describe a possible link between source phrase fi,j and target phrase ek,l. These variables are built from the entries of the phrase table according to selection strategies introduced in Section 2.4. In the following, index variables are so that: 0 ≤ i< j ≤ n, in the source sentence and 0 ≤ k < l ≤ m, in the target sentence, where n (resp. m) is the length of the source (resp. target) sentence. Solving the oracle decoding problem then amounts to optimizing the following objective function: mi,j,akx,li,Xj,k,lai,j,k,l· (l − k), (1) under the constraints: X ∀x ∈ J1,mK : ek,l ≤ 1 (2) = (3) 1∀,kn,lK : Xai,j,k,l = fk,l (4) ∀i,j : Xai,j,k,l (5) k,l s.tX. Xk≤x≤l ∀∀xy ∈∈ J11,,mnKK : X i,j s.tX. Xi≤y≤j fi,j 1 Xi,j = ei,j Xk,l The objective function (1) corresponds to the number of target words that are generated. The first set of constraints (2) ensures that each word in the reference e ap- pears in no more than one phrase. Maximizing the objective under these constraints amounts to maximizing the unigram recall. The second set of constraints (3) ensures that each word in the source f is translated exactly once, which guarantees that the search space of the ILP problem is the same as the search space of a phrase-based system. Constraints (4) bind the fk,l and ai,j,k,l variables, ensuring that whenever a link ai,j,k,l is active, the corresponding phrase fk,l is also active. Constraints (5) play a similar role for the reference. The Relaxed Problem Even though it accurately models the search space of a phrase-based decoder, this programs is not really useful as is: due to out-ofvocabulary words or missing entries in the phrase table, the constraint that all source words should be translated yields infeasible problems8. We propose to relax this problem and allow some source words to remain untranslated. This is done by replacing constraints (3) by: ∀y ∈ J1,nK : X i,j s.tX. Xi≤y≤j fi,j ≤ 1 To better ref∀lyec ∈t th J1e, bneKh :avior of phrase-based decoders, which attempt to translate all source words, we also need to modify the objective function as follows: X i,Xj,k,l ai,j,k,l · (l − k) +Xfi,j · (j − i) Xi,j (6) The second term in this new objective ensures that optimal solutions translate as many source words as possible. 8An ILP problem is said to be infeasible when tion violates at least one constraint. every possible solu- 936 The Relaxed-Distortion Problem A last caveat with the Relaxed optimization program is caused by frequently occurring source tokens, such as function words or punctuation signs, which can often align with more than one target word. For lack of taking distortion information into account in our objective function, all these alignments are deemed equivalent, even if some of them are clearly more satisfactory than others. This situation is illustrated on Figure 1. le chat et the cat and le the chien dog Figure 1: Equivalent alignments between “le” and “the”. The dashed lines corresponds to a less interpretable solution. To overcome this difficulty, we propose a last change to the objective function: X i,Xj,k,l ai,j,k,l · (l − k) +Xfi,j · (j − i) X ai,j,k,l|k − i| Xi,j −α (7) i Xk ,l X,j, Compared to the objective function of the relaxed problem (6), we introduce here a supplementary penalty factor which favors monotonous alignments. For each phrase pair, the higher the difference between source and target positions, the higher this penalty. If α is small enough, this extra term allows us to select, among all the optimal alignments of the re l axed problem, the one with the lowest distortion. In our experiments, we set α to min {n, m} to ensure that the penalty factor is always smminall{enr, ,tmha}n tthoe e rneswuarred t fhoart aligning atwltyo single iwso ardlwsa. 2.4 Selecting Indicator Variables In the approach introduced in the previous sections, the oracle decoding problem is solved by selecting, among a set of possible translation links, the ones that yield the solution with the highest unigram recall. We propose two strategies to build this set of possible translation links. In the first one, denoted exact match, an indicator ai,j,k,l is created if there is an entry (f, e) so that f spans from word position ito j in the source and e from word position k to l in the target. In this strategy, the ILP program considers exactly the same ruleset as conventional phrase-based decoders. We also consider an alternative strategy, which could help us to identify errors made during the phrase extraction process. In this strategy, denoted inside match, an indicator ai,j,k,l is created when the following three criteria are met: i) f spans from position ito j of the source; ii) a substring of e, denoted e, spans from position k to l of the reference; iii) (f, e¯) is not an entry of the phrase table. The resulting set of indicator variables thus contains, at least, all the variables used in the exact match strategy. In addition, we license here the use of phrases containing words that do not occur in the reference. In fact, using such solutions can yield higher BLEU scores when the reward for additional correct matches exceeds the cost incurred by wrong predictions. These cases are symptoms of situations where the extraction heuristic failed to extract potentially useful subphrases. 2.5 Oracle Decoding with Reordering Constraints The ILP problem introduced in the previous section can be extended in several ways to describe and test various improvements to phrase-based systems or to evaluate the impact of different parameter settings. This flexibility mainly stems from the possibility offered by our framework to express arbitrary constraints over variables. In this section, we illustrate these possibilities by describing how reordering constraints can easily be considered. As a first example, the Moses decoder uses a distortion limit to constrain the set of possible reorderings. This constraint “enforces (...) that the last word of a phrase chosen for translation cannot be more than d9 words from the leftmost untranslated word in the source” (Lopez, 2009) and is expressed as: ∀aijkl , ai0j0k0l0 s.t. k > k0, aijkl · ai0j0k0l0 · |j − i0 + 1| ≤ d, The maximum distortion limit strategy (Lopez, 2009) is also easily expressed and take the following form (assuming this constraint is parameterized by d): ∀l < m − 1, ai,j,k,l·ai0,j0,l+1,l0 · |i0 − j − 1| 71is%t e6hs.a distortion greater that Moses default distortion limit. alignment decisions enabled by the use of larger training corpora and phrase table. To evaluate the impact ofthe second heuristic, we computed the number of phrases discarded by Moses (be- cause of the default ttl limit) but used in the oracle hypotheses. In the English to French NEWSCO setting, they account for 34.11% of the total number of phrases used in the oracle hypotheses. When the oracle decoder is constrained to use the same phrase table as Moses, its BLEU-4 score drops to 42.78. This shows that filtering the phrase table prior to decoding discards many useful phrase pairs and is seriously limiting the best achievable performance, a conclusion shared with (Auli et al., 2009). Search Errors Search errors can be identified by comparing the score of the best hypothesis found by Moses and the score of the oracle hypothesis. If the score of the oracle hypothesis is higher, then there has been a search error; on the contrary, there has been an estimation error when the score of the oracle hypothesis is lower than the score of the best hypothesis found by Moses. 940 Based on the comparison of the score of Moses hypotheses and of oracle hypotheses for the English to French NEWSCO setting, our preliminary conclusion is that the number of search errors is quite limited: only about 5% of the hypotheses of our oracle decoder are actually getting a better score than Moses solutions. Again, this shows that the scoring function (model error) is one of the main bottleneck of current PBTS. Comparing these hypotheses is nonetheless quite revealing: while Moses mostly selects phrase pairs with high translation scores and generates monotonous alignments, our ILP decoder uses larger reorderings and less probable phrases to achieve better solutions: on average, the reordering score of oracle solutions is −5.74, compared to −76.78 fscoro rMeo osfe osr outputs. iGonivsen is −the5 weight assigned through MERT training to the distortion score, no wonder that these hypotheses are severely penalized. The Impact of Phrase Length The observed outputs do not only depend on decisions made during the search, but also on decisions made during training. One such decision is the specification of maximal length for the source and target phrases. In our framework, evaluating the impact of this decision is simple: it suffices to change the definition of indicator variables so as to consider only alignments between phrases of a given length. In the English-French NEWSCO setting, the most restrictive choice, when only alignments between single words are authorized, yields an oracle BLEU-4 of 48.68; however, authorizing phrases up to length 2 allows to achieve an oracle value of 66.57, very close to the score achieved when considering all extracted phrases (67.77). This is corroborated with a further analysis of our oracle alignments, which use phrases whose average source length is 1.21 words (respectively 1.31 for target words). If many studies have already acknowledged the predomi- nance of “small” phrases in actual translations, our oracle scores suggest that, for this language pair, increasing the phrase length limit beyond 2 or 3 might be a waste of computational resources. 4 Related Work To the best of our knowledge, there are only a few works that try to study the expressive power ofphrase-based machine translation systems or to provide tools for analyzing potential causes of failure. The approach described in (Auli et al., 2009) is very similar to ours: in this study, the authors propose to find and analyze the limits of machine translation systems by studying the reference reachability. A reference is reachable for a given system if it can be exactly generated by this system. Reference reachability is assessed using Moses in forced decoding mode: during search, all hypotheses that deviate from the reference are simply discarded. Even though the main goal of this study was to compare the search space of phrase-based and hierarchical systems, it also provides some insights on the impact of various search parameters in Moses, delivering conclusions that are consistent with our main results. As described in Section 1.2, these authors also propose a typology of the errors of a statistical translation systems, but do not attempt to provide methods for identifying them. The authors of (Turchi et al., 2008) study the learn- ing capabilities of Moses by extensively analyzing learning curves representing the translation performances as a function of the number of examples, and by corrupting the model parameters. Even though their focus is more on assessing the scoring function, they reach conclusions similar to ours: the current bottleneck of translation performances is not the representation power of the PBTS but rather in their scoring functions. Oracle decoding is useful to compute reachable pseudo-references in the context of discriminative training. This is the main motivation of (Tillmann and Zhang, 2006), where the authors compute high BLEU hypotheses by running a conventional decoder so as to maximize a per-sentence approximation of BLEU-4, under a simple (local) reordering model. Oracle decoding has also been used to assess the limitations induced by various reordering constraints in (Dreyer et al., 2007). To this end, the authors propose to use a beam-search based oracle decoder, which computes lower bounds of the best achievable BLEU-4 using dynamic programming techniques over finite-state (for so-called local and IBM constraints) or hierarchically structured (for ITG constraints) sets of hypotheses. Even 941 though the numbers reported in this study are not directly comparable with ours17, it seems that our decoder is not only conceptually much simpler, but also achieves much more optimistic lower-bounds of the oracle BLEU score. The approach described in (Li and Khudanpur, 2009) employs a similar technique, which is to guide a heuristic search in an hypergraph representing possible translation hypotheses with n-gram counts matches, which amounts to decoding with a n-gram model trained on the sole reference translation. Additional tricks are presented in this article to speed-up decoding. Computing oracle BLEU scores is also the subject of (Zens and Ney, 2005; Leusch et al., 2008), yet with a different emphasis. These studies are concerned with finding the best hypotheses in a word graph or in a consensus network, a problem that has various implications for multi-pass decoding and/or system combination techniques. The former reference describes an exponential approximate algorithm, while the latter proves the NPcompleteness of this problem and discuss various heuristic approaches. Our problem is somewhat more complex and using their techniques would require us to built word graphs containing all the translations induced by arbitrary segmentations and permutations of the source sentence. 5 Conclusions In this paper, we have presented a methodology for analyzing the errors of PBTS, based on the computation of an approximation of the BLEU-4 oracle score. We have shown that this approximation could be computed fairly accurately and efficiently using Integer Linear Programming techniques. Our main result is a confirmation of the fact that extant PBTS systems are expressive enough to achieve very high translation performance with respect to conventional quality measurements. The main efforts should therefore strive to improve on the way phrases and hypotheses are scored during training. This gives further support to attempts aimed at designing context-dependent scoring functions as in (Stroppa et al., 2007; Gimpel and Smith, 2008), or at attempts to perform discriminative training of feature-rich models. (Bangalore et al., 2007). We have shown that the examination of difficult-totranslate sentences was an effective way to detect errors or inconsistencies in the reference translations, making our approach a potential aid for controlling the quality or assessing the difficulty of test data. Our experiments have also highlighted the impact of various parameters. Various extensions of the baseline ILP program have been suggested and/or evaluated. In particular, the ILP formalism lends itself well to expressing various constraints that are typically used in conventional PBTS. In 17The best BLEU-4 oracle they achieve on Europarl German to English is approximately 48; but they considered a smaller version of the training corpus and the WMT’06 test set. our future work, we aim at using this ILP framework to systematically assess various search configurations. We plan to explore how replacing non-reachable references with high-score pseudo-references can improve discrim- inative training of PBTS. We are also concerned by determining how tight is our approximation of the BLEU4 score is: to this end, we intend to compute the best BLEU-4 score within the n-best solutions of the oracle decoding problem. Acknowledgments Warm thanks to Houda Bouamor for helping us with the annotation tool. This work has been partly financed by OSEO, the French State Agency for Innovation, under the Quaero program. References Tobias Achterberg. 2007. Constraint Integer Programming. Ph.D. thesis, Technische Universit a¨t Berlin. http : / / opus .kobv .de /tuberl in/vol ltexte / 2 0 0 7 / 16 11/ . Abhishek Arun and Philipp Koehn. 2007. Online learning methods for discriminative training of phrase based statistical machine translation. In Proc. of MT Summit XI, Copenhagen, Denmark. Michael Auli, Adam Lopez, Hieu Hoang, and Philipp Koehn. 2009. A systematic analysis of translation model search spaces. In Proc. of WMT, pages 224–232, Athens, Greece. Satanjeev Banerjee and Alon Lavie. 2005. METEOR: An automatic metric for MT evaluation with improved correlation with human judgments. In Proc. of the ACL Workshop on Intrinsic and Extrinsic Evaluation Measures for Machine Translation and/or Summarization, pages 65–72, Ann Arbor, Michigan. Srinivas Bangalore, Patrick Haffner, and Stephan Kanthak. 2007. Statistical machine translation through global lexical selection and sentence reconstruction. In Proc. of ACL, pages 152–159, Prague, Czech Republic. L e´on Bottou and Olivier Bousquet. 2008. The tradeoffs oflarge scale learning. In Proc. of NIPS, pages 161–168, Vancouver, B.C., Canada. Chris Callison-Burch, Philipp Koehn, Christof Monz, and Josh Schroeder. 2009. Findings of the 2009 Workshop on Statistical Machine Translation. In Proc. of WMT, pages 1–28, Athens, Greece. David Chiang, Steve DeNeefe, Yee Seng Chan, and Hwee Tou Ng. 2008. Decomposability of translation metrics for improved evaluation and efficient algorithms. In Proc. of ECML, pages 610–619, Honolulu, Hawaii. John De Nero and Dan Klein. 2008. The complexity of phrase alignment problems. In Proc. of ACL: HLT, Short Papers, pages 25–28, Columbus, Ohio. Markus Dreyer, Keith B. Hall, and Sanjeev P. Khudanpur. 2007. Comparing reordering constraints for smt using efficient bleu oracle computation. In NAACL-HLT/AMTA Workshop on Syntax and Structure in Statistical Translation, pages 103– 110, Rochester, New York. 942 Ulrich Germann, Michael Jahr, Kevin Knight, Daniel Marcu, and Kenji Yamada. 2001 . Fast decoding and optimal decoding for machine translation. In Proc. of ACL, pages 228–235, Toulouse, France. Ulrich Germann, Michael Jahr, Kevin Knight, Daniel Marcu, and Kenji Yamada. 2004. Fast and optimal decoding for machine translation. Artificial Intelligence, 154(1-2): 127– 143. Ulrich Germann. 2003. Greedy decoding for statistical machine translation in almost linear time. In Proc. of NAACL, pages 1–8, Edmonton, Canada. Kevin Gimpel and Noah A. Smith. 2008. Rich source-side context for statistical machine translation. In Proc. of WMT, pages 9–17, Columbus, Ohio. Philipp Koehn, Franz Josef Och, and Daniel Marcu. 2003. Statistical phrase-based translation. In Proc. of NAACL, pages 48–54, Edmonton, Canada. Philipp Koehn, Hieu Hoang, Alexandra Birch, Chris CallisonBurch, Marcello Federico, Nicola Bertoldi, Brooke Cowan, Wade Shen, Christine Moran, Richard Zens, Chris Dyer, Ondrej Bojar, Alexandra Constantin, and Evan Herbst. 2007. Moses: Open source toolkit for statistical machine translation. In Proc. of ACL, demonstration session. Philipp Koehn. 2004. Pharaoh: A beam search decoder for phrase-based statistical machine translation models. In Proc. of AMTA, pages 115–124, Washington DC. Shankar Kumar and William Byrne. 2005. Local phrase reordering models for statistical machine translation. In Proc. of HLT, pages 161–168, Vancouver, Canada. Alon Lavie, Kenji Sagae, and Shyamsundar Jayaraman. The significance of recall in automatic metrics for MT evaluation. In In Proc. of AMTA, pages 134–143, Washington DC. Gregor Leusch, Evgeny Matusov, and Hermann Ney. 2008. Complexity of finding the BLEU-optimal hypothesis in a confusion network. In Proc. of EMNLP, pages 839–847, Honolulu, Hawaii. Zhifei Li and Sanjeev Khudanpur. 2009. Efficient extraction of oracle-best translations from hypergraphs. In Proc. of NAACL, pages 9–12, Boulder, Colorado. Percy Liang, Alexandre Bouchard-C oˆt´ e, Dan Klein, and Ben Taskar. 2006. An end-to-end discriminative approach to machine translation. In Proc. of ACL, pages 761–768, Sydney, Australia. Adam Lopez. 2009. Translation as weighted deduction. In Proc. of EACL, pages 532–540, Athens, Greece. Franz Josef Och and Hermann Ney. 2003. A systematic comparison of various statistical alignment models. Comput. Linguist. , 29(1): 19–5 1. Franz Josef Och. 2003. Minimum error rate training in statistical machine translation. In Proc. of ACL, pages 160–167, Sapporo, Japan. Kishore Papineni, Salim Roukos, Todd Ward, and Wei-jing Zhu. 2002. Bleu: A method for automatic evaluation of machine translation. Technical report, Philadelphia, Pennsylvania. D. Roth and W. Yih. 2005. Integer linear programming inference for conditional random fields. In Proc. of ICML, pages 737–744, Bonn, Germany. Nicolas Stroppa, Antal van den Bosch, and Andy Way. 2007. Exploiting source similarity for smt using context-informed features. In Andy Way and Barbara Proc. of TMI, pages Christoph Tillmann 231–240, Sk¨ ovde, and Tong Zhang. Gawronska, editors, Sweden. 2006. A discriminative global training algorithm for statistical mt. In Proc. of ACL, 721–728, Sydney, Australia. Turchi, Tijl De Bie, and Nello pages Marco Cristianini. 2008. Learn- ing performance of a machine translation system: a statistical and computational analysis. In Proc. of WMT, pages Columbus, Ohio. 35–43, Richard Zens and Hermann Ney. 2005. Word graphs for statistical machine translation. In Proc. of the ACL Workshop on Building and Using Parallel Texts, pages 191–198, Ann Arbor, Michigan. 943
5 0.41684863 35 emnlp-2010-Discriminative Sample Selection for Statistical Machine Translation
Author: Sankaranarayanan Ananthakrishnan ; Rohit Prasad ; David Stallard ; Prem Natarajan
Abstract: Production of parallel training corpora for the development of statistical machine translation (SMT) systems for resource-poor languages usually requires extensive manual effort. Active sample selection aims to reduce the labor, time, and expense incurred in producing such resources, attaining a given performance benchmark with the smallest possible training corpus by choosing informative, nonredundant source sentences from an available candidate pool for manual translation. We present a novel, discriminative sample selection strategy that preferentially selects batches of candidate sentences with constructs that lead to erroneous translations on a held-out development set. The proposed strategy supports a built-in diversity mechanism that reduces redundancy in the selected batches. Simulation experiments on English-to-Pashto and Spanish-to-English translation tasks demon- strate the superiority of the proposed approach to a number of competing techniques, such as random selection, dissimilarity-based selection, as well as a recently proposed semisupervised active learning strategy.
7 0.41368169 65 emnlp-2010-Inducing Probabilistic CCG Grammars from Logical Form with Higher-Order Unification
8 0.41291517 82 emnlp-2010-Multi-Document Summarization Using A* Search and Discriminative Learning
9 0.41209993 57 emnlp-2010-Hierarchical Phrase-Based Translation Grammars Extracted from Alignment Posterior Probabilities
10 0.41188204 78 emnlp-2010-Minimum Error Rate Training by Sampling the Translation Lattice
11 0.41182059 107 emnlp-2010-Towards Conversation Entailment: An Empirical Investigation
12 0.41100627 86 emnlp-2010-Non-Isomorphic Forest Pair Translation
13 0.40933004 69 emnlp-2010-Joint Training and Decoding Using Virtual Nodes for Cascaded Segmentation and Tagging Tasks
14 0.40834066 67 emnlp-2010-It Depends on the Translation: Unsupervised Dependency Parsing via Word Alignment
15 0.40365642 58 emnlp-2010-Holistic Sentiment Analysis Across Languages: Multilingual Supervised Latent Dirichlet Allocation
16 0.4019298 120 emnlp-2010-What's with the Attitude? Identifying Sentences with Attitude in Online Discussions
17 0.40190688 49 emnlp-2010-Extracting Opinion Targets in a Single and Cross-Domain Setting with Conditional Random Fields
18 0.4014914 87 emnlp-2010-Nouns are Vectors, Adjectives are Matrices: Representing Adjective-Noun Constructions in Semantic Space
19 0.40091792 7 emnlp-2010-A Mixture Model with Sharing for Lexical Semantics
20 0.40005594 80 emnlp-2010-Modeling Organization in Student Essays