emnlp emnlp2012 emnlp2012-85 knowledge-graph by maker-knowledge-mining

85 emnlp-2012-Local and Global Context for Supervised and Unsupervised Metonymy Resolution

Source: pdf

Author: Vivi Nastase ; Alex Judea ; Katja Markert ; Michael Strube

Abstract: Computational approaches to metonymy resolution have focused almost exclusively on the local context, especially the constraints placed on a potentially metonymic word by its grammatical collocates. We expand such approaches by taking into account the larger context. Our algorithm is tested on the data from the metonymy resolution task (Task 8) at SemEval 2007. The results show that incorporation of the global context can improve over the use of the local context alone, depending on the types of metonymies addressed. As a second contribution, we move towards unsupervised resolution of metonymies, made feasible by considering ontological relations as possible readings. We show that such an unsupervised approach delivers promising results: it beats the supervised most frequent sense baseline and performs close to a supervised approach using only standard lexico-syntactic features.

Reference: text

Summary: the most important sentenses genereted by tfidf model

sentIndex sentText sentNum sentScore

1 org Abstract Computational approaches to metonymy resolution have focused almost exclusively on the local context, especially the constraints placed on a potentially metonymic word by its grammatical collocates. [sent-11, score-1.036]

2 Our algorithm is tested on the data from the metonymy resolution task (Task 8) at SemEval 2007. [sent-13, score-0.622]

3 The results show that incorporation of the global context can improve over the use of the local context alone, depending on the types of metonymies addressed. [sent-14, score-0.365]

4 As a second contribution, we move towards unsupervised resolution of metonymies, made feasible by considering ontological relations as possible readings. [sent-15, score-0.255]

5 We show that such an unsupervised approach delivers promising results: it beats the supervised most frequent sense baseline and performs close to a supervised approach using only standard lexico-syntactic features. [sent-16, score-0.122]

6 1 Introduction With the exception of explicit tasks in metonymy and metaphor analysis, computational treatment of language relies on the assumption that the texts to be processed have a literal interpretation. [sent-17, score-0.586]

7 This contrasts with the fact that figurative expressions are common in language, as exemplified by the metonymy in the excerpt from a Wikipedia article in Example 1 and another in Example 2 from the SemEval 2007 metonymy resolution task (Markert and Nissim, 2009). [sent-18, score-1.209]

8 Hence, in a metonymy a potentially metonymic expression or word (here Canada) stands for a conceptually related entity (here, people of Canada). [sent-22, score-0.8]

9 Textual entailment may need metonymy resolution (Bentivogli et al. [sent-25, score-0.622]

10 Leveling and Hartrumpf (2008) show that metonymy recognition on location proper names helps geographical information retrieval by excluding metonymically used place names from consideration (such as Example 1 or the use of Vietnam for the Vietnam war). [sent-30, score-0.516]

11 Metonymies also frequently interact with anaphora resolution (Nunberg, 1995; Markert and Hahn, 2002), as in Example 1 where the metonymic use of Canada is referred to by a plural pronoun afterward (their). [sent-31, score-0.337]

12 However, this approach needs novel, manual provision of readings as well as annotated examples for each new semantic class. [sent-36, score-0.142]

13 In contrast, we will see readings as relations between the potentially metonymic word (PMW) and other concepts in a large concept network, a priori allowing all possible relations as readings. [sent-37, score-0.708]

14 We base this approach on the observation that metonymic words stand in for concepts that they are related with e. [sent-38, score-0.302]

15 These readings are obtained on the fly and are therefore independent of manually — — – provided, preclassified interpretations or semantic classes, leading eventually to the possibility of unsupervised metonymy resolution. [sent-41, score-0.725]

16 Then we extract from a large concept network derived from Wikipedia the relations surrounding the PMW. [sent-43, score-0.309]

17 Let us consider the indications for metonymic readings and its interpretation in Example 1, on the one hand, and Example 2, on the other hand. [sent-47, score-0.409]

18 In Example 1, the grammatical relation to the verb defeat and the verb’s selectional preferences indicate the metonymy. [sent-48, score-0.407]

19 We will call all such grammatically related words and the grammatical relations the local context of the PMW. [sent-49, score-0.361]

20 In these examples, the wider context (database, key184 word) is a better indication for a metonymy but has not been satisfactorily integrated in prior approaches (see Section 2). [sent-53, score-0.561]

21 We here call all words surrounding the PMW but not grammatically related to it the global context. [sent-54, score-0.166]

22 In our approach we integrate both the local and the global context in our probabilistic framework. [sent-55, score-0.171]

23 In the next step we compute probabilities of the global context surrounding the PMWs under each (locally required) abstraction, and combine this with the selectional preferences of the grammatically related – words. [sent-57, score-0.424]

24 We compute the probabilities of the relations (= readings) between the concept corresponding to the PMW and its directly related concepts. [sent-59, score-0.191]

25 – – — — 2 Related Work The word sense disambiguation setting for metonymy resolution as developed by Nissim and Markert (2005) and used for the SemEval 2007 task (Markert and Nissim, 2009) uses a small, prespecified number of frequently occurring readings. [sent-62, score-0.654]

26 We believe that might be due to the lack of a link between the local and global context in these approaches in our work, we condition the global context on the abstractions and selectional preferences yielded by the local context and achieve better results. [sent-67, score-0.756]

27 (2000), Lapata (2003) propose a probabilistic model for finding the correct interpretation of such metonymies in an unsupervised manner. [sent-73, score-0.244]

28 However, these event type metonymies differ from the problem dealt with in our paper and the SemEval 2007 task in that their recognition (i. [sent-74, score-0.149]

29 their distinction from literal occurrences) is achieved simply by grammatical patterns (a noun instead of a gerund or to-infinitive following the verb) and the problem is limited to interpretation. [sent-76, score-0.159]

30 Our view of relations in a concept network being the interpretations of metonymies is strongly reminiscent of older work in metonymy resolution such as Hobbs et al. [sent-77, score-1.089]

31 , 1993; Pustejovsky, 1991 ; Harabagiu, 1998) rely on the view that metonymies violate selectional restrictions in their immediate, local context, usually those 1Brun et al. [sent-81, score-0.356]

32 (2007) is semi-supervised but again relies local grammatical context. [sent-82, score-0.158]

33 As can be seen in the Example 2, this misses metonymies which do not violate selectional restrictions. [sent-84, score-0.287]

34 Nastase and Strube (2009) use more flexible probabilistic selectional preferences instead of strict constraint violations as well as WordNet as a larger taxonomy but are also restricted to the local context. [sent-85, score-0.282]

35 Markert and Hahn (2002) do propose a treatment of metonymies that takes into account the larger discourse in the form of anaphoric relations between a metonymy and the prior context. [sent-86, score-0.738]

36 In addition, their framework uses a strict rule-based ranking of competing readings that cannot be easily extended. [sent-88, score-0.117]

37 The work presented here also relies on a concept network, built automatically from Wikipedia. [sent-89, score-0.118]

38 This resource provides us with links between entities in the text, and also a variety of ontological relations for the PMW, that will allow us to identify a wide variety of metonymic interpretations. [sent-90, score-0.37]

39 Our approach combines information from the concept network with automatically acquired selectional preferences as well as a possibility to combine in a probabilistic framework the influence of the local and global context on the interpretation of a potentially metonymic word. [sent-91, score-0.913]

40 3 The Approach The approach we present takes into account both the local, grammatical, context and the larger textual context of a potentially metonymic word. [sent-92, score-0.346]

41 On the one hand, the word/term to be interpreted (the potentially metonymic word/term PMW) is mapped onto a concept in the concept network (Section 3. [sent-94, score-0.586]

42 3), which gives us access to the conceptual relations (Ri) between the PMW and other concepts (cx ∈ CRi). [sent-95, score-0.243]

43 On the other hand, any word w gram– matically related to the PMW via a grammatical relation r provides us with semantic restrictions on the interpretation of the PMW, namely preferred semantic classes Aj (we call them abstractions) and a selectional preference score. [sent-96, score-0.421]

44 2 These are automatically 2We restrict the grammatical context that provides selectional preferences to verbs or adjectives grammatically related r A w . [sent-97, score-0.432]

45 Because the abstractions Aj and the PMW’s related concepts (cx) come from the same structured resource, we can compute the probabilities for each Ri given the grammatically reel partoebda wbiolirtdie w aonrd e atchhe grammatical relation r. [sent-101, score-0.425]

46 The global context can also easily be added to the computation, as the probability of each word in the context relative to an abstraction Aj can be computed through the resource’s is a hierarchy and its link to Wikipedia articles. [sent-102, score-0.243]

47 1 A concept network obtained from Wikipedia We use a Wikipedia article dump (January 2011) which provided over 3. [sent-106, score-0.283]

48 This partly structured repository is transformed into a large-scale multilingual concept network, whose nodes are concepts corresponding to articles or categories in Wikipedia (Nastase et al. [sent-108, score-0.228]

49 Concepts in this network are connected through a variety of semantic relations (e. [sent-110, score-0.192]

50 WikiNet is used here as a concept inventory, and its links and structure to generalize more specific concepts identified in texts to general concepts. [sent-115, score-0.189]

51 2 Selectional preferences and abstractions To compute selectional preferences we use the set of English Wikipedia articles, which describe specific concepts. [sent-119, score-0.41]

52 We exploit these links to gather concept arguments for verbs and adjectives, and generalize these using the concept network built from Wikipedia. [sent-125, score-0.33]

53 Essentially, we aimed to determine a small set of (more general) concepts that describe the set of collocates for a word w and grammatical relation r. [sent-137, score-0.272]

54 Starting from the concept collocates gathered, we go upwards following WikiNet’s is a links, and for each node found that covers at least N concept collocates (N is a parameter, N=2 in the experiments presented here), the MDL score of the node is computed (Algorithm 2). [sent-138, score-0.344]

55 The disjoint set of nodes that has the lowest overall MDL score is chosen (Γ), and for each node in this cut (which we call abstraction), we compute the selectional preference score, based on the number of concepts it dominates. [sent-140, score-0.209]

56 – – – – – Determining abstractions and selectional preferences leads to the following defeat information6: nsubjMartial obj artists (118977183) – 0. [sent-150, score-0.414]

57 3 Linking the PMW to the concept network In our environment, linking the PMW to the concept network is equivalent to finding its correspond- ing concept in our ontology, WikiNet. [sent-164, score-0.569]

58 We see this corresponding concept as the literal reading of the PMW. [sent-165, score-0.223]

59 In our particular setting, where we use the metonymy data from SemEval 2007, the domain of the PMW is well defined: locations and companies, respectively. [sent-168, score-0.516]

60 Using these constraints, finding the corresponding Wikipedia articles is much simplified, by using the category hierarchy and constraining the concepts to fall under the Geography and Companies categories respectively. [sent-169, score-0.141]

61 4 – Scoring conceptual relations with local and global context We work under the assumption that the concept corresponding to the PMW is related to the possible interpretations through a semantic relation, in particu- lar one that is captured in the concept network. [sent-174, score-0.637]

62 After 6The format is: Concept name (Concept Id) selectional preference score. [sent-175, score-0.138]

63 Table 1: Example conceptual relations establishing the connection to the resource by linking the PMW to the concept cPMW corresponding to its literal interpretation (see Section 3. [sent-186, score-0.472]

64 3), we extract the relations in which it is involved (Ri, i = 1, k), athned rtehlea concepts ith i csh hc oitn insec intveod lvtoe through these relations (CRi = {cx | (cPMWRicx) }). [sent-187, score-0.217]

65 We are interested in computing the likelihood of a conceptual relation being the correct interpreta- tion of a PMW, given its local and global context p(Ri |Cont, w, r). [sent-190, score-0.328]

66 1 The local context The local context considered in this work are all grammatically related verbs and adjectives w and their associated grammatical relation r. [sent-193, score-0.46]

67 2) provides the set of abstractions corresponding to the grammatically related word w and grammatical relation r: Aj ,j = 1, n. [sent-195, score-0.354]

68 Remember that these are local context constraints on the interpretation of the PMW. [sent-196, score-0.175]

69 2 The global context The abstractions obtained before are concepts. [sent-204, score-0.224]

70 We extract all nodes in the network subsumed by these concepts, and their corresponding articles in Wikipedia (if they have one). [sent-205, score-0.156]

71 This produces “abstraction-specific” article sets, based on which we compute the probability of the global context of a PMW for each abstraction. [sent-206, score-0.173]

72 As the SemEval 2007 data gives metonymic examples in a three-sentence context we use all the words in the 3 sentences as our global context. [sent-210, score-0.333]

73 Although the latter is our ultimate goal, to allow comparison with related work from the metonymy resolution task (Task 8) at SemEval 2007, we first investigate the supervised set-up. [sent-215, score-0.666]

74 1 Data We use the data from the metonymy resolution task (Task 8) at SemEval 2007. [sent-219, score-0.622]

75 The training-test division was achieved randomly so that the test data can have metonymic readings for which no training data exists, showing again the limitations of a supervised approach of prespecified readings. [sent-222, score-0.424]

76 Grammatical features The features used by Nissim and Markert (2005), and commonly used for the supervised classification of metonymy readings (Markert and Nissim, 2009): • grammatical role of PMW (subj, obj, . [sent-223, score-0.766]

77 ); grammatical number of PMW (sg, pl); number of grammatical roles in which the nPMumWbe appears ainm imts current context; number of words in PMW. [sent-232, score-0.178]

78 190 Semantic relations as features To evaluate the proposed approach we use the PMW’s conceptual relations as features. [sent-248, score-0.245]

79 For the “countries” portion of the data this adds 109 semantic relation features, and for companies 29 features. [sent-250, score-0.165]

80 Nastase and Strube (2009) added WordNet supersenses as features, and their values are selectional preferences computed with reference to WordNet. [sent-256, score-0.213]

81 These are similar to our abstractions, which in our approach serve to link the local and the global context to the ontological relations, but do not appear as features. [sent-257, score-0.247]

82 It is interesting to note that, in our setting, the global context is more useful than the local – 9We show the best result for each category, not necessarily from the overall best performing system. [sent-261, score-0.171]

83 Combining local and global evidence improves over both, indicating that the information they provide is not redundant. [sent-283, score-0.126]

84 Such a fine grained distinction between different professions for people in a country is not necessary, or indeed, desirable, for the metonymy resolution task. [sent-289, score-0.66]

85 3 Simulating unsupervised metonymy resolution In an unsupervised metonymy resolution approach, we would assign as interpretation the conceptual relation whose probability given the PMW, global and local contexts is highest. [sent-309, score-1.656]

86 To simulate then the unsupervised metonymy resolution task, we make the relation features (used in the supervised approach) binary, where for each instance the relation that has highest probability has the value 1, the others 0. [sent-310, score-0.816]

87 Using only the relation features simulates an unsupervised approach this set-up learns a mapping between the relations used as features and the metonymy classes in the data used. [sent-311, score-0.706]

88 As expected the results are lower, but still close to the supervised method when using only grammatical features (M&N;) for the location setting. [sent-313, score-0.133]

89 One feature that contributes greatly to the results, especially for the company semantic class, is the grammatical role of the PMW, but we could not incorporate this in the unsupervised setting. [sent-315, score-0.193]

90 The results in the simulated unsupervised setting indicate that relations are a viable substitute for manually provided classes in an unsupervised framework, while leaving space for improvement. [sent-316, score-0.166]

91 – 5 Conclusion We have explored the usage of local and global context for the task of metonymy resolution in a prob- abilistic framework. [sent-317, score-0.793]

92 The global context has been rarely used for the task of determining the intended reading of a potentially metonymic word (PMW) in context. [sent-318, score-0.393]

93 We rely on automatically computed selectional preferences, extracted from a corpus of Wikipedia articles, and generalized based on a concept network also extracted from Wikipedia. [sent-319, score-0.35]

94 The method described here is also a step towards the unsupervised resolution of metonymic words in context, by taking into account knowledge about the concept corresponding to the literal interpretation of the PMW, and its relations to other concepts. [sent-321, score-0.693]

95 This framework would also allow for exploring the metonymy resolution phenomena in various languages (since Wikipedia and WikiNet are multilingual), and investigate whether the same relations apply or different languages have different metonymic patterns. [sent-322, score-0.926]

96 XRCE-M: A hybrid system for named entity metonymy resolution. [sent-332, score-0.516]

97 Combining collocations, lexical and encyclopedic knowledge for metonymy resolution. [sent-378, score-0.516]

98 UTD-HLT-CG: Semantic architecture for metonymy resolution and classification of nominal relations. [sent-386, score-0.622]

99 Unsupervised learning of selectional restrictions and detection of argument coercions. [sent-407, score-0.138]

100 Sense-based interpretation of logical metonymy using a statistical method. [sent-411, score-0.577]

similar papers computed by tfidf model

tfidf for this paper:

wordName wordTfidf (topN-words)

[('metonymy', 0.516), ('pmw', 0.394), ('aj', 0.305), ('metonymic', 0.231), ('markert', 0.158), ('cont', 0.152), ('metonymies', 0.149), ('selectional', 0.138), ('abstractions', 0.122), ('nissim', 0.122), ('concept', 0.118), ('readings', 0.117), ('wikinet', 0.116), ('resolution', 0.106), ('cx', 0.102), ('conceptual', 0.099), ('network', 0.094), ('semeval', 0.094), ('grammatical', 0.089), ('nastase', 0.088), ('grammatically', 0.085), ('companies', 0.082), ('mdl', 0.082), ('wikipedia', 0.077), ('preferences', 0.075), ('relations', 0.073), ('article', 0.071), ('concepts', 0.071), ('literal', 0.07), ('local', 0.069), ('interpretation', 0.061), ('hi', 0.061), ('relation', 0.058), ('hyperlink', 0.058), ('global', 0.057), ('collocates', 0.054), ('pmws', 0.054), ('wkn', 0.054), ('hyperlinks', 0.053), ('strube', 0.053), ('ri', 0.052), ('wl', 0.052), ('june', 0.047), ('defeat', 0.047), ('katja', 0.046), ('company', 0.045), ('context', 0.045), ('supervised', 0.044), ('ontological', 0.042), ('cri', 0.041), ('fass', 0.041), ('hahn', 0.041), ('nicolae', 0.041), ('hobbs', 0.039), ('articles', 0.039), ('country', 0.038), ('leveling', 0.035), ('august', 0.035), ('countries', 0.035), ('reading', 0.035), ('link', 0.034), ('unsupervised', 0.034), ('canada', 0.034), ('interpretations', 0.033), ('fc', 0.033), ('pustejovsky', 0.032), ('vivi', 0.032), ('obj', 0.032), ('prespecified', 0.032), ('abstraction', 0.031), ('hierarchy', 0.031), ('uk', 0.029), ('stands', 0.028), ('visit', 0.027), ('july', 0.027), ('sanda', 0.027), ('administrator', 0.027), ('ggmbh', 0.027), ('ianlpguotr', 0.027), ('leeds', 0.027), ('othermet', 0.027), ('valetta', 0.027), ('yankees', 0.027), ('germany', 0.027), ('linking', 0.027), ('republic', 0.025), ('potentially', 0.025), ('classes', 0.025), ('semantic', 0.025), ('collocations', 0.025), ('surrounding', 0.024), ('resource', 0.024), ('evaluations', 0.024), ('team', 0.024), ('bentivogli', 0.023), ('vietnam', 0.023), ('malta', 0.023), ('malvina', 0.023), ('roberts', 0.023), ('subsumed', 0.023)]

similar papers list:

simIndex simValue paperId paperTitle

same-paper 1 1.0000001 85 emnlp-2012-Local and Global Context for Supervised and Unsupervised Metonymy Resolution

Author: Vivi Nastase ; Alex Judea ; Katja Markert ; Michael Strube

2 0.089141794 47 emnlp-2012-Explore Person Specific Evidence in Web Person Name Disambiguation

Author: Liwei Chen ; Yansong Feng ; Lei Zou ; Dongyan Zhao

Abstract: In this paper, we investigate different usages of feature representations in the web person name disambiguation task which has been suffering from the mismatch of vocabulary and lack of clues in web environments. In literature, the latter receives less attention and remains more challenging. We explore the feature space in this task and argue that collecting person specific evidences from a corpus level can provide a more reasonable and robust estimation for evaluating a feature’s importance in a given web page. This can alleviate the lack of clues where discriminative features can be reasonably weighted by taking their corpus level importance into account, not just relying on the current local context. We therefore propose a topic-based model to exploit the person specific global importance and embed it into the person name similarity. The experimental results show that the corpus level topic in- formation provides more stable evidences for discriminative features and our method outperforms the state-of-the-art systems on three WePS datasets.

3 0.084462039 35 emnlp-2012-Document-Wide Decoding for Phrase-Based Statistical Machine Translation

Author: Christian Hardmeier ; Joakim Nivre ; Jorg Tiedemann

Abstract: Independence between sentences is an assumption deeply entrenched in the models and algorithms used for statistical machine translation (SMT), particularly in the popular dynamic programming beam search decoding algorithm. This restriction is an obstacle to research on more sophisticated discourse-level models for SMT. We propose a stochastic local search decoding method for phrase-based SMT, which permits free document-wide dependencies in the models. We explore the stability and the search parameters ofthis method and demonstrate that it can be successfully used to optimise a document-level semantic language model. 1 Motivation In the field oftranslation studies, it is undisputed that discourse-wide context must be considered care- fully for good translation results (Hatim and Mason, 1990). By contrast, the state of the art in statistical machine translation (SMT), despite significant advances in the last twenty years, still assumes that texts can be translated sentence by sentence under strict independence assumptions, even though it is well known that certain linguistic phenomena such as pronominal anaphora cannot be translated correctly without referring to extra-sentential context. This is true both for the phrase-based and the syntaxbased approach to SMT. In the rest of this paper, we shall concentrate on phrase-based SMT. One reason why it is difficult to experiment with document-wide models for phrase-based SMT is that the dynamic programming (DP) algorithm 1179 which has been used almost exclusively for decoding SMT models in the recent literature has very strong assumptions of locality built into it. DP beam search for phrase-based SMT was described by Koehn et al. (2003), extending earlier work on word-based SMT (Tillmann et al., 1997; Och et al., 2001 ; Tillmann and Ney, 2003). This algorithm con- structs output sentences by starting with an empty hypothesis and adding output words at the end until translations for all source words have been generated. The core models of phrase-based SMT, in particular the n-gram language model (LM), only depend on a constant number of output words to the left of the word being generated. This fact is exploited by the search algorithm with a DP technique called hypothesis recombination (Och et al., 2001), which permits the elimination of hypotheses from the search space if they coincide in a certain number of final words with a better hypothesis and no future expansion can possibly invert the relative ranking of the two hypotheses under the given models. Hypothesis recombination achieves a substantial reduction of the search space without affecting search optimality and makes it possible to use aggressive pruning techniques for fast search while still obtaining good results. The downside of this otherwise excellent approach is that it only works well with models that have a local dependency structure similar to that of an n-gram language model, so they only depend on a small context window for each target word. Sentence-local models with longer dependencies can be added, but doing so greatly increases the risk for search errors by inhibiting hypothesis recombination. Cross-sentence dependencies cannot be directly integrated into DP SMT decoding in LParnogcue agdein Lgesa ornf tihneg, 2 p0a1g2e Jso 1in17t C9–o1n1f9e0re,n Jce ju on Is Elanmdp,ir Kicoarlea M,e 1t2h–o1d4s J iunly N 2a0tu1r2a.l ? Lc a2n0g1u2ag Aes Psorcoicaetsiosin fgo arn Cdo Cmopmutpauti oantiaoln Lailn Ngautiustriacls any obvious way, especially if joint optimisation of a number of interdependent decisions over an entire document is required. Research into models with a more varied, non-local dependency structure is to some extent stifled by the difficulty of decoding such models effectively, as can be seen by the problems some researchers encountered when they attempted to solve discourse-level problems. Consider, for instance, the work on cache-based language models by Tiedemann (2010) and Gong et al. (201 1), where error propagation was a serious issue, or the works on pronominal anaphora by Le Nagard and Koehn (2010), who implemented cross-sentence dependencies with an ad-hoc two-pass decoding strategy, and Hardmeier and Federico (2010) with the use of an external decoder driver to manage backward-only dependencies between sentences. In this paper, we present a method for decoding complete documents in phrase-based SMT. Our decoder uses a local search approach whose state consists of a complete translation of an entire document at any time. The initial state is improved by the application of a series of operations using a hill climbing strategy to find a (local) maximum of the score function. This setup gives us complete freedom to define scoring functions over the entire document. Moreover, by optionally initialising the state with the output of a traditional DP decoder, we can ensure that the final hypothesis is no worse than what would have been found by DP search alone. We start by describing the decoding algorithm and the state operations used by our decoder, then we present empirical results demonstrating the effectiveness of our approach and its usability with a document-level semantic language model, and finally we discuss some related work. 2 SMT Decoding by Hill Climbing In this section, we formally describe the phrasebased SMT model implemented by our decoder as well as the decoding algorithm we use. 2.1 SMT Model Our decoder is based on local search, so its state at any time is a representation of a complete translation of the entire document. Even though the decoder operates at the document level, it is important to keep 1180 track of sentence boundaries, and the individual operations that are applied to the state are still confined to sentence scope, so it is useful to decompose the state of a document into the state of its sentences, and we define the overall state S as a sequence of sentence states: S = S1S2 . . .SN, (1) where N is the number of sentences. This implies that we constrain the decoder to emit exactly one output sentence per input sentence. Let ibe the number of a sentence and mi the number of input tokens of this sentence, p and q (with 1 ≤ p ≤ q ≤ mi) be positions in the input sentence a1n ≤d [p; q] qde ≤no mte the set ofpositions from p up to and including q. We say that [p; q] precedes [p0; q0], or [p; q] ≺ [p0; q0], if q < p0. Let Φi([p; q]) be the set of t[pra;nqs]l ≺atio [pns for the source phrase covering positions [p; q] in the input sentence ias given by the phrase table. We call A = h[p; q] ,φi an anchored phrase pair w.it Wh coverage C(A) = [p; q] nif a φ ∈ Φi([p; q]) sise a target phrase translating =th [ep source w∈o Φrds at positions [p; q] . Then a sequence of ni anchored phrase pairs Si = A1A2 . . .Ani (2) is a valid sentence state for sentence iif the following two conditions hold: 1. The coverage sets C(Aj) for j in 1, . . . , ni are mutually disjoint, and 2. the anchored phrase pairs jointly cover the complete input sentence, or [niC(Aj) = [1;mi]. (3) [j=1 Let f(S) be a scoring function mapping a state S to a real number. As usual in SMT, it is assumed that the scoring function can be decomposed into a linear combination of K feature functions hk(S), each with a constant weight λk, so f(S) =k∑K=1λkhk(S). (4) The problem addressed by the decoder is the search for the state with maximal score, such that Sˆ Sˆ = argSmaxf(S). (5) The feature functions implemented in our baseline system are identical to the ones found in the popular Moses SMT system (Koehn et al., 2007). In particular, our decoder has the following feature functions: 1. phrase translation scores provided by the phrase table including forward and backward conditional probabilities, lexical weights and a phrase penalty (Koehn et al., 2003), 2. n-gram language model scores implemented with the KenLM toolkit (Heafield, 2011), 3. a word penalty score, 4. a distortion model with geometric (Koehn et al., 2003), and decay 5. a feature indicating the number of times a given distortion limit is exceeded in the current state. In our experiments, the last feature is used with a fixed weight of negative infinity in order to limit the gaps between the coverage sets of adjacent anchored phrase pairs to a maximum value. In DP search, the distortion limit is usually enforced directly by the search algorithm and is not added as a feature. In our decoder, however, this restriction is not required to limit complexity, so we decided to add it among the scoring models. 2.2 Decoding Algorithm The decoding algorithm we use (algorithm 1) is very simple. It starts with a given initial document state. In the main loop, which extends from line 3 to line 12, it generates a successor state S0 for the current state S by calling the function Neighbour, which non-deterministically applies one of the operations described in section 3 of this paper to S. The score of the new state is compared to that of the previous one. If it meets a given acceptance criterion, S0 becomes the current state, else search continues from the previous state S. For the experiments in this paper, we use the hill climbing acceptance criterion, which simply accepts a new state if its score is higher than that of the current state. Other acceptance criteria are possible and could be used to endow the search algorithm with stochastic behaviour. 1181 The main loop is repeated until a maximum number of steps (step limit) is reached or until a maximum number of moves are rejected in a row (rejection limit). Algorithm 1 Decoding algorithm Input: an initial document state S; search parameters maxsteps and maxrejected Output: a modified document state 1: nsteps ← 0 2: nrejected ← 0 3: nwrhejileec nsteps < maxsteps and nrejected < maxrejected do 4: S0 ← Neighbour (S) 5: if Accept (f(S0) , f(S)) then 6: S ← S0 7: nrejected ← 0 8: elsner 9: nrejected ← nrejected + 1 10: enndr eifj 11: nsteps ← nsteps + 1 12: ennds wtephsile ← 13: return S A notable difference between this algorithm and other hill climbing algorithms that have been used for SMT decoding (Germann et al., 2004; Langlais et al., 2007) is its non-determinism. Previous work for sentence-level decoding employed a steepest ascent strategy which amounts to enumerating the complete neighbourhood of the current state as defined by the state operations and selecting the next state to be the best state found in the neighbourhood of the current one. Enumerating all neighbours of a given state, costly as it is, has the advantage that it makes it easy to prove local optimality of a state by recognising that all possible successor states have lower scores. It can be rather inefficient, since at every step only one modification will be adopted; many of the modifications that are discarded will very likely be generated anew in the next iteration. As we extend the decoder to the document level, the size of the neighbourhood that would have to be explored in this way increases considerably. Moreover, the inefficiency of the steepest ascent approach potentially increases as well. Very likely, a promising move in one sentence will remain promising after a modification has been applied to another sentence, even though this is not guaranteed to be true in the presence of cross-sentence models. We therefore adopt a first-choice hill climbing strategy that non-deterministically generates successor states and accepts the first one that meets the acceptance criterion. This frees us from the necessity of generating the full set of successors for each state. On the downside, if the full successor set is not known, it is no longer possible to prove local optimality of a state, so we are forced to use a different condition for halting the search. We use a combination of two limits: The step limit is a hard limit on the resources the user is willing to expend on the search problem. The value of the rejection limit determines how much of the neighbourhood is searched for better successors before a state is accepted as a solution; it is related to the probability that a state returned as a solution is in fact locally optimal. To simplify notations in the description of the individual state operations, we write Si −→ Si0 (6) to signify that a state operation, when presented with a document state as in equation 1 and acting on sentence i, returns a new document state of S0 = S1 . . .Si−1 Si0 Si+1 . . .SN. (7) Similarly, Si : Aj . . .Aj+h−1 −→ A01 . . .A0h0 (8) is equivalent to Si −→ A1 . . .Aj−1 A01 . . .A0h0 Aj+h . . .Ani (9) and indicates that the operation returns a state in which a sequence of h consecutive anchored phrase pairs has been replaced by another sequence of h0 anchored phrase pairs. 2.3 Efficiency Considerations When implementing the feature functions for the decoder, we have to exercise some care to avoid recomputing scores for the whole document at every iteration. To achieve this, the scores are computed completely only once, at the beginning of the decoding run. In subsequent iterations, scoring functions are presented with the scores of the previous 1182 iteration and a list of modifications produced by the state operation, a set of tuples hi, r, s,A01 . . .A0h0i, each indicating tthioant ,t ahe s edto ocfu tmupelnets s hhio,ru,sld, Abe modifii,e eda as described by Si :Ar . . .As −→ A01 . . .A0h0 . (10) If a feature function is decomposable in some way, as all the standard features developed under the constraints of DP search are, it can then update the state simply by subtracting and adding score components pertaining to the modified parts of the document. Feature functions have the possibility to store their own state information along with the document state to make sure the required information is available. Thus, the framework makes it possible to exploit decomposability for efficient scoring without impos- ing any particular decomposition on the features as beam search does. To make scoring even more efficient, scores are computed in two passes: First, every feature function is asked to provide an upper bound on the score that will be obtained for the new state. In some cases, it is possible to calculate reasonable upper bounds much more efficiently than computing the exact feature value. If the upper bound fails to meet the acceptance criterion, the new state is discarded right away; if not, the full score is computed and the acceptance criterion is tested again. Among the basic SMT models, this two-pass strategy is only used for the n-gram LM, which requires fairly expensive parameter lookups for scoring. The scores of all the other baseline models are fully computed during the first scoring pass. The n-gram model is more complex. In its state information, it keeps track of the LM score and LM library state for each word. The first scoring pass then identifies the words whose LM scores are affected by the current search step. This includes the words changed by the search operation as well as the words whose LM history is modified. The range of the history de- pendencies can be determined precisely by considering the “valid state length” information provided by the KenLM library. In the first pass, the LM scores of the affected words are subtracted from the total score. The model only looks up the new LM scores for the affected words and updates the total score if the new search state passes the first acceptance check. This two-pass scoring approach allows us to avoid LM lookups altogether for states that will be rejected anyhow because of low scores from the other models, e. g. because the distortion limit is violated. Model score updates become more complex and slower as the number of dependencies of a model increases. While our decoding algorithm does not impose any formal restrictions on the number or type of dependencies that can be handled, there will be practical limits beyond which decoding becomes unacceptably slow or the scoring code becomes very difficult to maintain. These limits are however fairly independent of the types of dependencies handled by a model, which permits the exploration of more varied model types than those handled by DP search. 2.4 State Initialisation Before the hill climbing decoding algorithm can be run, an initial state must be generated. The closer the initial state is to an optimum, the less work remains to be done for the algorithm. If the algorithm is to be self-contained, initialisation must be relatively uninformed and can only rely on some general prior assumptions about what might be a good initial guess. On the other hand, if optimal results are sought after, it pays off to invest some effort into a good starting point. One way to do this is to run DP search first. For uninformed initialisation, we chose to implement a very simple procedure based only on the observation that, at least for language pairs involving the major European languages, it is usually a good guess to keep the word order of the output very similar to that of the input. We therefore create the initial state by selecting, for each sentence in the document, a sequence of anchored phrase pairs covering the input sentence in monotonic order, that is, such that for all pairs of adjacent anchored phrase pairs Aj and Aj+1, we have that C(Aj) ≺ C(Aj+1 ). For initialisation with DP search, we first run the Moses decoder (Koehn et al., 2007) with default search parameters and the same models as those used by our decoder. Then we extract the best output hypothesis from the search graph of the decoder and map it into a sequence of anchored phrase pairs in the obvious way. When the document-level decoder is used with models that are incompatible with beam search, Moses can be run with a subset of the models in order to find an approximation of the solution 1183 which is then refined with the complete feature set. 3 State Operations Given a document state S, the decoder uses a neighbourhood function Neighbour to simulate a move in the state space. The neighbourhood function nondeterministically selects a type of state operation and a location in the document to apply it to and returns the resulting new state. We use a set of three operations that has the property that every possible document state can be reached from every other state in a sequence of moves. Designing operations for state transitions in local search for phrase-based SMT is a problem that has been addressed in the literature (Langlais et al., 2007; Arun et al., 2010). Our decoder’s first- choice hill climbing strategy never enumerates the full neighbourhood of a state. We therefore place less emphasis than previous work on defining a compact neighbourhood, but allow the decoder to make quite extensive changes to a state in a single step with a certain probability. Otherwise our operations are similar to those used by Arun et al. (2010). All of the operations described in this paper make changes to a single sentence only. Each time it is called, the Neighbour function selects a sentence in the document with a probability proportional to the number of input tokens in each sentence to ensure a fair distribution ofthe decoder’s attention over the words in the document regardless of varying sentence lengths. 3.1 Changing Phrase Translations The change-phrase-translation operation replaces the translation of a single phrase with a random translation with the same coverage taken from the phrase table. Formally, the operation selects an anchored phrase pair Aj by drawing uniformly from the elements of Si and then draws a new translation φ0 uniformly from the set Φi(C(Aj)). The new state is given by Si : Aj −→ hC(Aj), φ0i. (11) 3.2 Changing Word Order The swap-phrases operation affects the output word order without changing the phrase translations. It exchanges two anchored phrase pairs Aj and Aj+h, resulting in an output state of Si : Aj . . .Aj+h −→ Aj+h Aj+1 . . .Aj+h−1 Aj. (12) The start location j is drawn uniformly from the eligible sentence positions; the swap range h comes from a geometric distribution with configurable decay. Other word-order changes such as a one-way move operation that does not require another movement in exchange or more advanced permutations can easily be defined. 3.3 Resegmentation The most complex operation is resegment, which allows the decoder to modify the segmentation ofthe source phrase. It takes a number of anchored phrase pairs that form a contiguous block both in the input and in the output and replaces them with a new set of phrase pairs covering the same span of the input sentence. Formally, Si : Aj . . .Aj+h−1 −→ A01 . . .A0h0 (13) such that j+[h−1 [h0 [ C(Aj0) = [ C(A0j0) = [p;q] j[0=j (14) j[0=1 for some p and q, where, for j0 = 1, . . . ,h0, we have that A0j0 = h[pj0; qj0] , φj0i, all [pj0; qj0] are mutually disjoint =an hd[ peach φj0 isi randomly drawn from Φi([pj0;qj0]). Regardless of the ordering of Aj . . .Aj+h−1 , the resegment operation always generates a sequence of anchored phrase pairs in linear order, such that C(A0j0) ≺ C(A0j0+1 ) for j0 = 1, . . . ,h0 −1 . As )f o≺r Cth(eA other operations, j is− generated uniformly and h is drawn from a geometric distribution with a decay parameter. The new segmentation is generated by extending the sequence of anchored phrase pairs with random elements starting at the next free position, proceeding from left to right until the whole range [p; q] is covered. 4 Experimental Results In this section, we present the results of a series of experiments with our document decoder. The 1184 goal of our experiments is to demonstrate the behaviour of the decoder and characterise its response to changes in the fundamental search parameters. The SMT models for our experiments were created with a subset of the training data for the English-French shared task at the WMT 2011workshop (Callison-Burch et al., 2011). The phrase table was trained on Europarl, news-commentary and UN data. To reduce the training data to a manageable size, singleton phrase pairs were removed before the phrase scoring step. Significance-based filtering (Johnson et al., 2007) was applied to the resulting phrase table. The language model was a 5gram model with Kneser-Ney smoothing trained on the monolingual News corpus with IRSTLM (Federico et al., 2008). Feature weights were trained with Minimum Error-Rate Training (MERT) (Och, 2003) on the news-test2008 development set using the DP beam search decoder and the MERT implementation of the Moses toolkit (Koehn et al., 2007). Experimental results are reported for the newstest2009 test set, a corpus of 111 newswire documents totalling 2,525 sentences or 65,595 English input tokens. 4.1 Stability An important difference between our decoder and the classical DP decoder as well as previous work in SMT decoding with local search is that our decoder is inherently non-deterministic. This implies that repeated runs of the decoder with the same search parameters, input and models will not, in general, find the same local maximum of the score space. The first empirical question we ask is therefore how different the results are under repeated runs. The results in this and the next section were obtained with random state initialisation, i. e. without running the DP beam search decoder. Figure 1 shows the results of 7 decoder runs with the models described above, translating the newstest2009 test set, with a step limit of 227 and a rejection limit of 100,000. The x-axis of both plots shows the number of decoding steps on a logarithmic scale, so the number of steps is doubled between two adjacent points on the same curve. In the left plot, the y-axis indicates the model score optimised by the decoder summed over all 2525 sentences of the document. In the right plot, the case-sensitive BLEU score (Papineni et al., 2002) of the current decoder Figure 1: Score stability in repeated decoder runs state against a reference translation is displayed. We note, as expected, that the decoder achieves a considerable improvement of the initial state with diminishing returns as decoding continues. Between 28 and 214 steps, the score increases at a roughly logarithmic pace, then the curve flattens out, which is partly due to the fact that decoding for some documents effectively stopped when the maximum number of rejections was reached. The BLEU score curve shows a similar increase, from an initial score below 5 % to a maximum of around 21.5 %. This is below the score of 22.45 % achieved by the beam search decoder with the same models, which is not surprising considering that our decoder approximates a more difficult search problem, from which a number of strong independence assumptions have been lifted, without, at the moment, having any stronger models at its disposal to exploit this additional freedom for better translation. In terms of stability, there are no dramatic differences between the decoder runs. Indeed, the small differences that exist are hardly discernible in the plots. The model scores at the end of the decoding run range between −158767.9 and −158716.9, a g re rlautniv rea ndgieffe breetnwceee nof − only a6b7.o9ut a n0d.0 −3 %15.8 F1i6n.a9l, BLEU scores range from 21.41 % to 21.63 %, an interval that is not negligible, but comparable to the variance observed when, e. g., feature weights from repeated MERT runs are used with one and the same SMT system. Note that these results were obtained with random state initialisation. With DP initialisation, score differences between repeated runs rarely 1185 exceed 0.02 absolute BLEU percentage points. Overall, we conclude that the decoding results of our algorithm are reasonably stable despite the nondeterminism inherent in the procedure. In our subsequent experiments, the evaluation scores reported are calculated as the mean of three runs for each experiment. 4.2 Search Algorithm Parameters The hill climbing algorithm we use has two parameters which govern the trade-off between decoding time and the accuracy with which a local maximum is identified: The step limit stops the search process after a certain number of steps regardless of the search progress made or lack thereof. The rejection limit stops the search after a certain number of unsuccessful attempts to make a step, when continued search does not seem to be promising. In most of our experiments, we used a step limit of 227 ≈ 1.3 · 108 and a rejection limit of 105. In practice, decoding terminates by reaching the rejection limit for the vast majority of documents. We therefore examined the effect of different rejection limits on the learning curves. The results are shown in figure 2. The results show that continued search does pay off to a certain extent. Indeed, the curve for rejection limit 107 seems to indicate that the model score increases roughly logarithmically, albeit to a higher base, even after the curve has started to flatten out at 214 steps. At a certain point, however, the probability of finding a good successor state drops rather sharply by about two orders of magnitude, as Figure 2: Search performance at different rejection limits evidenced by the fact that a rejection limit of 106 does not give a large improvement over one of 105, while one of 107 does. The continued model score improvement also results in an increase in BLEU scores, and with a BLEU score of 22. 1% the system with rejection limit 107 is fairly close to the score of 22.45 % obtained by DP beam search. Obviously, more exact search comes at a cost, and in this case, it comes at a considerable cost, which is an explosion of the time required to decode the test set from 4 minutes at rejection limit 103 to 224 minutes at rejection limit 105 and 38 hours 45 minutes at limit 107. The DP decoder takes 3 1 minutes for the same task. We conclude that the rejection limit of 105 selected for our experiments, while technically suboptimal, realises a good trade-off between decoding time and accuracy. 4.3 A Semantic Document Language Model In this section, we present the results of the application of our decoder to an actual SMT model with cross-sentence features. Our model addresses the problem of lexical cohesion. In particular, it rewards the use of semantically related words in the translation output by the decoder, where semantic distance is measured with a word space model based on Latent Semantic Analysis (LSA). LSA has been applied to semantic language modelling in previous research with some success (Coccaro and Jurafsky, 1998; Bellegarda, 2000; Wandmacher and Antoine, 2007). In SMT, it has mostly been used for domain adaptation (Kim and Khudanpur, 2004; Tam et al., 1186 2007), or to measure sentence similarities (Banchs and Costa-juss a`, 2011). The model we use is inspired by Bellegarda (2000). It is a Markov model, similar to a standard n-gram model, and assigns to each content word a score given a history of n preceding content words, where n = 30 below. Scoring relies on a 30dimensional LSA word vector space trained with the S-Space software (Jurgens and Stevens, 2010). The score is defined based on the cosine similarity between the word vector of the predicted word and the mean word vector of the words in the history, which is converted to a probability by histogram lookup as suggested by Bellegarda (2000). The model is structurally different from a regular n-gram model in that word vector n-grams are defined over content words occurring in the word vector model only and can cross sentence boundaries. Stop words, identified by an extensive stop word list and amounting to around 60 % of the tokens, are scored by a different mechanism based on their relative frequency (undiscounted unigram probability) in the training corpus. In sum, the score produced by the semantic document LM has the following form: wh(er|h)α=is tεpαheuncipgors(wp)o|hrtinof w fci os nakutneskotn wpon w ,onerldse,in(ls1teh5) training corpus and ε is a small fixed probability. It is integrated into the decoder as an extra feature function. Since we lack an automatic method for training the feature weights of document-wide features, its weight was selected by grid search over a number of values, comparing translation performance for the newstest2009 test set. In these experiments, we used DP beam search to initialise the state of our local search decoder. Three results are presented (table 1): The first table row shows the baseline performance using DP beam search with standard sentence-local features only. The scores in the second row were obtained by running the hill climbing decoder with DP initialisation, but without adding any models. A marginal increase in scores for all three test sets demonstrates that the hill climbing decoder manages to fix some of the search errors made by the DP search. The last row contains the scores obtained by adding in the semantic language model. Scores are presented for three publicly available test sets from recent WMT Machine Translation shared tasks, of which one (newstest2009) was used to monitor progress during development and select the final model. Adding the semantic language model results in a small increase in NIST scores (Doddington, 2002) for all three test sets as well as a small BLEU score gain (Papineni et al., 2002) for two out of three corpora. We note that the NIST score turned out to react more sensitively to improvements due to the semantic LM in all our experiments, which is reasonable because the model specifically targets content words, which benefit from the information weighting done by the NIST score. While the results we present do not constitute compelling evidence in favour of our semantic LM in its current form, they do suggest that this model could be improved to realise higher gains from cross-sentence semantic information. They support our claim that cross- sentence models should be examined more closely and that existing methods should be adapted to deal with them, a problem addressed by our main contribution, the local search document decoder. 5 Related Work Even though DP beam search (Koehn et al., 2003) has been the dominant approach to SMT decoding in recent years, methods based on local search have been explored at various times. For word-based SMT, greedy hill-climbing techniques were advo1187 cated as a faster replacement for beam search (Germann et al., 2001 ; Germann, 2003; Germann et al., 2004), and a problem formulation specifically targeting word reordering with an efficient word reordering algorithm has been proposed (Eisner and Tromble, 2006). A local search decoder has been advanced as a faster alternative to beam search also for phrasebased SMT (Langlais et al., 2007; Langlais et al., 2008). That work anticipates many of the features found in our decoder, including the use of local search to refine an initial hypothesis produced by DP beam search. The possibility of using models that do not fit well into the beam search paradigm is mentioned and illustrated with the example of a reversed n-gram language model, which the authors claim would be difficult to implement in a beam search decoder. Similarly to the work by Germann et al. (2001), their decoder is deterministic and explores the entire neighbourhood of a state in order to identify the most promising step. Our main contribution with respect to the work by Langlais et al. (2007) is the introduction of the possibility of handling document-level models by lifting the assumption of sentence independence. As a consequence, enumerating the entire neighbourhood becomes too expensive, which is why we resort to a “first-choice” strategy that non-deterministically generates states and accepts the first one encountered that meets the acceptance criterion. More recently, Gibbs sampling was proposed as a way to generate samples from the posterior distribution of a phrase-based SMT decoder (Arun et al., 2009; Arun et al., 2010), a process that resembles local search in its use of a set of state-modifying operators to generate a sequence of decoder states. Where local search seeks for the best state attainable from a given initial state, Gibbs sampling produces a representative sample from the posterior. Like all work on SMT decoding that we know of, the Gibbs sampler presented by Arun et al. (2010) assumes independence of sentences and considers the complete neighbourhood of each state before taking a sample. 6 Conclusion In the last twenty years of SMT research, there has been a strong assumption that sentences in a text newstest2009 newstest2010 newstest201 1 BLEU NIST BLEU NIST BLEU NIST 22.56 6.513 27.27 7.034 24.94 7.170 + hill climbing 22.60 6.518 27.33 7.046 24.97 7.169 with semantic LM 22.71 6.549 27.53 7.087 24.90 7.199 DP search only DP Table 1: Experimental results with a cross-sentence semantic language model are independent of one another, and discourse context has been largely neglected. Several factors have contributed to this. Developing good discourse-level models is difficult, and considering the modest translation quality that has long been achieved by SMT, there have been more pressing problems to solve and lower hanging fruit to pick. However, we argue that the popular DP beam search algorithm, which delivers excellent decoding performance, but imposes a particular kind of local dependency structure on the feature models, has also had its share in driving researchers away from discourse-level problems. In this paper, we have presented a decoding procedure for phrase-based SMT that makes it possible to define feature models with cross-sentence dependencies. Our algorithm can be combined with DP beam search to leverage the quality of the traditional approach with increased flexibility for models at the discourse level. We have presented preliminary results on a cross-sentence semantic language model addressing the problem of lexical cohesion to demonstrate that this kind of models is worth exploring further. Besides lexical cohesion, cross-sentence models are relevant for other linguistic phenomena such as pronominal anaphora or verb tense selection. We believe that SMT research has reached a point of maturity where discourse phenomena should not be ignored any longer, and we consider our decoder to be a step towards this goal. References Abhishek Arun, Chris Dyer, Barry Haddow, Phil Blunsom, Adam Lopez, and Philipp Koehn. 2009. Monte carlo inference and maximization for phrase-based translation. In Proceedings of the Thirteenth Conference on Computational Natural Language Learning (CoNLL-2009), pages 102–1 10, Boulder, Colorado, June. Association for Computational Linguistics. Abhishek Arun, Barry Haddow, Philipp Koehn, Adam Lopez, Chris Dyer, and Phil Blunsom. 2010. Monte 1188 Ma- Carlo techniques for phrase-based translation. chine translation, 24(2): 103–121 . Rafael E. Banchs and Marta R. Costa-juss a`. 2011. A semantic feature for Statistical Machine Translation. In Proceedings of Fifth Workshop on Syntax, Semantics and Structure in Statistical Translation, pages 126– 134, Portland, Oregon, USA, June. Association for Computational Linguistics. Jerome R. Bellegarda. 2000. Exploiting latent semantic information in statistical language modeling. Proceedings of the IEEE, 88(8): 1279–1296. Chris Callison-Burch, Philipp Koehn, Christof Monz, and Omar Zaidan. 2011. Findings of the 2011 Workshop on Statistical Machine Translation. In Proceedings of the Sixth Workshop on Statistical Machine Translation, pages 22–64, Edinburgh, Scotland, July. Association for Computational Linguistics. Noah Coccaro and Daniel Jurafsky. 1998. Towards better integration of semantic predictors in statistical language modeling. In Proceedings of the 5th International Conference on Spoken Language Processing, Sydney. George Doddington. 2002. Automatic evaluation of machine translation quality using n-gram co-occurrence statistics. In Proceedings of the second International conference on Human Language Technology Research, pages 138–145, San Diego. Jason Eisner and Roy W. Tromble. 2006. Local search with very large-scale neighborhoods for optimal permutations in machine translation. In Proceedings of the HLT-NAACL Workshop on Computationally Hard Problems and Joint Inference in Speech and Language Processing, pages 57–75. Marcello Federico, Nicola Bertoldi, and Mauro Cettolo. 2008. IRSTLM: an open source toolkit for handling large scale language models. In Interspeech 2008, pages 1618–1621 . ISCA. Ulrich Germann, Michael Jahr, Kevin Knight, Daniel Marcu, and Kenji Yamada. 2001 . Fast decoding and optimal decoding for machine translation. In Proceedings of 39th Annual Meeting of the Association for Computational Linguistics, pages 228–235, Toulouse, France, July. Association for Computational Linguis- tics. 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 Proceedings of the 2003 Human Language Technology Conference of the North American Chapter of the Association for Computational Linguistics. Zhengxian Gong, Min Zhang, and Guodong Zhou. 2011. Cache-based document-level Statistical Machine Translation. In Proceedings of the 2011 Conference on Empirical Methods in Natural Language Processing, pages 909–919, Edinburgh, Scotland, UK., July. Association for Computational Linguistics. Christian Hardmeier and Marcello Federico. 2010. Modelling Pronominal Anaphora in Statistical Machine Translation. In Proceedings of the seventh International Workshop on Spoken Language Translation (IWSLT), pages 283–289. Basil Hatim and Ian Mason. 1990. Discourse and the Translator. Language in Social Life Series. Longman, London. Kenneth Heafield. 2011. KenLM: faster and smaller language model queries. In Proceedings of the Sixth Workshop on Statistical Machine Translation, pages 187–197, Edinburgh, Scotland, July. Association for Computational Linguistics. Howard Johnson, Joel Martin, George Foster, and Roland Kuhn. 2007. Improving translation quality by discarding most of the phrasetable. In Proceedings of the 2007 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning (EMNLP-CoNLL), pages 967– 975, Prague, Czech Republic, June. Association for Computational Linguistics. David Jurgens and Keith Stevens. 2010. The S-Space package: An open source package for word space models. In Proceedings of the ACL 2010 System Demonstrations, pages 30–35, Uppsala, Sweden, July. Association for Computational Linguistics. Woosung Kim and Sanjeev Khudanpur. 2004. Crosslingual latent semantic analysis for language modeling. In IEEE international conference on acoustics, speech, and signal processing (ICASSP), volume 1, pages 257–260, Montr ´eal. Philipp Koehn, Franz Josef Och, and Daniel Marcu. 2003. Statistical phrase-based translation. In Proceedings of the 2003 conference of the North American chapter of the Association for Computational Linguistics on Human Language Technology, pages 48– 54, Edmonton. 1189 Philipp Koehn, Hieu Hoang, Alexandra Birch, et al. 2007. Moses: open source toolkit for Statistical Machine Translation. In Annual meeting of the Association for Computational Linguistics: Demonstration session, pages 177–180, Prague. Philippe Langlais, Alexandre Patry, and Fabrizio Gotti. 2007. A greedy decoder for phrase-based statistical machine translation. In TMI-2007: Proceedings of the 11th International Conference on Theoretical and Methodological Issues in Machine Translation, pages 104–1 13, Sk¨ ovde. Philippe Langlais, Alexandre Patry, and Fabrizio Gotti. 2008. Recherche locale pour la traduction statistique par segments. In TALN 2008, pages 119–128, Avignon, France, June. ATALA. Ronan Le Nagard and Philipp Koehn. 2010. Aiding pronoun translation with co-reference resolution. In Proceedings of the Joint Fifth Workshop on Statistical Machine Translation and MetricsMATR, pages 252–261, Uppsala, Sweden, July. Association for Computational Linguistics. Franz Josef Och, Nicola Ueffing, and Hermann Ney. 2001. An efficient A* search algorithm for Statistical Machine Translation. In Proceedings of the DataDriven Machine Translation Workshop, 39th Annual Meeting of the Association for Computational Linguistics (ACL), pages 55–62, Toulouse. Franz Josef Och. 2003. Minimum error rate training in Statistical Machine Translation. In Proceedings of the 41st annual meeting of the Association for Computational Linguistics, pages 160–167, Sapporo (Japan). Kishore Papineni, Salim Roukos, Todd Ward, and WeiJing Zhu. 2002. BLEU: a method for automatic evaluation of Machine Translation. In Proceedings of the 40th annual meeting of the Association for Computational Linguistics, pages 3 11–3 18, Philadelphia. ACL. Yik-Cheung Tam, Ian Lane, and Tanja Schultz. 2007. Bilingual LSA-based adaptation for Statistical Machine Translation. Machine Translation, 21(4): 187– 207. J o¨rg Tiedemann. 2010. To cache or not to cache? Experiments with adaptive models in Statistical Machine Translation. In Proceedings of the ACL 2010 Joint Fifth Workshop on Statistical Machine Translation and Metrics MATR, pages 189–194, Uppsala, Sweden. Association for Computational Linguistics. Christoph Tillmann and Hermann Ney. 2003. Word reordering and a Dynamic Programming beam search algorithm for Statistical Machine Translation. Computational linguistics, 29(1):97–133. Christoph Tillmann, Stephan Vogel, Hermann Ney, and Alex Zubiaga. 1997. A DP-based search using monotone alignments in Statistical Translation. In Proceedings of the 35th Annual Meeting of the Association for Computational Spain, tics. Linguistics, July. Association for 289–296, Madrid, Computational Linguis- pages Tonio Wandmacher and Jean-Yves Antoine. 2007. Methods to integrate a language model with semantic information for a word prediction component. In Proceedings of the 2007 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning (EMNLPCoNLL), pages 506–5 13, Prague, Czech Republic, June. Association for Computational Linguistics. 1190

4 0.068567425 92 emnlp-2012-Multi-Domain Learning: When Do Domains Matter?

Author: Mahesh Joshi ; Mark Dredze ; William W. Cohen ; Carolyn Rose

Abstract: We present a systematic analysis of existing multi-domain learning approaches with respect to two questions. First, many multidomain learning algorithms resemble ensemble learning algorithms. (1) Are multi-domain learning improvements the result of ensemble learning effects? Second, these algorithms are traditionally evaluated in a balanced class label setting, although in practice many multidomain settings have domain-specific class label biases. When multi-domain learning is applied to these settings, (2) are multidomain methods improving because they capture domain-specific class biases? An understanding of these two issues presents a clearer idea about where the field has had success in multi-domain learning, and it suggests some important open questions for improving beyond the current state of the art.

5 0.062418763 30 emnlp-2012-Constructing Task-Specific Taxonomies for Document Collection Browsing

Author: Hui Yang

Abstract: Taxonomies can serve as browsing tools for document collections. However, given an arbitrary collection, pre-constructed taxonomies could not easily adapt to the specific topic/task present in the collection. This paper explores techniques to quickly derive task-specific taxonomies supporting browsing in arbitrary document collections. The supervised approach directly learns semantic distances from users to propose meaningful task-specific taxonomies. The approach aims to produce globally optimized taxonomy structures by incorporating path consistency control and usergenerated task specification into the general learning framework. A comparison to stateof-the-art systems and a user study jointly demonstrate that our techniques are highly effective. .

6 0.05898137 40 emnlp-2012-Ensemble Semantics for Large-scale Unsupervised Relation Extraction

7 0.053846262 76 emnlp-2012-Learning-based Multi-Sieve Co-reference Resolution with Knowledge

8 0.053667698 110 emnlp-2012-Reading The Web with Learned Syntactic-Semantic Inference Rules

9 0.052097537 103 emnlp-2012-PATTY: A Taxonomy of Relational Patterns with Semantic Types

10 0.051638003 98 emnlp-2012-No Noun Phrase Left Behind: Detecting and Typing Unlinkable Entities

11 0.05091605 112 emnlp-2012-Resolving Complex Cases of Definite Pronouns: The Winograd Schema Challenge

12 0.050823037 84 emnlp-2012-Linking Named Entities to Any Database

13 0.05053135 80 emnlp-2012-Learning Verb Inference Rules from Linguistically-Motivated Evidence

14 0.050468449 71 emnlp-2012-Joint Entity and Event Coreference Resolution across Documents

15 0.048330184 3 emnlp-2012-A Coherence Model Based on Syntactic Patterns

16 0.047879424 113 emnlp-2012-Resolving This-issue Anaphora

17 0.04765404 136 emnlp-2012-Weakly Supervised Training of Semantic Parsers

18 0.045558371 97 emnlp-2012-Natural Language Questions for the Web of Data

19 0.042284593 62 emnlp-2012-Identifying Constant and Unique Relations by using Time-Series Text

20 0.039670892 93 emnlp-2012-Multi-instance Multi-label Learning for Relation Extraction

similar papers computed by lsi model

lsi for this paper:

topicId topicWeight

[(0, 0.142), (1, 0.094), (2, -0.013), (3, -0.035), (4, -0.01), (5, -0.031), (6, 0.04), (7, 0.084), (8, -0.008), (9, -0.024), (10, -0.019), (11, -0.018), (12, -0.056), (13, 0.068), (14, 0.093), (15, -0.053), (16, -0.089), (17, -0.073), (18, 0.023), (19, 0.179), (20, -0.0), (21, 0.04), (22, 0.067), (23, -0.055), (24, 0.02), (25, -0.046), (26, -0.032), (27, 0.146), (28, -0.019), (29, 0.084), (30, 0.033), (31, -0.209), (32, 0.098), (33, -0.04), (34, 0.11), (35, -0.22), (36, 0.157), (37, 0.046), (38, 0.043), (39, -0.124), (40, 0.035), (41, -0.015), (42, -0.204), (43, 0.009), (44, 0.143), (45, -0.04), (46, 0.089), (47, -0.133), (48, 0.096), (49, 0.07)]

similar papers list:

simIndex simValue paperId paperTitle

same-paper 1 0.93334335 85 emnlp-2012-Local and Global Context for Supervised and Unsupervised Metonymy Resolution

Author: Vivi Nastase ; Alex Judea ; Katja Markert ; Michael Strube

2 0.57161123 30 emnlp-2012-Constructing Task-Specific Taxonomies for Document Collection Browsing

Author: Hui Yang

3 0.40742639 47 emnlp-2012-Explore Person Specific Evidence in Web Person Name Disambiguation

Author: Liwei Chen ; Yansong Feng ; Lei Zou ; Dongyan Zhao

4 0.34548032 110 emnlp-2012-Reading The Web with Learned Syntactic-Semantic Inference Rules

Author: Ni Lao ; Amarnag Subramanya ; Fernando Pereira ; William W. Cohen

Abstract: We study how to extend a large knowledge base (Freebase) by reading relational information from a large Web text corpus. Previous studies on extracting relational knowledge from text show the potential of syntactic patterns for extraction, but they do not exploit background knowledge of other relations in the knowledge base. We describe a distributed, Web-scale implementation of a path-constrained random walk model that learns syntactic-semantic inference rules for binary relations from a graph representation of the parsed text and the knowledge base. Experiments show significant accuracy improvements in binary relation prediction over methods that consider only text, or only the existing knowledge base.

5 0.33819652 35 emnlp-2012-Document-Wide Decoding for Phrase-Based Statistical Machine Translation

Author: Christian Hardmeier ; Joakim Nivre ; Jorg Tiedemann

6 0.28882667 128 emnlp-2012-Translation Model Based Cross-Lingual Language Model Adaptation: from Word Models to Phrase Models

7 0.28251514 133 emnlp-2012-Unsupervised PCFG Induction for Grounded Language Learning with Highly Ambiguous Supervision

8 0.26690754 40 emnlp-2012-Ensemble Semantics for Large-scale Unsupervised Relation Extraction

9 0.26595563 103 emnlp-2012-PATTY: A Taxonomy of Relational Patterns with Semantic Types

10 0.25622678 44 emnlp-2012-Excitatory or Inhibitory: A New Semantic Orientation Extracts Contradiction and Causality from the Web

11 0.24871932 10 emnlp-2012-A Statistical Relational Learning Approach to Identifying Evidence Based Medicine Categories

12 0.24582171 32 emnlp-2012-Detecting Subgroups in Online Discussions by Modeling Positive and Negative Relations among Participants

13 0.24301022 113 emnlp-2012-Resolving This-issue Anaphora

14 0.23556113 84 emnlp-2012-Linking Named Entities to Any Database

15 0.23514156 62 emnlp-2012-Identifying Constant and Unique Relations by using Time-Series Text

16 0.23409669 2 emnlp-2012-A Beam-Search Decoder for Grammatical Error Correction

17 0.23185967 92 emnlp-2012-Multi-Domain Learning: When Do Domains Matter?

18 0.22794402 112 emnlp-2012-Resolving Complex Cases of Definite Pronouns: The Winograd Schema Challenge

19 0.21569607 136 emnlp-2012-Weakly Supervised Training of Semantic Parsers

20 0.21122308 86 emnlp-2012-Locally Training the Log-Linear Model for SMT

similar papers computed by lda model

lda for this paper:

topicId topicWeight

[(2, 0.016), (16, 0.032), (25, 0.017), (34, 0.064), (60, 0.079), (63, 0.051), (64, 0.021), (65, 0.044), (70, 0.021), (73, 0.012), (74, 0.036), (76, 0.037), (79, 0.424), (80, 0.021), (86, 0.024), (95, 0.014)]

similar papers list:

simIndex simValue paperId paperTitle

same-paper 1 0.79303861 85 emnlp-2012-Local and Global Context for Supervised and Unsupervised Metonymy Resolution

Author: Vivi Nastase ; Alex Judea ; Katja Markert ; Michael Strube

2 0.76043797 87 emnlp-2012-Lyrics, Music, and Emotions

Author: Rada Mihalcea ; Carlo Strapparava

Abstract: In this paper, we explore the classification of emotions in songs, using the music and the lyrics representation of the songs. We introduce a novel corpus of music and lyrics, consisting of 100 songs annotated for emotions. We show that textual and musical features can both be successfully used for emotion recognition in songs. Moreover, through comparative experiments, we show that the joint use of lyrics and music brings significant improvements over each of the individual textual and musical classifiers, with error rate reductions of up to 31%.

3 0.65791768 35 emnlp-2012-Document-Wide Decoding for Phrase-Based Statistical Machine Translation

Author: Christian Hardmeier ; Joakim Nivre ; Jorg Tiedemann

4 0.34873313 95 emnlp-2012-N-gram-based Tense Models for Statistical Machine Translation

Author: Zhengxian Gong ; Min Zhang ; Chew Lim Tan ; Guodong Zhou

Abstract: Tense is a small element to a sentence, however, error tense can raise odd grammars and result in misunderstanding. Recently, tense has drawn attention in many natural language processing applications. However, most of current Statistical Machine Translation (SMT) systems mainly depend on translation model and language model. They never consider and make full use of tense information. In this paper, we propose n-gram-based tense models for SMT and successfully integrate them into a state-of-the-art phrase-based SMT system via two additional features. Experimental results on the NIST Chinese-English translation task show that our proposed tense models are very effective, contributing performance improvement by 0.62 BLUE points over a strong baseline. 1

5 0.33908421 54 emnlp-2012-Forced Derivation Tree based Model Training to Statistical Machine Translation

Author: Nan Duan ; Mu Li ; Ming Zhou

Abstract: A forced derivation tree (FDT) of a sentence pair {f, e} denotes a derivation tree that can tpraainrsl {afte, f} idnetono itetss a acc duerraivtea target etrea tnhsaltat cioann e. In this paper, we present an approach that leverages structured knowledge contained in FDTs to train component models for statistical machine translation (SMT) systems. We first describe how to generate different FDTs for each sentence pair in training corpus, and then present how to infer the optimal FDTs based on their derivation and alignment qualities. As the first step in this line of research, we verify the effectiveness of our approach in a BTGbased phrasal system, and propose four FDTbased component models. Experiments are carried out on large scale English-to-Japanese and Chinese-to-English translation tasks, and significant improvements are reported on both translation quality and alignment quality.

6 0.32733014 118 emnlp-2012-Source Language Adaptation for Resource-Poor Machine Translation

7 0.32061207 108 emnlp-2012-Probabilistic Finite State Machines for Regression-based MT Evaluation

8 0.31958598 82 emnlp-2012-Left-to-Right Tree-to-String Decoding with Prediction

9 0.31414148 136 emnlp-2012-Weakly Supervised Training of Semantic Parsers

10 0.31103998 11 emnlp-2012-A Systematic Comparison of Phrase Table Pruning Techniques

11 0.30951524 71 emnlp-2012-Joint Entity and Event Coreference Resolution across Documents

12 0.30372328 2 emnlp-2012-A Beam-Search Decoder for Grammatical Error Correction

13 0.30024147 110 emnlp-2012-Reading The Web with Learned Syntactic-Semantic Inference Rules

14 0.30020061 124 emnlp-2012-Three Dependency-and-Boundary Models for Grammar Induction

15 0.29952654 123 emnlp-2012-Syntactic Transfer Using a Bilingual Lexicon

16 0.29951105 18 emnlp-2012-An Empirical Investigation of Statistical Significance in NLP

17 0.29929894 24 emnlp-2012-Biased Representation Learning for Domain Adaptation

18 0.29914194 14 emnlp-2012-A Weakly Supervised Model for Sentence-Level Semantic Orientation Analysis with Multiple Experts

19 0.29886967 23 emnlp-2012-Besting the Quiz Master: Crowdsourcing Incremental Classification Games

20 0.29869583 89 emnlp-2012-Mixed Membership Markov Models for Unsupervised Conversation Modeling