acl acl2010 acl2010-252 knowledge-graph by maker-knowledge-mining
Source: pdf
Author: Joel Tetreault ; Jennifer Foster ; Martin Chodorow
Abstract: Jennifer Foster NCLT Dublin City University Ireland j fo st er@ comput ing . dcu . ie Martin Chodorow Hunter College of CUNY New York, NY, USA martin . chodorow @hunter . cuny . edu We recreate a state-of-the-art preposition usage system (Tetreault and Chodorow (2008), henceWe evaluate the effect of adding parse features to a leading model of preposition us- age. Results show a significant improvement in the preposition selection task on native speaker text and a modest increment in precision and recall in an ESL error detection task. Analysis of the parser output indicates that it is robust enough in the face of noisy non-native writing to extract useful information.
Reference: text
sentIndex sentText sentNum sentScore
1 ie Martin Chodorow Hunter College of CUNY New York, NY, USA martin . [sent-4, score-0.041]
2 edu We recreate a state-of-the-art preposition usage system (Tetreault and Chodorow (2008), henceWe evaluate the effect of adding parse features to a leading model of preposition us- age. [sent-7, score-1.777]
3 Results show a significant improvement in the preposition selection task on native speaker text and a modest increment in precision and recall in an ESL error detection task. [sent-8, score-1.092]
4 Analysis of the parser output indicates that it is robust enough in the face of noisy non-native writing to extract useful information. [sent-9, score-0.171]
5 1 Introduction The task of preposition error detection has received a considerable amount of attention in recent years because selecting an appropriate preposition poses a particularly difficult challenge to learners of English as a second language (ESL). [sent-10, score-1.753]
6 It is not only ESL learners that struggle with English preposition usage automatically detecting preposition errors made by ESL speakers is a challenging task for NLP systems. [sent-11, score-1.648]
7 In this paper, we investigate the following research question: Are parser output features helpful in modeling preposition usage in well-formed text and learner text? [sent-15, score-1.125]
8 forth T&C08;) originally trained with lexical features and augment it with parser output features. [sent-16, score-0.279]
9 We employ the Stanford parser in our experiments because it consists of a competitive phrase structure parser and a constituent-to-dependency conversion tool (Klein and Manning, 2003a; Klein and Manning, 2003b; de Marneffe et al. [sent-17, score-0.402]
10 We compare the original model with the parser-augmented model on the tasks ofpreposition selection in wellformed text (fluent writers) and preposition error detection in learner texts (ESL writers). [sent-19, score-1.105]
11 This paper makes the following contributions: • • • 2 We demonstrate that parse features have a significant impact on preposition sreelsec htiaovne i na well-formed text. [sent-20, score-0.928]
12 We also show which features have the greatest effect on performance. [sent-21, score-0.073]
13 We show that, despite the noisiness of learner text, parse faet,at dueresps can actually mssa okfe l small, albeit non-significant, improvements to the performance of a state-of-the-art preposition error detection system. [sent-22, score-1.144]
14 We evaluate the accuracy of parsing and especially preposition arattcaych omfen pta irsni nlegar annedr texts. [sent-23, score-0.789]
15 (2008) describe very similar preposition error detection systems in which a model of correct prepositional usage is trained from wellformed text and a writer’s preposition is compared with the predictions of this model. [sent-25, score-1.876]
16 c C2o0n1f0er Aenscseoc Sihatoirotn P faopre Crso,m papguetsat 3io5n3a–l3 L5i8n,guistics but they achieve accuracy in a similar range. [sent-28, score-0.033]
17 Of these systems, only the DAPPER system (De Felice and Pulman, 2008; De Felice and Pulman, 2009; De Felice, 2009) uses a parser, the C&C; parser (Clark and Curran, 2007)), to determine the head and complement of the preposition. [sent-29, score-0.293]
18 De Felice and Pulman (2009) remark that the parser tends to be misled more by spelling errors than by grammatical errors. [sent-30, score-0.253]
19 The parser is fundamental to their system and they do not carry out a comparison of the use of a parser to determine the preposition’s attachments versus the use of shallower techniques. [sent-31, score-0.328]
20 T&C08;, on the other hand, reject the use of a parser because of the difficulties they foresee in applying one to learner data. [sent-32, score-0.223]
21 (2008) make only limited use of the Xerox Incremental Parser in their preposition error detection system. [sent-34, score-0.971]
22 They split the input sentence into the chunks before and after the preposition, and parse both chunks separately. [sent-35, score-0.147]
23 Only very shal- low analyses are extracted from the parser output because they do not trust the full analyses. [sent-36, score-0.192]
24 They also argue that a parser which is capable of distinguishing between arguments and adjuncts is useful for generating the correct preposition. [sent-38, score-0.172]
25 Next, we added the parse features to this model to create a new model “+Parse”. [sent-40, score-0.172]
26 2 we describe the parser output features used to augment the model. [sent-43, score-0.279]
27 We illustrate our features using the example phrase many local groups around the country. [sent-44, score-0.245]
28 1 shows the phrase structure tree and dependency triples returned by the Stanford parser for this phrase. [sent-46, score-0.283]
29 (2007) and T&C08; treat the tasks of preposition selection and error detection as a classification problem. [sent-49, score-1.02]
30 That is, given the context around a preposition and a model of correct usage, a classifier determines which of the 34 prepositions covered by the model is most appropriate for the context. [sent-50, score-0.924]
31 A model of correct preposition usage is constructed by training a Maximum Entropy classifier (Ratnaparkhi, 1998) on millions of preposition contexts from well-formed text. [sent-51, score-1.634]
32 So, for the phrase many local groups around the country, examples of lexical features for the preposition around include: FN = country, PN = groups, left-2-word-sequence = local-groups, and left-2-POS-sequence = JJ-NNS. [sent-55, score-1.042]
33 Combination T&C08; expand on the lexical feature set by combining the PV, PN and FN features, resulting in features such as PN-FN and PV-PN-FN. [sent-56, score-0.099]
34 POS and token versions of these features are employed. [sent-57, score-0.073]
35 The intuition behind creating combination features is that the Maximum Entropy classifier does not automatically model the interactions between individual features. [sent-58, score-0.123]
36 2 Parse Features To augment the above model we experimented with 14 features divided among five main classes. [sent-61, score-0.108]
37 Table 1 shows the features and their values for our around example. [sent-62, score-0.114]
38 The Preposition Head and Complement feature represents the two basic attachment relations of the preposition, i. [sent-63, score-0.108]
39 its head (what it is attached to) and its complement (what is attached to it). [sent-65, score-0.258]
40 Relation specifies the relation between the head and complement. [sent-66, score-0.078]
41 The Preposition Head and Complement Combined features are similar to the T&C08; Combination features except that they are extracted from parser output. [sent-67, score-0.316]
42 This mix of tags and tokens in a word-word dependency has proven to be an effective feature in sentiment analysis (Joshi and Penstein-Ros e´, 2009). [sent-69, score-0.066]
43 All the features described so far are extracted from the set of dependency triples output by the Stanford parser. [sent-70, score-0.193]
44 The final set of features (Phrase Structure), however, is extracted directly from the phrase structure trees themselves. [sent-71, score-0.151]
45 1, we compare the T&C08; and +Parse models on the task of preposition selection on well-formed texts written by native speakers. [sent-73, score-0.855]
46 For every preposition in the test set, we compare the system’s top preposition for that context to the writer’s preposition, and report accuracy rates. [sent-74, score-1.545]
47 The task here is slightly different - if the most likely preposition according to the model differs from the likelihood of the writer’s preposition by a certain threshold amount, a preposition error is flagged. [sent-77, score-2.373]
48 1 Native Speaker Test Data Our test set consists of 259K preposition events from the same source as the original training data. [sent-79, score-0.78]
49 2% and when the parse features are added, the +Parse model improves performance by more than 3% to 68. [sent-81, score-0.172]
50 1Prior research has shown preposition selection performance accuracy ranging from 65% to nearly 80%. [sent-84, score-0.838]
51 r125acy ture Addition Experiment Table 2 shows the effect of various feature class combinations on prediction accuracy. [sent-88, score-0.026]
52 The results are clear: a significant performance improvement is obtained on the preposition selection task when features from parser output are added. [sent-89, score-1.049]
53 The two best models in Table 2 contain parse features. [sent-90, score-0.099]
54 The table also shows that the non-parser-based feature classes are not entirely subsumed by the parse features but rather provide, to varying degrees, complementary information. [sent-91, score-0.217]
55 Having established the effectiveness of parse features, we investigate which parse feature classes contribute the most. [sent-92, score-0.224]
56 To test each contribution, we perform a feature addition experiment, separately adding features to the T&C08; model (see Table 3). [sent-93, score-0.099]
57 First, while there is overlapping information between the dependency features and the phrase structure features, the phrase structure features are making a contribution. [sent-95, score-0.3]
58 This is interesting because it suggests that a pure dependency parser might be less useful than a parser which explicitly produces both constituent and dependency information. [sent-96, score-0.378]
59 Second, using a parser to identify the prepo- sition head seems to be more useful than using it to identify the preposition complement. [sent-97, score-1.004]
60 2 Finally, as was the case for the T&C08; features, the combination parse features are also important (particularly the tag-tag or tag/token pairs). [sent-98, score-0.195]
61 2 ESL Test Data Our test data consists of 5,183 preposition events extracted from a set of essays written by non2De Felice (2009) observes the tem. [sent-100, score-0.801]
62 2e c21a5 l Table 4: ESL Error Detection Results native speakers for the Test of English as a Foreign Language (TOEFL? [sent-103, score-0.05]
63 annotators and checked by the authors using the preposition annotation scheme described in Tetreault and Chodorow (2008b). [sent-107, score-0.756]
64 4,881 of the prepositions were judged to be correct and the remaining 302 were judged to be incorrect. [sent-108, score-0.168]
65 The writer’s preposition is flagged as an error by the system if its likelihood according to the model satisfied a set of criteria (e. [sent-109, score-0.861]
66 , the difference between the probability of the system’s choice and the writer’s preposition is 0. [sent-111, score-0.756]
67 Unlike the selection task where we use accuracy as the metric, we use precision and recall with respect to error detection. [sent-113, score-0.187]
68 To date, performance figures that have been reported in the literature have been quite low, reflecting the difficulty of the task. [sent-114, score-0.026]
69 Table 4 shows the performance figures for the T&C08; and +Parse models. [sent-115, score-0.026]
70 Both precision and recall are higher for the +Parse model, however, given the low number of errors in our annotated test set, the difference is not statistically significant. [sent-116, score-0.038]
71 For each preposition we note whether the parser was successful in determining its head and complement. [sent-118, score-0.983]
72 The results for the three groups are shown in Table 5. [sent-119, score-0.074]
73 The figures in the first row are for correct prepositions and those in the second are for incorrect ones. [sent-120, score-0.126]
74 The parser tends to do a better job of determining the preposition’s complement than its head which is not surprising given the well-known problem of PP attachment ambiguity. [sent-121, score-0.395]
75 Given the preposition, the preceding noun, the preceding 356 Table 5: Parser Accuracy on Prepositions Sample of ESL Sentences in a verb and the following noun, Collins (1999) reports an accuracy rate of 84. [sent-122, score-0.153]
76 When confronted with the same information, the accuracy of three trained annotators is 88. [sent-124, score-0.033]
77 2% as an approximate PP-attachment upper bound, the Stanford parser appears to be doing a good job. [sent-127, score-0.149]
78 Comparing the results over the three sentence groups, its ability to identify the preposition’s head is quite robust to grammatical noise. [sent-128, score-0.103]
79 Preposition errors in isolation do not tend to mislead the parser: in the second group which contains sentences which are largely fluent apart from preposition errors, there is little difference between the parser’s accuracy on the correctly used prepositions and the incorrectly used ones. [sent-129, score-0.964]
80 6 Conclusion We have shown that the use of a parser can boost the accuracy of a preposition selection model tested on well-formed text. [sent-131, score-0.987]
81 In the error detection task, the improvement is less marked. [sent-132, score-0.215]
82 Nevertheless, examination of parser output shows the parse features can be extracted reliably from ESL data. [sent-133, score-0.364]
83 For our immediate future work, we plan to carry out the ESL evaluation on a larger test set to better gauge the usefulness of a parser in this context, to carry out a detailed error analysis to understand why certain parse features are effective and to explore a larger set of features. [sent-134, score-0.486]
84 In the longer term, we hope to compare different types of parsers in both the preposition selection and error detection tasks, i. [sent-135, score-1.043]
85 a task-based parser evaluation in the spirit of that carried out by Miyao et al. [sent-137, score-0.149]
86 (2008) on the task of protein pair interaction extraction. [sent-138, score-0.021]
87 We would like to further investigate the role of parsing in error detection by looking at other error types and other text types, e. [sent-139, score-0.32]
88 A classifier-based approach to preposition and determiner error correction in L2 english. [sent-161, score-0.861]
89 Using contextual speller techniques and language modelling for ESL error correction. [sent-186, score-0.126]
90 Using the web as a linguistic resource to automatically correct lexico-syntactic errors. [sent-190, score-0.023]
91 The ups and downs of preposition error detection in ESL writing. [sent-224, score-0.992]
92 Native Judgments of non-native usage: Experiments in preposition error detection. [sent-228, score-0.861]
wordName wordTfidf (topN-words)
[('preposition', 0.756), ('esl', 0.293), ('felice', 0.218), ('parser', 0.149), ('tetreault', 0.127), ('chodorow', 0.121), ('detection', 0.11), ('error', 0.105), ('parse', 0.099), ('rachele', 0.097), ('pulman', 0.085), ('writer', 0.082), ('attachment', 0.082), ('head', 0.078), ('prepositions', 0.077), ('groups', 0.074), ('features', 0.073), ('usage', 0.072), ('np', 0.069), ('complement', 0.066), ('joel', 0.066), ('pp', 0.059), ('phrase', 0.057), ('attached', 0.057), ('stanford', 0.053), ('learner', 0.053), ('marneffe', 0.052), ('native', 0.05), ('selection', 0.049), ('amod', 0.048), ('dapper', 0.048), ('hermet', 0.048), ('country', 0.048), ('de', 0.047), ('preceding', 0.047), ('noun', 0.046), ('manchester', 0.046), ('united', 0.045), ('hunter', 0.042), ('fluent', 0.041), ('around', 0.041), ('martin', 0.041), ('pn', 0.04), ('dependency', 0.04), ('cuny', 0.039), ('errors', 0.038), ('triples', 0.037), ('augment', 0.035), ('fn', 0.035), ('gamon', 0.034), ('judged', 0.034), ('accuracy', 0.033), ('wellformed', 0.032), ('pv', 0.03), ('writers', 0.03), ('miyao', 0.03), ('carry', 0.03), ('erroneous', 0.028), ('joshi', 0.027), ('classifier', 0.027), ('klein', 0.027), ('learners', 0.026), ('christopher', 0.026), ('figures', 0.026), ('verb', 0.026), ('feature', 0.026), ('grammatical', 0.025), ('chunks', 0.024), ('events', 0.024), ('typed', 0.024), ('combination', 0.023), ('correct', 0.023), ('judgments', 0.023), ('parsers', 0.023), ('prepositional', 0.022), ('output', 0.022), ('speaker', 0.022), ('downs', 0.021), ('trip', 0.021), ('klementiev', 0.021), ('sition', 0.021), ('cient', 0.021), ('noisiness', 0.021), ('ola', 0.021), ('mahesh', 0.021), ('misled', 0.021), ('speller', 0.021), ('rip', 0.021), ('marrekech', 0.021), ('modelsfor', 0.021), ('recreate', 0.021), ('protein', 0.021), ('foresee', 0.021), ('extracted', 0.021), ('manning', 0.021), ('hi', 0.021), ('stephen', 0.02), ('tends', 0.02), ('subsumed', 0.019), ('mislead', 0.019)]
simIndex simValue paperId paperTitle
same-paper 1 0.99999958 252 acl-2010-Using Parse Features for Preposition Selection and Error Detection
Author: Joel Tetreault ; Jennifer Foster ; Martin Chodorow
Abstract: Jennifer Foster NCLT Dublin City University Ireland j fo st er@ comput ing . dcu . ie Martin Chodorow Hunter College of CUNY New York, NY, USA martin . chodorow @hunter . cuny . edu We recreate a state-of-the-art preposition usage system (Tetreault and Chodorow (2008), henceWe evaluate the effect of adding parse features to a leading model of preposition us- age. Results show a significant improvement in the preposition selection task on native speaker text and a modest increment in precision and recall in an ESL error detection task. Analysis of the parser output indicates that it is robust enough in the face of noisy non-native writing to extract useful information.
2 0.14758414 120 acl-2010-Fully Unsupervised Core-Adjunct Argument Classification
Author: Omri Abend ; Ari Rappoport
Abstract: The core-adjunct argument distinction is a basic one in the theory of argument structure. The task of distinguishing between the two has strong relations to various basic NLP tasks such as syntactic parsing, semantic role labeling and subcategorization acquisition. This paper presents a novel unsupervised algorithm for the task that uses no supervised models, utilizing instead state-of-the-art syntactic induction algorithms. This is the first work to tackle this task in a fully unsupervised scenario.
3 0.10264599 203 acl-2010-Rebanking CCGbank for Improved NP Interpretation
Author: Matthew Honnibal ; James R. Curran ; Johan Bos
Abstract: Once released, treebanks tend to remain unchanged despite any shortcomings in their depth of linguistic analysis or coverage of specific phenomena. Instead, separate resources are created to address such problems. In this paper we show how to improve the quality of a treebank, by integrating resources and implementing improved analyses for specific constructions. We demonstrate this rebanking process by creating an updated version of CCGbank that includes the predicate-argument structure of both verbs and nouns, baseNP brackets, verb-particle constructions, and restrictive and non-restrictive nominal modifiers; and evaluate the impact of these changes on a statistical parser.
4 0.096410282 102 acl-2010-Error Detection for Statistical Machine Translation Using Linguistic Features
Author: Deyi Xiong ; Min Zhang ; Haizhou Li
Abstract: Automatic error detection is desired in the post-processing to improve machine translation quality. The previous work is largely based on confidence estimation using system-based features, such as word posterior probabilities calculated from Nbest lists or word lattices. We propose to incorporate two groups of linguistic features, which convey information from outside machine translation systems, into error detection: lexical and syntactic features. We use a maximum entropy classifier to predict translation errors by integrating word posterior probability feature and linguistic features. The experimental results show that 1) linguistic features alone outperform word posterior probability based confidence estimation in error detection; and 2) linguistic features can further provide complementary information when combined with word confidence scores, which collectively reduce the classification error rate by 18.52% and improve the F measure by 16.37%.
5 0.08900018 200 acl-2010-Profiting from Mark-Up: Hyper-Text Annotations for Guided Parsing
Author: Valentin I. Spitkovsky ; Daniel Jurafsky ; Hiyan Alshawi
Abstract: We show how web mark-up can be used to improve unsupervised dependency parsing. Starting from raw bracketings of four common HTML tags (anchors, bold, italics and underlines), we refine approximate partial phrase boundaries to yield accurate parsing constraints. Conversion procedures fall out of our linguistic analysis of a newly available million-word hyper-text corpus. We demonstrate that derived constraints aid grammar induction by training Klein and Manning’s Dependency Model with Valence (DMV) on this data set: parsing accuracy on Section 23 (all sentences) of the Wall Street Journal corpus jumps to 50.4%, beating previous state-of-the- art by more than 5%. Web-scale experiments show that the DMV, perhaps because it is unlexicalized, does not benefit from orders of magnitude more annotated but noisier data. Our model, trained on a single blog, generalizes to 53.3% accuracy out-of-domain, against the Brown corpus nearly 10% higher than the previous published best. The fact that web mark-up strongly correlates with syntactic structure may have broad applicability in NLP.
6 0.083472542 133 acl-2010-Hierarchical Search for Word Alignment
7 0.077681214 221 acl-2010-Syntax-to-Morphology Mapping in Factored Phrase-Based Statistical Machine Translation from English to Turkish
8 0.06733904 75 acl-2010-Correcting Errors in a Treebank Based on Synchronous Tree Substitution Grammar
9 0.065485328 114 acl-2010-Faster Parsing by Supertagger Adaptation
10 0.06452129 246 acl-2010-Unsupervised Discourse Segmentation of Documents with Inherently Parallel Structure
11 0.061019666 130 acl-2010-Hard Constraints for Grammatical Function Labelling
12 0.060399462 84 acl-2010-Detecting Errors in Automatically-Parsed Dependency Relations
13 0.055821702 99 acl-2010-Efficient Third-Order Dependency Parsers
14 0.055121798 36 acl-2010-Automatic Collocation Suggestion in Academic Writing
15 0.054206699 169 acl-2010-Learning to Translate with Source and Target Syntax
16 0.052737433 240 acl-2010-Training Phrase Translation Models with Leaving-One-Out
17 0.051982958 153 acl-2010-Joint Syntactic and Semantic Parsing of Chinese
18 0.051883873 195 acl-2010-Phylogenetic Grammar Induction
19 0.050058261 83 acl-2010-Dependency Parsing and Projection Based on Word-Pair Classification
20 0.049655017 115 acl-2010-Filtering Syntactic Constraints for Statistical Machine Translation
topicId topicWeight
[(0, -0.153), (1, -0.012), (2, 0.057), (3, 0.004), (4, -0.048), (5, -0.01), (6, 0.027), (7, 0.011), (8, 0.047), (9, 0.075), (10, 0.039), (11, 0.047), (12, -0.047), (13, 0.049), (14, 0.051), (15, 0.033), (16, 0.007), (17, 0.022), (18, 0.093), (19, 0.022), (20, -0.044), (21, 0.016), (22, 0.044), (23, 0.066), (24, 0.023), (25, -0.048), (26, 0.109), (27, -0.02), (28, -0.019), (29, 0.023), (30, 0.034), (31, 0.01), (32, 0.042), (33, -0.02), (34, 0.086), (35, -0.055), (36, -0.118), (37, -0.033), (38, -0.077), (39, 0.038), (40, 0.027), (41, 0.085), (42, -0.077), (43, -0.042), (44, -0.056), (45, 0.024), (46, 0.047), (47, 0.112), (48, 0.035), (49, -0.025)]
simIndex simValue paperId paperTitle
same-paper 1 0.94595248 252 acl-2010-Using Parse Features for Preposition Selection and Error Detection
Author: Joel Tetreault ; Jennifer Foster ; Martin Chodorow
Abstract: Jennifer Foster NCLT Dublin City University Ireland j fo st er@ comput ing . dcu . ie Martin Chodorow Hunter College of CUNY New York, NY, USA martin . chodorow @hunter . cuny . edu We recreate a state-of-the-art preposition usage system (Tetreault and Chodorow (2008), henceWe evaluate the effect of adding parse features to a leading model of preposition us- age. Results show a significant improvement in the preposition selection task on native speaker text and a modest increment in precision and recall in an ESL error detection task. Analysis of the parser output indicates that it is robust enough in the face of noisy non-native writing to extract useful information.
2 0.66413665 200 acl-2010-Profiting from Mark-Up: Hyper-Text Annotations for Guided Parsing
Author: Valentin I. Spitkovsky ; Daniel Jurafsky ; Hiyan Alshawi
Abstract: We show how web mark-up can be used to improve unsupervised dependency parsing. Starting from raw bracketings of four common HTML tags (anchors, bold, italics and underlines), we refine approximate partial phrase boundaries to yield accurate parsing constraints. Conversion procedures fall out of our linguistic analysis of a newly available million-word hyper-text corpus. We demonstrate that derived constraints aid grammar induction by training Klein and Manning’s Dependency Model with Valence (DMV) on this data set: parsing accuracy on Section 23 (all sentences) of the Wall Street Journal corpus jumps to 50.4%, beating previous state-of-the- art by more than 5%. Web-scale experiments show that the DMV, perhaps because it is unlexicalized, does not benefit from orders of magnitude more annotated but noisier data. Our model, trained on a single blog, generalizes to 53.3% accuracy out-of-domain, against the Brown corpus nearly 10% higher than the previous published best. The fact that web mark-up strongly correlates with syntactic structure may have broad applicability in NLP.
3 0.64304656 76 acl-2010-Creating Robust Supervised Classifiers via Web-Scale N-Gram Data
Author: Shane Bergsma ; Emily Pitler ; Dekang Lin
Abstract: In this paper, we systematically assess the value of using web-scale N-gram data in state-of-the-art supervised NLP classifiers. We compare classifiers that include or exclude features for the counts of various N-grams, where the counts are obtained from a web-scale auxiliary corpus. We show that including N-gram count features can advance the state-of-the-art accuracy on standard data sets for adjective ordering, spelling correction, noun compound bracketing, and verb part-of-speech disambiguation. More importantly, when operating on new domains, or when labeled training data is not plentiful, we show that using web-scale N-gram features is essential for achieving robust performance.
4 0.61771864 203 acl-2010-Rebanking CCGbank for Improved NP Interpretation
Author: Matthew Honnibal ; James R. Curran ; Johan Bos
Abstract: Once released, treebanks tend to remain unchanged despite any shortcomings in their depth of linguistic analysis or coverage of specific phenomena. Instead, separate resources are created to address such problems. In this paper we show how to improve the quality of a treebank, by integrating resources and implementing improved analyses for specific constructions. We demonstrate this rebanking process by creating an updated version of CCGbank that includes the predicate-argument structure of both verbs and nouns, baseNP brackets, verb-particle constructions, and restrictive and non-restrictive nominal modifiers; and evaluate the impact of these changes on a statistical parser.
5 0.57804608 114 acl-2010-Faster Parsing by Supertagger Adaptation
Author: Jonathan K. Kummerfeld ; Jessika Roesner ; Tim Dawborn ; James Haggerty ; James R. Curran ; Stephen Clark
Abstract: We propose a novel self-training method for a parser which uses a lexicalised grammar and supertagger, focusing on increasing the speed of the parser rather than its accuracy. The idea is to train the supertagger on large amounts of parser output, so that the supertagger can learn to supply the supertags that the parser will eventually choose as part of the highestscoring derivation. Since the supertagger supplies fewer supertags overall, the parsing speed is increased. We demonstrate the effectiveness of the method using a CCG supertagger and parser, obtain- ing significant speed increases on newspaper text with no loss in accuracy. We also show that the method can be used to adapt the CCG parser to new domains, obtaining accuracy and speed improvements for Wikipedia and biomedical text.
6 0.54394102 101 acl-2010-Entity-Based Local Coherence Modelling Using Topological Fields
7 0.53815323 130 acl-2010-Hard Constraints for Grammatical Function Labelling
8 0.49301338 102 acl-2010-Error Detection for Statistical Machine Translation Using Linguistic Features
9 0.48823467 19 acl-2010-A Taxonomy, Dataset, and Classifier for Automatic Noun Compound Interpretation
10 0.48312178 245 acl-2010-Understanding the Semantic Structure of Noun Phrase Queries
11 0.48162162 99 acl-2010-Efficient Third-Order Dependency Parsers
12 0.47634369 12 acl-2010-A Probabilistic Generative Model for an Intermediate Constituency-Dependency Representation
13 0.44930089 221 acl-2010-Syntax-to-Morphology Mapping in Factored Phrase-Based Statistical Machine Translation from English to Turkish
14 0.44829878 164 acl-2010-Learning Phrase-Based Spelling Error Models from Clickthrough Data
15 0.44500557 241 acl-2010-Transition-Based Parsing with Confidence-Weighted Classification
16 0.44394678 148 acl-2010-Improving the Use of Pseudo-Words for Evaluating Selectional Preferences
17 0.42790964 85 acl-2010-Detecting Experiences from Weblogs
18 0.42718387 139 acl-2010-Identifying Generic Noun Phrases
19 0.4216029 211 acl-2010-Simple, Accurate Parsing with an All-Fragments Grammar
20 0.41792503 120 acl-2010-Fully Unsupervised Core-Adjunct Argument Classification
topicId topicWeight
[(7, 0.012), (14, 0.02), (25, 0.095), (33, 0.017), (42, 0.021), (59, 0.091), (73, 0.065), (76, 0.018), (78, 0.041), (81, 0.251), (83, 0.111), (84, 0.036), (98, 0.096)]
simIndex simValue paperId paperTitle
1 0.87329167 108 acl-2010-Expanding Verb Coverage in Cyc with VerbNet
Author: Clifton McFate
Abstract: A robust dictionary of semantic frames is an essential element of natural language understanding systems that use ontologies. However, creating lexical resources that accurately capture semantic representations en masse is a persistent problem. Where the sheer amount of content makes hand creation inefficient, computerized approaches often suffer from over generality and difficulty with sense disambiguation. This paper describes a semi-automatic method to create verb semantic frames in the Cyc ontology by converting the information contained in VerbNet into a Cyc usable format. This method captures the differences in meaning between types of verbs, and uses existing connections between WordNet, VerbNet, and Cyc to specify distinctions between individual verbs when available. This method provides 27,909 frames to OpenCyc which currently has none and can be used to extend ResearchCyc as well. We show that these frames lead to a 20% increase in sample sentences parsed over the Research Cyc verb lexicon. 1
2 0.82369965 183 acl-2010-Online Generation of Locality Sensitive Hash Signatures
Author: Benjamin Van Durme ; Ashwin Lall
Abstract: Motivated by the recent interest in streaming algorithms for processing large text collections, we revisit the work of Ravichandran et al. (2005) on using the Locality Sensitive Hash (LSH) method of Charikar (2002) to enable fast, approximate comparisons of vector cosine similarity. For the common case of feature updates being additive over a data stream, we show that LSH signatures can be maintained online, without additional approximation error, and with lower memory requirements than when using the standard offline technique.
same-paper 3 0.80232638 252 acl-2010-Using Parse Features for Preposition Selection and Error Detection
Author: Joel Tetreault ; Jennifer Foster ; Martin Chodorow
Abstract: Jennifer Foster NCLT Dublin City University Ireland j fo st er@ comput ing . dcu . ie Martin Chodorow Hunter College of CUNY New York, NY, USA martin . chodorow @hunter . cuny . edu We recreate a state-of-the-art preposition usage system (Tetreault and Chodorow (2008), henceWe evaluate the effect of adding parse features to a leading model of preposition us- age. Results show a significant improvement in the preposition selection task on native speaker text and a modest increment in precision and recall in an ESL error detection task. Analysis of the parser output indicates that it is robust enough in the face of noisy non-native writing to extract useful information.
4 0.62270582 71 acl-2010-Convolution Kernel over Packed Parse Forest
Author: Min Zhang ; Hui Zhang ; Haizhou Li
Abstract: This paper proposes a convolution forest kernel to effectively explore rich structured features embedded in a packed parse forest. As opposed to the convolution tree kernel, the proposed forest kernel does not have to commit to a single best parse tree, is thus able to explore very large object spaces and much more structured features embedded in a forest. This makes the proposed kernel more robust against parsing errors and data sparseness issues than the convolution tree kernel. The paper presents the formal definition of convolution forest kernel and also illustrates the computing algorithm to fast compute the proposed convolution forest kernel. Experimental results on two NLP applications, relation extraction and semantic role labeling, show that the proposed forest kernel significantly outperforms the baseline of the convolution tree kernel. 1
5 0.62056839 211 acl-2010-Simple, Accurate Parsing with an All-Fragments Grammar
Author: Mohit Bansal ; Dan Klein
Abstract: We present a simple but accurate parser which exploits both large tree fragments and symbol refinement. We parse with all fragments of the training set, in contrast to much recent work on tree selection in data-oriented parsing and treesubstitution grammar learning. We require only simple, deterministic grammar symbol refinement, in contrast to recent work on latent symbol refinement. Moreover, our parser requires no explicit lexicon machinery, instead parsing input sentences as character streams. Despite its simplicity, our parser achieves accuracies of over 88% F1 on the standard English WSJ task, which is competitive with substantially more complicated state-of-theart lexicalized and latent-variable parsers. Additional specific contributions center on making implicit all-fragments parsing efficient, including a coarse-to-fine inference scheme and a new graph encoding.
6 0.62047404 101 acl-2010-Entity-Based Local Coherence Modelling Using Topological Fields
7 0.61794484 153 acl-2010-Joint Syntactic and Semantic Parsing of Chinese
8 0.61536187 158 acl-2010-Latent Variable Models of Selectional Preference
9 0.61474323 169 acl-2010-Learning to Translate with Source and Target Syntax
10 0.6126489 109 acl-2010-Experiments in Graph-Based Semi-Supervised Learning Methods for Class-Instance Acquisition
11 0.61194694 120 acl-2010-Fully Unsupervised Core-Adjunct Argument Classification
12 0.61017913 237 acl-2010-Topic Models for Word Sense Disambiguation and Token-Based Idiom Detection
13 0.60925496 230 acl-2010-The Manually Annotated Sub-Corpus: A Community Resource for and by the People
14 0.60871297 247 acl-2010-Unsupervised Event Coreference Resolution with Rich Linguistic Features
15 0.60856467 17 acl-2010-A Structured Model for Joint Learning of Argument Roles and Predicate Senses
16 0.60760856 128 acl-2010-Grammar Prototyping and Testing with the LinGO Grammar Matrix Customization System
17 0.60641241 23 acl-2010-Accurate Context-Free Parsing with Combinatory Categorial Grammar
18 0.60570037 162 acl-2010-Learning Common Grammar from Multilingual Corpus
19 0.60533166 261 acl-2010-Wikipedia as Sense Inventory to Improve Diversity in Web Search Results
20 0.60387969 130 acl-2010-Hard Constraints for Grammatical Function Labelling