Michael Collins and James Brooks. Prepositional phrase attachment through a backed-o# model. In Proceedings of the Third Workshop on Very Large Corpora, pages 27--38, Cambridge, 1995.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....has been very little if any work in the area of ambiguous CPs. In addition to developing an unsupervised CP disambiguation model, In [MG, in prep] we have developed two supervised models (one backed off and one maximum entropy) for determining CP attachment. The backed off model, closely based on [CB95] per forms at 75.6 accuracy. The reduction error from the unsupervised model presented here to the backed off model is 13 . This is compa rable to the 14.3 error reduction found when going from [AR98] to [CB95] It is interesting to note that after reducing the volume of training data by ....

....entropy) for determining CP attachment. The backed off model, closely based on [CB95] per forms at 75.6 accuracy. The reduction error from the unsupervised model presented here to the backed off model is 13 . This is compa rable to the 14.3 error reduction found when going from [AR98] to [CB95] It is interesting to note that after reducing the volume of training data by half there was no drop in accuracy. In fact, accuracy remained exactly the same as the volume of data was in creased from half to full. The backed off model in [MG, in prep] trained on only 1380 train ing phrases. ....

M. Collins, J. Brooks. 1995. Preposi- tional Phrase Attachment through a BackedOff Model, A CL 3rd Workshop on Very Large Corpora, Pages 27-38, Cambridge, Mas- sachusetts, June.

.... for example, on the assumption that case slots are mutually independent (Hindle and Rooth, 1991; Sekine et al. 1992; Resnik, 1993a; Grishman and Sterling, 1994; Alshawi and Carter, 1994) or at most two case slots are dependent (Brill and Resnik, 1994; Ratnaparkhi, Reynar, and Roukos, 1994; Collins and Brooks, 1995). 2.5 Structural Disambiguation 2.5.1 The lexical approach There have been many probabilistic methods proposed in the literature to address the structural disambiguation problem. Some methods tackle the basic problem of resolving ambiguities in quadruples (v, n 1 , p, n 2 ) e.g. eat, ....

....where random variable a takes on attv and attn as its values, and random variables (v, n 1 , p, n 2 ) take on quadruples as their values. Since the number of parameters in the distribution is very large, accurate estimation of the distribution would be impossible. In order to address this problem, Collins and Brooks (1995) devised a back o# method. It first calculates the conditional probability P (a v, n 1 , p, n 2 ) by using the relative frequency f(a, v, n 1 , p, n 2 ) f(v, n 1 , p, n 2 ) if the denominator is larger than 0; otherwise it successively uses lower order frequencies to heuristically calculate the ....

[Article contains additional citation context not shown here]

Collins, Michael and James Brooks. 1995. Prepositional phrase attachment through a backed-o# model. Proceedings of the 3rd Workshop on Very Large Corpora.

....data any new examples that are classified with high confidence, and re estimating the probability distributions. The algorithm scored around 80 on a test set of 880 handlabelled examples. Human experts scored 85 88 on the same test set. Chapter 2. Previous Work 16 2.5.1. 2 Back off Models: Collins and Brooks 1995 The approach described by Collins Brooks in this paper continues in the same vein as Hindle and Rooth s, but attacks the sparse data problem in a different and more effective way. This algorithm also uses the explicitly supervised Ratnaparkhi dataset. Hindle Rooth handled the sparse data ....

....about which classifier to use and which features and tuples to include in that classifier. Chapter 5 considers several new feature types, and Chapter 6 compares three general methods for selecting features from a large pool of candidates. Chapter 4 is essentially about the choice of classifier: [Collins and Brooks, 1995] achieve very good results using a back off Chapter 2. Previous Work 28 algorithm, but the underlying model maximum likelihood has several shortcomings. Maximum entropy models have several attractive properties, and it seems quite possible to improve on the [Ratnaparkhi et al. 1994] and ....

[Article contains additional citation context not shown here]

Collins, M. and Brooks, J. (1995). Prepositional phrase attachment through a backed-off model. In Proceedings of the Third Workshop on 88 Bibliography 89 Very Large Corpora, pages 27--38, Somerset, NJ. Association for Computational Linguistics.

....analysis. We rst show that MBL methods can be reformulated as feature based learning algorithms. Furthermore, the features that emerge from this mapping happen to coincide with features commonly used by many other features based learning algorithms in NLP [ Brill, 1995; Golding, 1995; Roth, 1998; Collins and Brooks, 1995; Ratnaparkhi et al. 1994 ] What may be surprising is, as we show, that the prediction computed by MBL can be simply computed by a linear function over this set of features. Placing seemingly di erent learning paradigms MBL on one hand and feature based methods, including both machine ....

....of tokens in a sentence. However, we observe that these conjunctive features are the common features used anyhow by most learning algorithms applied to various disambiguation and predictions tasks in NLP. For example, various methods have been used to study the PPA problem discussed in Ex. 3. 1 [ Collins and Brooks, 1995; Brill and Resnik, 1994; Ratnaparkhi et al. 1994; Krymolowski and Roth, 1998 ] All of them make use of essentially the same conjunctive features presented here. The same holds for many other problems. We conclude that MBL methods are linear over feature spaces that are commonly used for ....

M. Collins and J Brooks. Prepositional phrase attachment through a backed-o model. In Proceedings of Third the Workshop on Very Large Corpora, 1995.

....if the difference between the competing cooccurrence values is above a certain threshold. In this way the co occurrence values served to decide 58 of our test cases with an attachment accuracy of 75 . In the more successful experiments for PP attachment in English (Stetina and Nagao 1997, Collins and Brooks 1995) the co occurrence statistics included the noun within the PP. The motivation behind this becomes immediately clear if we compare the PPs in the example sentences 3 and 4. Since both PPs start with the same preposition only the noun within the PP helps to find the correct attachment. 3) Peter ....

Collins M, Brooks J 1995 Prepositional phrase attachment through a backed-off model. In Proc. of the Third Workshop on very large corpora.

....or a noun (as for saw the girl with a basketball) a lexical association score is determined, using the structures in a hand tagged corpus as reference. If the preposition (e.g. with) is more often attached in the corpus to the noun (girl) than the verb (saw) then verb attachment is more likely. Collins Brooks (1995) take a similar approach, but they also include the object of the PP in their scoring. 25 Niemann Determining PP Attachment In D. Estival (ed. Determining PP Attachment through Semantic Associations and Preferences. Michael Niemann (1998) pp 25 32. Abstracts for the ANLP Post Graduate ....

Collins, Michael & Brooks, James. 1995. Prepositional Phrase Attachment through a Backed-Off Model.

....and denominator for each disjunct separately, as if that disjunct were the entire reduction, and nd our estimate by adding the numerators and adding the denominators. The disjunct with the greater denominator (i.e. whose condition is more common) will have the greater in uence on our estimate [Collins Brooks 1995]. 4. Models A and D must be able to decide, for words at positions k and i k such that k already has c children between k and i, whether i is the (c 1)st child of k: Pr(link from i to k j tw i ; tw k ; tw kid(k;c) reduction list : word(tw i ) tag(tw i ) word(tw k ) tag(tw k ) ....

M. Collins and J. Brooks. 1995. Prepositional phrase attachment through a backedo  model. Proceedings of the Third Workshop on Very Large Corpora, 27-38. cmp-lg/9506021.

....over the Penn Treebank and they are using a semantic dictionary to cluster the words. Our method is easier to use and widely applicable. Our method might be extended to include the head noun within the PP. Including the head noun has been shown to have positive effects on the attachment quality [Collins and Brooks 95] This entails moving from a quadruple to a quintuple (V, N, P, head N, PPfunction) which often posed sparse data problems. With the help of the WWW we hope to overcome these problems. In the future we will also look into combining detailed corpus analysis and WWW frequencies to get the ....

Michael Collins and James Brooks. 1995. Prepositional Phrase Attachment through a backed-off model. In: Proc. of the Third Workshop on Very Large Corpora.

.... machinery to accommodate larger models, the availability of resources such as the Penn Treebank (Marcus et al. 1993) and the success of machine learning techniques for lowerlevel NLP problems, such as part of speech tagging (Church, 1988; Brill, 1995) and PPattachment (Brill and Resnik, 1994; Collins and Brooks, 1995). However, perhaps even more signi cant has been the lexicalization of the grammar formalisms being probabilistically modeled: crucially, all the recent, successful statistical parsers have in some way made use of bilexical dependencies. This includes both the parsers that attach probabilities to ....

M. Collins and J. Brooks. 1995. Prepositional phrase attachment through a backed-o model. In Third Workshop on Very Large Corpora, pages 2738.

....that the model is indifferent to the event, as compared to other available choices. This provides a specific neutral value as the cost for unseen events. 2. 5 Unseen Event Costs Our approach to deriving costs for unseen choices generalizes the notion of backed off likelihood estimates [5, 10, 11]. Clustering techniques [12] are an alternative. When faced with an unseen choice (ejc) we use the cost, suitably normalized, for (ejc 0 ) where c 0 is a more general context than c. We could also apply a similar generalization to the equivalence class for the event e, but we have not yet ....

M. Collins and J. Brooks. Prepositional phrase attachment through a backed-off model. In Proc. 3rd Wks. on Very Large Corpora, pages 27--38, Cambridge, Massachusetts, 1995. ACL.

....DT NN JJ NN (Wall Street Journal corpus) The domains we proposed for this evaluation are the non recursive phrase structures (PS) and simple clauses. Evaluation using a full parsing does not seem realistic presently since parsing prepositional phrases requires generally lexical resources (Collins and Brooks, 1995). The use of PS and clauses allows a fully coverage of many tagsets. 3.2. The Theoretically Minimal Tagset We would like to point out that the quality of a tagset does not depend on the quantity of tags. For this purpose, we build up the minimum tagset necessary to parse sentence whatever the ....

Collins, Michael and James Brooks, 1995. Prepositional phrase attachment through a backed-off model. In Third Workshop on Very Large Corpora.

No context found.

Michael Collins and James Brooks. Prepositional phrase attachment through a backed-o# model. In Proceedings of the Third Workshop on Very Large Corpora, pages 27--38, Cambridge, 1995.

No context found.

Collins, Michael and James Brooks. 1995. Prepositional phrase attachment through a backed-o# model. In Proceedings of the Third Workshop on Very Large Corpora, pages 27--38, Cambridge.

No context found.

M. Collins and J. Brooks. 1995. Prepositional phrase attachment through a backed-o# model. In Proceedings of the Third Workshop on Very Large Corpora, Cambridge.

No context found.

Collins, Michael and James Brooks. 1995. Prepositional phrase attachment through a backed-off model. In David Yarowsky and Kenneth Church, editors, Proceedings of the Third Workshop on Very Large Corpora, pages 27--38, Cambridge, MA, June.

No context found.

Michael Collins and James Brooks. Prepositional phrase attachment through a backed-o# model. In Proceedings of the Third Workshop on Very Large Corpora, pages 27--38, Cambridge, 1995.

No context found.

Collins, Michael and James Brooks. 1995. Prepositional phrase attachment through a backed-o# model. In Proceedings of the Third Workshop on Very Large Corpora, pages 27--38, Cambridge.

No context found.

M.J Collins and J. Brooks. Prepositional phrase attachment through a backed-off model. In Proceedings of the Third Workshop on Very Large Corpora, Cambridge, 1995.

No context found.

Collins, M., & Brooks, J. (1995). Prepositional phrase attachment through a backed-oV model. In Proceedings of the Third Workshop on Very Large Corpora.

No context found.

M.J Collins and J. Brooks. Prepositional phrase attachment through a backed-off model. In Proceedings of the Third Workshop on Very Large Corpora, Cambridge, 1995.

No context found.

Michael Collins. 1995. Prepositional phrase attachment through a backed-o# model. In Proceedings of the Third Workshop on Very Large Corpora, pages 27#38, Cambridge, Massachusetts.

No context found.

-136. Collins M. and Brooks J. (1995) "Prepositional Phrase Attachment through a Backing-Off Model", In David Yarowsky & Ken Church(eds.) Proceedings of the third workshop on very large corpora, MIT.

No context found.

M. Collins and J Brooks. Prepositional phrase attachment through a backed-off model. In WVLC 1995.

No context found.

M. Collins and J Brooks. Prepositional phrase attachment through a backed-o model. In WVLC 1995.

No context found.

Michael Collins. 1995. Prepositional phrase attachment through a backed-off model. In Proceedings of the Third Workshop on Very Large Corpora, pages 27--38, Cambridge, Massachusetts.

First 50 documents

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC