Mooney, R.J., and Califf, M.E., Induction of first-order decision lists: results on learning the past tense of English verbs, Journal of Artificial Intelligence Research 3 (1995) 1-24.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....(outliers) then this general clause should be preceeded with a more special clause handling these exceptions. Together with the once interpretation, this strategy will produce a form of a specific to general decision list, whose advantage to functional representation has been argued in [10]. To attain such clause ordering, we use the degree of freedom given by the seed example selection in ILP systems like Aleph [5] and Progol [11] In these systems, the seed example is selected randomly or in the presentation order and used for the construction of a bottom clause which is then ....

R.J. Mooney and M.E. Cali#. Induction of first-order decision lists: Results on learning the past tense of English verbs. In L. De Raedt, editor, Proceedings of the 5th International Workshop on Inductive Logic Programming, pages 145--146. Department of Computer Science, Katholieke Universiteit Leuven, 1995.

....btw. BL and the other methods smaller, but not significantly. 4.2 Learning English Past Tense Rules The second experiment is based on 1392 tuples of English verbs and their past tenses. Learning rules of English past tense by a multi clause Prolog program has been studied with noise free data [9, 13]. The background knowledge contains the predicate split 3 which splits a word into a prefix and su#x (e.g. split( m,a,i,l,e,d] m,a,i,l] e,d] see [9] for typical hypotheses constructed by ILP in this domain. Unlike the noisefree experiments, in our case the output argument is distorted by ....

....verbs and their past tenses. Learning rules of English past tense by a multi clause Prolog program has been studied with noise free data [9, 13] The background knowledge contains the predicate split 3 which splits a word into a prefix and su#x (e.g. split( m,a,i,l,e,d] m,a,i,l] e,d] see [9] for typical hypotheses constructed by ILP in this domain. Unlike the noisefree experiments, in our case the output argument is distorted by altering a number of characters in the word such that the probability of n wrong characters decreases exponentially with n to approximate the normal ....

R.J. Mooney and M.E. Cali#. Induction of first--order decision lists: Results on learning the past tense of English verbs. Journal of Artificial Intelligence Research, 3:1--24, 1995.

....When there are more than two classes, it is necessary to learn a theory for each class. The question is how to apply these theories to a new example [30] We have employed two approaches for dealing with multiclass problems. The first one is the use of ordered rules, also named decision lists [23, 19]. In this case, the first rule that covers the example assigns its label to it. The learning process consists of generating a rule for the class with most uncovered examples and iterate until all the examples are covered. Figure 1 shows an example of such a decision list. The second approach, ....

Raymond J. Mooney and Mary Elaine Cali#. Induction of first-order decision lists: results on learning the past tense of english verbs. Journal of Artificial Intelligence Research, 3:1--24, 1995.

....(outliers) then this general clause should be preceeded with a more special clause handling these exceptions. Together with the once Gammainterpretation, this strategy will produce a form of a specific to general decision list, whose advantage to functional representation has been argued in [9]. To attain such clause ordering, we use the degree of freedom given by the seed example selection in ILP systems like Aleph [5] and Progol [10] In these systems, the seed example is selected randomly or in the presentation order and used for the construction of a bottom clause which is then ....

....or in the presentation order and used for the construction of a bottom clause which is then suitably generalised. The idea of our method is that we direct the seed example selection as to first choose (and cover) those examples that are outliers to some potentially good clause. Unlike e.g. FOIDL [9], which produces a similarly ordered clause set, we want to avoid backtracking (deleting previously constructed clauses) to protect efficiency. During the computation of f e E (C n ) for each candidate clause C n , we also evaluate the function HopeE (C n ) maxOaeE (f e EnO (C n ) ....

R.J. Mooney and M.E. Califf. Induction of first-order decision lists: Results on learning the past tense of English verbs. In L. De Raedt, editor, Proceedings of the 5th International Workshop on Inductive Logic Programming, pages 145--146. Department of Computer Science, Katholieke Universiteit Leuven, 1995.

....btw. BL and the other methods smaller, but not significantly. 4.2 Learning English Past Tense Rules The second experiment is based on 1392 tuples of English verbs and their past tenses. Learning rules of English past tense by a multi clause Prolog program has been studied with noise free data [8, 12]. The background knowledge contains the predicate split 3 which splits a word into a prefix and suffix (e.g. split( m,a,i,l,e,d] m,a,i,l] e,d] see [8] for typical hypotheses constructed by ILP in this domain. Unlike the noisefree experiments, in our case the output argument is distorted by ....

....verbs and their past tenses. Learning rules of English past tense by a multi clause Prolog program has been studied with noise free data [8, 12] The background knowledge contains the predicate split 3 which splits a word into a prefix and suffix (e.g. split( m,a,i,l,e,d] m,a,i,l] e,d] see [8] for typical hypotheses constructed by ILP in this domain. Unlike the noisefree experiments, in our case the output argument is distorted by altering a number of characters in the word such that the probability of n wrong characters decreases exponentially with n 2 to approximate the normal ....

R.J. Mooney and M.E. Califf. Induction of first--order decision lists: Results on learning the past tense of English verbs. Journal of Artificial Intelligence Research, 3:1--24, 1995.

....with newly acquired training examples. If the examples are misclassified by the existing theory, they can be added as exceptions at the top of the decision list to ensure their correct classification. Later on, learning can be used to replace these exceptions with rules, if possible. Foidl [24] and Clog [15] are two of the first order decision list learners. It is also worth mentioning that Clog, unlike Progol, is an incremental learner. Eager ILP vs. Analogical Prediction The relative merits of eager and lazy learning have already been discussed. ILP belongs to the eager learning ....

Raymond J. Mooney and Mary Elaine Califf. Induction of first--order decision lists: Results on learning the past tense of English verbs. Journal of Artificial Intelligence Research, June 1995.

....used an inductive logic programming (ILP) system that learns first order decision lists, i.e. ordered sets of rules. We first explain the notion of first order decision lists on the problem of synthesis of the past tense of English verbs, one of the first examples of learning morphology with ILP (Mooney Califf, 1995). We then lay out the ILP formulation of the problem of learning rules for morphological analysis of Slovene nouns and adjectives and describe how it was addressed with the ILP system Clog. The induction results are illustrated for an example MSD. We finally discuss the evaluation of the learned ....

....ILP system Clog. The induction results are illustrated for an example MSD. We finally discuss the evaluation of the learned rules on the evaluation set. 3. 1 Learning decision lists The ILP formulation of the problem of learning rules for the synthesis of past tense of English verbs considered in (Mooney Califf, 1995) is as follows. A logic program has to be learned defining the relation past(PresentVerb,PastVerb) where PresentVerb is an orthographic representation of the present tense form of a verb and PastVerb is an orthographic representation of its past tense form. PresentVerb is the input and PastVerb ....

[Article contains additional citation context not shown here]

Mooney, R. J., & Califf, M. E. (1995). Induction of first-order decision lists: Results on learning the past tense of English verbs. Journal of Artificial Intelligence Research, pp. 1--24.

....to automate this task. Inductive Logic Programming (ILP) deals with the induction of predicate definitions from examples and background knowledge. Recently, researchers have been trying to apply ILP techniques to the construction of natural language processing (NLP) systems. Mooney and Califf [11] have applied ILP techniques to learning the past tense of English verbs, and showed that ILP techniques are more effective than neuralnetwork and decision tree methods. Mooney [12] has also worked a system which learns a parser from a training corpus of parsed sentences. In [14] Muggleton ....

Mooney, R. J., and Califf M.E., Induction of First-Order decision Lists: Results on Learning The Past Tense of English Verbs, Journal of Artificial Intelligence Research, 3 , 1995, pp:1-24.

....sizes of partial 10 models on BNC test data. The single mistake was on the pair (forbid,forbad) The model gave the regular past. 5. Experiment 1: English past tense For this experiment I used the standard English Past Tense data set , available as an appendix to (Ling, 1994) and also used by Mooney and Califf (1995) and various other researchers, which consists of 1389 pairs of words in the UNIBET phonetic alphabet. For test data I selected some unseen verbs from the British National Corpus(Aston Burnard, 1998) I chose all verbs that occurred ten times with the tag vvd, the past tense tag. After removing ....

....with Symbolic learning models Various symbolic learning algorithms have been presented for the English past tense. Ling (1994) proposes a symbolic pattern associator that he claims performs clearly better than the connectionist models. Unfortunately it suffers from many of the same problems. Mooney and Califf (1995) presents an inductive logic programming approach (Muggelton, 1999; Muggleton Bain, 1999) that performs well. It is not clear how well this would generalise to other forms of morphology. It appears to require a set of predicates to be defined in advance in these papers, there is a predicate ....

Mooney, R. J., & Califf, M. E. (1995). Induction of first-order decision lists: Results on learning the past tense of english verbs. Journal of Artificial Intelligence Research, 3, 1--24.

.... include drug design [35] protein secondary structure prediction [47] mutagenicity prediction [55] carcinogenesis prediction [56] medical diagnosis [43] discovery of qualitative models in medicine [31] finite element mesh design [17] telecommunications [54] natural language processing [44], recovering software specifications [7] and many others. There are many overview papers on ILP applications, including an early overview [3] A tutorial on ILP applications is available at http: www ai.ijs.si ilpnet ilpkdd . A report providing short descriptions of applications developed as ....

R.J. Mooney, M.E. Califf. Induction of first-order decision lists: Results on learning the past tense of English verbs. Journal of Artificial Intelligence Research 3: 1--24, 1995.

....any other applied inductive learning algorithm. 2.2. 2 Decision Lists Decision lists [184] are ordered lists of conjunctive rules (where rules are tested in order and the first one that matches an instance is used to classify it) which have been applied in a number of concept learning systems [47, 152, 174]. Decision lists work well in domains with many attributes (or with attributes with many values) because they avoid to some extent the data fragmentation problem. Thus, regarding NLP, they have been applied to lexical ambiguity resolution. In particular: Word sense disambiguation, lexical choice ....

....NLP tasks, such those related with machine translation, spelling correction, etc. DTs ME IBL TBL NB Acquisition of verbal properties [221, 209] General machine translation [10] 100] Spelling correction [133] 86, 88, 89] DLs ILP NNs Clust GAs LSM LogL Acquisition of verbal properties [152, 153, 29] [209] 209, 135] General machine translation [235] Spelling correction [241, 208] 116] 89] Generation [176] Table 5: References corresponding to Machine Translation and other NLP tasks 14 3 Word Sense Disambiguation: A Case Study in Supervised Machine Learning The present section is ....

R. J. Mooney and M. E. Califf. Induction of First--order Decision Lists: Results on Learning the Past Tense of English Verbs. Journal of Artificial Intelligence Research, 3:1--24, 1995.

.... include drug design [16] protein secondary structure prediction [29] mutagenicity prediction [38] carcinogenesis prediction [39] medical diagnosis [24] discovery of qualitative models in medicine [11] finite element mesh design [5] telecommunications [37] natural language processing [25], recovering software specifications [3] and many others. Detailed surveys of ILP are provided by [18, 30] while [19] offers an extensive overview of on line available systems and datasets, as well as a bibliography with nearly 600 entries. 28, 4] are collections of research papers. ....

R.J. Mooney, M.E. Califf. Induction of first-order decision lists: Results on learning the past tense of English verbs. Journal of Artificial Intelligence Research 3: 1--24, 1995.

....Academic Publishers. Printed in the Netherlands. kaz man.tex; 15 06 2000; 17:55; p. 2 3 For this purpose, some corpus based approaches employ eager learning in that they derive a single theory or tool that can subsequently be applied to the training or unseen data (Rumelhart and McClelland, 1986; Mooney and Califf, 1995). Other, lazy learning , methods use the training data directly when segmenting a word (Deligne, 1996; Yvon, 1997) A comparison between hand crafted tools for word segmentation and those learnt from data shows that the former, unlike the latter, can be based on rather sophisticated approaches, ....

....modified by a human expert. ILP has been applied to learning word morphology on several occasions. In most cases, the main task is to learn the relationship between a word and its morphosyntactic features, which effectively involves segmenting the words and making use of the constituents produced (Mooney and Califf, 1995; Manandhar et al. 1998; Muggleton and Bain, 1999) ILP has proved very efficient in the area of morphology, as the learning of English past tense shows (Mooney and Califf, 1995; Muggleton and kaz man.tex; 15 06 2000; 17:55; p.3 4 Bain, 1999) When trying to predict the past tense form of an ....

[Article contains additional citation context not shown here]

Mooney, R. J. and M. E. Califf: 1995, `Induction of First--Order Decision Lists: Results on Learning the Past Tense of English Verbs'. Journal of Artificial Intelligence Research 3, 1--24.

....Logic Programming (ILP) Muggleton, 1994; Muggleton De Raedt, 1994) However, the models differ in details that are crucial. One source of difference is the structure of examples. An example in the standard form of ILP (Quinlan, 1990; Dzeroski, Muggleton, Russell, 1992; Muggleton, 1994; Mooney Califf, 1995) includes a single ground instance of a relation and the rest of the information on this example is provided through the background knowledge. In contrast an example in our model describes a complete situation and the ground action taken in that situation, and is therefore more explicit. On the ....

....to others. Despite those differences, the structure of induced expressions is similar, and similar techniques can be used in both models. Our learning approach is similar to the covering method in FOIL (Quinlan, 1990) and the representation is similar to the first order decision lists studied by Mooney and Califf (1995). Moreover, our arguments are similar to the ones in (De Raedt Dzeroski, 1994; Valiant, 1985) and can yield positive results on learning first order decision lists in the ILP context. On the other hand, several sophisticated methods for learning have been applied in ILP that may be useful in ....

Mooney, R. J., & Califf, M. E. (1995). Induction of first-order decision lists: Results on learning the past tense of English verbs. Journal of Artificial Intelligence Research, 3, 1--24.

....or more examples. Examples Hypotheses past( w,o,r,r,y] w,o,r,r,i,e,d] past( w,h,i,z] w,h,i,z,z,e,d] past(A,B) split(A,C, r,r,y] split(B,C, r,r,i,e,d] past( g,r,i,n,d] g,r,o,u,n,d] Figure 7: Form of examples and hypotheses for past tense domain 4. 4 Morphology Mooney and Califf [18] have applied ILP to learning the past tense of English verbs. Learning of English past tense has become a benchmark problem in the computational modelling of human language acquisition [29, 14] In [18] it was shown that a particular ILP system, FOIDL, could learn this transformation more ....

....7: Form of examples and hypotheses for past tense domain 4.4 Morphology Mooney and Califf [18] have applied ILP to learning the past tense of English verbs. Learning of English past tense has become a benchmark problem in the computational modelling of human language acquisition [29, 14] In [18] it was shown that a particular ILP system, FOIDL, could learn this transformation more effectively than previous neural network and decision tree methods. FOIDL s first order default rule style representation was demonstrated by the authors as producing a predictive accuracy advantage in this ....

R.J. Mooney and M.E. Califf. Induction of first-order decision lists: Results on learning the past tense of english verbs. Journal of Artificial Intelligence Research, 3:1--24, 1995.

....a propositional learning algorithm that is similar in structure to CSC. The main differences are that CSC is relational and it is used here in an iterative way which was not the case with Liu and colleagues. The approach in [Dehaspe De Raedt 97] also involved the use of association rules. Mooney 95] and [Lopes Brazdil 98] used first order decision lists in the problem of learning the past tense of English verbs. 9 Conclusion In the experiments we carried out the algorithm CSC(RC) obtains significantly better results when it is applied iteratively than when it is applied in two steps. The ....

R. J. Mooney. Induction of First-order Decision Lists: Results on Learning the Past Tense of English Verbs. Journal of Artificial Intelligence Research 3, pp. 1-24, 1995.

.... on the original, propositional, applications, 2) extends the applicability of TBL towards cases where a more expressive first order logic knowledge representation format is useful, 3) can be specialized towards the task of finding more traditional (first order) decision lists (Rivest, 1987; Mooney and Califf, 1995), and (4) suggests a new approach in between TBL and decision lists. Essentially, Brill s is a greedy algorithm that, starting from an initial classification of the data, constructs a model by adding transformation rules of the type if pattern matches the example then change classification of the ....

Mooney, R., and Califf, M. 1995. Induction of first-order decision lists: Results on learning the past tense of english verbs. Journal of Artificial Intelligence Research 1--23.

....sum, we find indications for a high disjunctity or polymorphism of the language data sets investigated in this study. Other studies in which machine learning algorithms are applied to language data, and in which special attention is payed to learning exceptions, mention similar indications (e.g. Mooney and Califf (1995; Van den Bosch et al. 1995) However, the question whether language data in general exhibits a higher degree of disjunctiveness or polymorphism than comparable data sets of non linguistic origin remains an open one, and will be a focal point in future research. 6.1.2. Usefulness of exceptional ....

Mooney, R. J. and M. E. Califf. 1995. Induction of first-order decision lists: Results on learning the past tense of english verbs. Journal of Artificial Intelligence Research, 3:1--24.

.... approaches (Rumelhart and McClelland (1986) MacWhinney and Leinbach (1991) arguing for connectionist models, Pinker and Prince (1988) Lachter and Bever (1988) Marcus et al. 1992) arguing against connectionist models, Ling and Marinov (1993) Ling (1994) using ID3 C4.5 decision trees, and Mooney and Califf (1995, 1996) using inductive logic programming decision lists, among others) However except for a couple of forays into German this literature has been exclusively concerned with the learning of the English past tense. This has not worried some. Ling is happy to describe it as a landmark task . ....

Mooney, R. J., and M. E. Califf. 1995. Induction of first-order decision lists: Results on learning the past tense of english verbs. Journal of Artificial Intelligence Research 3:1--24.

....in computational linguistics. Most current learning research in NLP employs statistical techniques inspired by research in speech recognition, such as hidden Markov models (HMMs) and probabilistic context free grammars (PCFGs) There has been some recent research on logic based language learning (Mooney Califf, 1995; Cohen, 1996; Freitag, 1998) in particular a recent body of European inductive logic programming (ILP) research on language (Cussens, 1997; Manandhar, Dzeroski, Erjavec, 1998; Kazakov Manandhar, 1998; Eineborg Lindberg, 1998; Lindberg Eineborg, 1998; Cussens, Dzeroski, Erjavec, 1999; ....

Mooney, R. J., & Califf, M. E. (1995). Induction of first-order decision lists: Results on learning the past tense of English verbs. Journal of Artificial Intelligence Research, 3, 1--24.

.... structure (Muggleton, King, Sternberg, 1992) Detailed experimental comparisons of ILP and feature based induction have demonstrated the advantages of relational representations in two language related tasks, text categorization (Cohen, 1995) and generating the past tense of an English verb (Mooney Califf, 1995). Our research attempts to bridge the gap between rational and empirical approaches to NLP by applying ILP to the problem of parser acquisition. The main advantage of this approach is its flexibility. Given the power of first order rules, there is less need to hand engineer appropriate features ....

Mooney, R. J., & Califf, M. E. (1995). Induction of first-order decision lists: Results on learning the past tense of English verbs. Journal of Artificial Intelligence Research, 3, 1--24.

....as unanalyzed atomic units. Being able to recognize the similarities between words having similar roots or resulting from similar derivations might lead to better generalization. Some initial work along these lines has applied ILP to the problem of learning to form the past tense of English verbs (Mooney Califf, 1995). At the lexical level, automated techniques for lexicon construction could broaden the applicability of Chill. Thompson (1995) has demonstrated an initial approach to corpus based acquisition of lexical mapping rules suitable for use with Chill style parser acquisition systems. The basic idea is ....

Mooney, R. J., & Califf, M. E. (1995). Induction of first-order decision lists: Results on learning the past tense of English verbs. Journal of Artificial Intelligence Research, in press.

No context found.

Mooney, R.J., and Califf, M.E., Induction of first-order decision lists: results on learning the past tense of English verbs, Journal of Artificial Intelligence Research 3 (1995) 1-24.

No context found.

Raymond J. Mooney and Mary Elaine Califf. Induction of first--order decision lists: Results on learning the past tense of English verbs. Journal of Artificial Intelligence Research, June 1995.

No context found.

Mooney, R. J. Induction of First-order Decision Lists: Results on Learning the Past Tense of English Verbs. Journal of Artificial Intelligence Research 3, pp. 1-24, 1995.

First 50 documents  Next 50

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