MacWhinney, B., & Leinbach, J. (1991). Implementations are not conceptualizations: Revising the verb learning model. Cognition, 29, 121-157.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....frequently studied in cognitive science and neural networks as a touchstone problem in language acquisition. In fact, there has been signi cant debate whether or not rule learning is an adequate cognitive model of how children learn this task (Rumelhart McClelland, 1986; Pinker Prince, 1988; MacWhinney Leinbach, 1991). Typically, the problem is studied in its phonetic form, in which a string of phonemes for the present tense is mapped to a string of phonemes for the past tense. The problem is interesting since one must learn the regular transformation of adding ed, as well as particular irregular patterns ....

MacWhinney, B., & Leinbach, J. (1991). Implementations are not conceptualizations: Revising the verb model. Cognition, 40, 291-296.

.... account proposed by Pinker (1989) Introduction The debate as to whether connectionist or symbolic models provide a better account for linguistic knowledge and how their accounts differ has centered on the English past tense (e.g. Rumelhart and McClelland, 1986) Pinker and Prince, 1988) (MacWhinney and Leinbach, 1991), Marcus et al. 1992) Daugherty and Seidenberg, 1993) Daugherty and Hare, 1994) This paper investigates a new area, the acquisition of verb subcategorization. 1) a. She gave a book to the student. b. She gave the student a book. c. She donated a book to the church. d. She donated the ....

MacWhinney, B. and Leinbach, J. (1991). Implementations are not conceptualizations: Revising the verb model.

....in fig. 1. It consists of an input layer and an output layer that are fully interconnected, and a hidden layer with Gaussian receptive fields that is constructed during learning. The input layer receives the verb stem in a templated phonological form that was developed by MacWhinney and Leinbach [10] for their past tense neural network model. In contrast to other models of past tense learning, in the CNN the formation of the past tense is considered a classification task: instead of producing the phonological form of the past tense, 23 classes were defined according to the way in which the ....

....task at hand. As such it corresponds conceptually to notions of constructivist learning in developmental psychology. 4. 2 Data The training corpus for the network consisted of 8,000 verb tokens, corresponding to 1,066 types, and was a modified version of one used in previous past tense models [9, 10]. It accurately reflects the type and token frequencies of regular and irregular verbs in language. While other models have simulated development by manipulating the training corpus during training [17, 20] here the focus is on the development of the system itself, and therefore the verbs were ....

[Article contains additional citation context not shown here]

MacWhinney, B. Leinbach, J. (1991). Implementations are not conceptualizations: Revising the verb learning model. Cognition, 40, 121--157.

....localism : the English past tense forms are learned in a network which is dedicated to this morphological task alone #Pinker Prince, 1988#. In the Rumelhart and McClelland model, as in most of the succeeding connectionist models of morphology acquisition #Daugherty Seidenberg, 1992; MacWhinney Leinbach, 1991; Plunkett Marchman, 1991#, morphology learning consists in learning to map a stem onto an af #xed form. If such a network is trained to learn the English past tense, say, howwould it bene#t from this knowledge in learning the identically formed regular English past participle Pinker and ....

....models distinguish morphology from phonology and how the phonological generalizations which characterize the various English s and ed su#xes can be captured in a small set of simple phonological rules. The problem for connectionist models, they argue, is that there is no place for phonology. As MacWhinney and Leinbach #1991# have shown, however, morphological localism is not an inherent feature of connectionist models. By adding a set of input features which distinguish di#erent morphological categories, e.g. past tense and past participle, they give their network the capacityto generalize from one form to ....

[Article contains additional citation context not shown here]

MacWhinney, B. & Leinbach, J. #1991#. Implementations are not conceptualization: revising the verb learning model. Cognition, 40, 121#157.

....of the phenomenon progresses. Unfortunately, this does not occur often enough in computational modeling of human learning. Notable exceptions include several papers in this issue, and a series of papers on learning the past tense of verbs (Rumelhart McClelland, 1986; Pinker Prince, 1988; MacWhinney Leinbach, 1991; and Ling Marinov, 1993) I d like to propose that the UCI Repository of Machine Learning Databases and Domain theories be extended to include databases that have been used to evaluate computational models of human learning. 1 I ve stored four databases used in Pazzani (1992) in ....

MacWhinney, B. & Leinbach, J. (1991). Implementations are not conceptualizations: Revising the verb learning model. Cognition, 39, .235-290.

....is in contrast with the English past tense, where 86 of all verb types (40 of all tokens) are regular. The instances in which the regular cases do not form the majority of all cases can be seen as a touchstone for connectionist models. This is because homogenous connectionist models (e.g. MacWhinney and Leinbach 1991, Rumelhart and McClelland 1987) learn by exploiting the statistical regularities of the input data and therefore necessarily assign the regular status to the most frequently occuring input. In this paper, a modular connectionist network is described that implements the rule associative memory ....

MacWhinney, B., and Leinbach, J. (1991). Implementations are not conceptualizations: Revising the verb learning model. Cognition, 40:121--157.

.... and an associative net used in generating irregular forms (e.g. take took) see Pinker 1991 for a summary) Connectionists have also presented modified models that addressed many of the limitations of Rumelhart and McClelland s original work (see, for 4 example, Daugherty Seidenberg 1992; MacWhinney Leinbach 1991) and have challenged findings taken as supporting the dual mechanism theory (e.g. Hoeffner, 1997) Because the past tense of English is a minor aspect of language and because this controversy has continued without resolution for almost 10 years, it is worth remembering why these issues are ....

MacWhinney, Brian, & Jared Leinbach. 1991. Implementations are not conceptualizations: Revising the verb learning model. Cognition 39.121-157.

....the traditional, rule based view that learning regular and irregular past tense relies on separate mechanisms. Because of a number of flaws, this model is now of little but historical value. More importantly, it sparked off a number of other attempts to tackle the problem with different models: MacWhinney and Leinbach (1991) presented an improved network model, and Ling and Marinov (1993) devised a symbolic counterpart, the Symbolic Pattern Associator. 5 The most recent addition to this line is the constructivist network model introduced in Westermann (in press) Westermann employed the growing neural gas algorithm ....

....overgeneralising them in different ways (e.g. knowed as well as knewed instead of knew ) Westermann points out that the constructivist model is actually the first in the line in which U shaped learning was observable and a result solely of the internal network dynamics. The model by MacWhinney and Leinbach (1991) failed to produce the initial phase of correct production. The symbolic pattern associator displayed the different phases, but as Westermann (in press) notes, it did so due to the manipulation of a psychologically unmotivated learning parameter which directly controlled how often a verb had to be ....

MacWhinney, B. and Leinbach, J. (1991). Implementation are not conceptualizations: Revising the verb learning model. Cognition, 40, 121--157.

....from significantly fewer examples than all previous methods. 1 Introduction The problem of learning the past tense of English verbs has been widely studied as an interesting subproblem in language acquisition. Previous research has applied both connectionist and symbolic method to this problem [22, 12, 9]; however, previous efforts used specially designed feature based encodings that impose a fixed limit on the length of words and fail to capture the generativity and position independence of the underlying transformation. We believed that representing the problem as constructing a logic program ....

....preferring literals that cover many positives and few negatives. The papers referenced above provide details and information on additional features. 2. 2 Learning the Past Tense of English Verbs The problem of learning the English past tense has been attempted by both connectionist systems [22, 12] and systems based on decision tree induction [11, 9] The task to be learned in these experiments is: given a phonetic encoding of the base form of an English verb, generate the phonetic encoding of the past tense form of that verb. The task can also be done using the alphabetic forms forms of ....

B. MacWhinney and J. Leinbach. Implementations are not conceptualizations: Revising the verb model. Cognition, 40:291--296, 1991.

....tries to modify existing constraints or construct new ones. The goal of the acquisition procedure is to compile a corpus of English words into a series of increasingly sophisticated phonological constraints. Our theory of acquisition differs significantly from those based on statistics (such as [12, 6]) The acquisition process is rather simple. It is incremental, greedy, and fast. It has almost no parameters to adjust. It makes falsifiable predictions about the learning of phonological constraints: 1) That learning requires very few examples, 2) That the same target constraints are learned ....

Brian MacWhinney and Jared Leinbach. Implementations are not conceptualizations: Revising the verb learning model. Cognition, 40, 1991.

....more realistic models of cognitive development. Introduction Models of learning the English past tense have in the past ten years become representative of different theories of language acquisition and cognition in general. While connectionist approaches (e.g. Rumelhart McClelland, 1986; MacWhinney Leinbach, 1991; Plunkett Marchman, 1993) have maintained that both regular and irregular past tense forms can be produced in a homogeneous architecture by a single process, dual route accounts (Pinker, 1991) argue for two distinct mechanisms in different pathways, where regular forms are produced by a rule ....

.... by growing its architecture in response to the specific learning problem (see also Westermann, 1997) The performance of this network is evaluated against three existing models of past tense learning: the original network by Rumelhart McClelland (1986, R M) the improved backpropagation model by MacWhinney Leinbach (1991, M L) and the Symbolic Pattern Associator (SPA, Ling Marinov, 1993) a symbolic decision tree model. It is shown that the present constructivist model outperforms the existing symbolic and subsymbolic models both in terms of psychologically realistic learning and generalization behavior. ....

[Article contains additional citation context not shown here]

MacWhinney, B. & Leinbach, J. (1991). Implementations are not Conceptualizations: Revising the Verb Learning Model. Cognition, 40, 121--157.

....perhaps more discussion and further research than any other paper dealing with connectionism and high level cognition. Among the connectionist models directly addressing the English past tense are Cottrell Plunkett (1991) Daugherty Seidenberg (1992) Hare Elman (1992) Hoeffner (1992) MacWhinney Leinbach (1991), and Plunkett Marchman (1991) other connectionist approaches to morphology include Corina (1991) and Gasser Lee (1991) I will not attempt to review in any detail the ensuing debate and the various descendants of the Rumelhart and McClelland model. Most of the controversy has concerned two ....

MacWhinney, B. & Leinbach, J. (1991). Implementations are not conceptualization: revising the verb learning model. Cognition, 40, 121--157.

....development of Foidl was motivated by a failure we observed when applying existing ILP methods to a particular problem, that of learning the past tense of English verbs. This problem has been studied fairly extensively using both connectionist and symbolic methods (Rumelhart McClelland, 1986; MacWhinney Leinbach, 1991; Ling, 1994) however, previous efforts used specially designed feature based encodings that impose a fixed limit on the length of words and fail to capture the position independence of the underlying transformation. We believed that representing the problem as constructing a logic program for ....

....to require explicit rules. This model was heavily criticized by opponents of the connectionist approach to language acquisition for the relatively poor results achieved and the heavily engineered representations and training techniques employed (Pinker Prince, 1988; Lachter Bever, 1988) MacWhinney and Leinbach (1991) attempted to address some of these criticisms by using a standard multi layer backpropagation learning algorithm and a simpler UNIBET encoding of phonemes (in which each of 36 phonemes is encoded as a single ASCII character) Ling and Marinov (1993) and Ling (1994) criticize all of the current ....

MacWhinney, B., & Leinbach, J. (1991). Implementations are not conceptualizations: Revising the verb model. Cognition, 40, 291--296.

....than the hissing sounds) are pluralized by adding a [z] sound, and (3) Nouns ending in a voiceless consonant (other than the hissing sounds) are pluralized by adding a [s] sound. Our theory of acquisition differs significantly from those based on statistics (such as (Rumelhart McClelland 1986; MacWhinney Leinbach 1991)) It is a theory of learning not training. It is incremental, greedy, and fast. It has almost no parameters to adjust. It makes falsifiable claims about the learning of phonological constraints: 1) that learning requires very few examples tens of examples in a few steps as opposed to ....

MacWhinney, B., and Leinbach, J. 1991. Implementations are not conceptualizations: Revising the verb learning model. Cognition 40.

....explicit rules and symbolic representations play any role in language processing and cognition in general. Their argument is based to a large extent on two artificial neural network (ANN) models that are claimed to be able to learn the past tenses of English verbs. Rumelhart and McClelland, 1986; MacWhinney and Leinbach, 1991). In this article we critically review Rumelhart and McClelland s as well as MacWhinney and Leinbach s ANN models and conclude that they do not succeed in the assigned task of learning the past tenses of English verbs. In order to answer their challenge to the symbolic processing approach, we ....

....and symbolic representations (Rumelhart and McClelland, 1986, pp. 216 271) After several critical reviews of this model by Pinker and Prince (1988) Lachter and Bever (1988) and a detailed background analysis of the model by Plunkett and Marchman (1991) a new revised PDP model was proposed by MacWhinney and Leinbach (1991) (M L henceforth) They claimed that their new model meets most of the criticisms addressed at the earlier eliminativist model of Rumelhart and McClelland and that it is truly able to learn the past tenses of English verbs without any explicit representation of the acquired linguistic knowledge. ....

[Article contains additional citation context not shown here]

MacWhinney, B. & Leinbach, J. (1991). Implementations are not conceptualizations: Revising the verb model. Cognition, 40, 121-157.

....accounts of language learning, but have also raised the question whether regular and irregular past tense forms necessarily rely on different mechanisms for their formation or whether a single mechanism can account for both. While most connectionist accounts (Rumelhart McClelland 1986, MacWhinney Leinbach 1991, Plunkett Marchman 1993) generally argue that a homogeneous architecture is sufficient for both forms, hybrid theories (Pinker 1991) explain regular cases with a rule and irregulars with an associative memory. There exist, however, modular connectionist (Westermann Goebel 1995) and ....

....for learning the past tense is described that builds its architecture in response to the learning task. The network is compared with three other implemented models of past tense acquisition: the original pattern associator (Rumelhart McClelland 1986, R M) the improved backpropagation network by MacWhinney Leinbach (1991) (M L) which took the extensive criticism of the R M model into account, and the Symbolic Pattern Associator (SPA, Ling Marinov 1993) which took up the challenge posed by M L to present an implemented symbolic system (rather than just a theory) for past tense acquisition. It is shown that the ....

[Article contains additional citation context not shown here]

MacWhinney, B. & Leinbach, J. (1991), `Implementations are not conceptualizations: Revising the verb learning model', Cognition 40, 121--157.

....of the representation and the network architecture that facilitate learning, as well as the opaque knowledge representation of the networks. Several subsequent attempts at improving the original results with new ANN models have been made (Plunkett Marchman, 1991; Cottrell Plunkett, 1991; MacWhinney Leinbach, 1991; Daugherty Seidenberg, 1993) Most notably, MacWhinney and Leinbach (1991) constructed a multilayer neural network with backpropagation (BP) and attempted to answer early criticisms. On the other hand, supporters of the symbolic approach believe that symbol structures such as parse trees, ....

....as well as the opaque knowledge representation of the networks. Several subsequent attempts at improving the original results with new ANN models have been made (Plunkett Marchman, 1991; Cottrell Plunkett, 1991; MacWhinney Leinbach, 1991; Daugherty Seidenberg, 1993) Most notably, MacWhinney and Leinbach (1991) constructed a multilayer neural network with backpropagation (BP) and attempted to answer early criticisms. On the other hand, supporters of the symbolic approach believe that symbol structures such as parse trees, propositions, etc. and the rules for their manipulations, are critical at the ....

[Article contains additional citation context not shown here]

MacWhinney, B., & Leinbach, J. (1991). Implementations are not conceptualizations: Revising the verb model. Cognition, 40, 121 -- 157.

.... mechanisms in language is fundamentally limited (also see Fodor Pylyshyn, 1988) However, many of the specific limitations of the Rumelhart and McClelland model have been addressed in subsequent simulation work (Cottrell Plunkett, 1991; Daugherty Seidenberg, 1992; Hoeffner, 1992; MacWhinney Leinbach, 1991; Marchman, 1993; Plunkett Marchman, 1991, 1993) Thus, the possibility remains strong that such a model could provide a full account of past tense inflection. Furthermore, some recent applications to aspects of language disorders (Hoeffner McClelland, 1993; Marchman, 1993) and language change ....

MacWhinney, B. & Leinbach, J. (1991). Implementations are not conceptualizations: Revising the verb learning model. Cognition, 40, 121--153.

.... localism : the English past tense forms are learned in a network which is dedicated to this morphological task alone (Pinker Prince, 1988) In the Rumelhart and McClelland model, as in most of the succeeding connectionist models of morphology acquisition (Daugherty Seidenberg, 1992; MacWhinney Leinbach, 1991; Plunkett Marchman, 1991) morphology learning consists in learning to map a stem onto an affixed form. If such a network is trained to learn the English past tense, say, how would it benefit from this knowledge in learning the identically formed regular English past participle Pinker and ....

....distinguish morphology from phonology and how the phonological generalizations which characterize the various English s and ed suffixes can be captured in a small set of simple phonological rules. The problem for connectionist models, they argue, is that there is no place for phonology. As MacWhinney and Leinbach (1991) have shown, however, morphological localism is not an inherent feature of connectionist models. By adding a set of input features which distinguish different morphological categories, e.g. past tense and past participle, they give their network the capacity to generalize from one form to ....

[Article contains additional citation context not shown here]

MacWhinney, B. & Leinbach, J. (1991). Implementations are not conceptualization: revising the verb learning model. Cognition, 40, 121--157.

.... localism : the English past tense forms are learned in a network which is dedicated to this morphological task alone (Pinker Prince, 1988) In the Rumelhart and McClelland model, as in most of the succeeding connectionist models of morphology acquisition (Daugherty Seidenberg, 1992; MacWhinney Leinbach, 1991; Plunkett Marchman, 1991) morphology learning consists in learning to map a stem onto an affixed form. If such a network is trained to learn the English past tense, say, how would it benefit from this knowledge in learning the identically formed regular English past participle Pinker and ....

....distinguish morphology from phonology and how the phonological generalizations which characterize the various English s and ed suffixes can be captured in a small set of simple phonological rules. The problem for connectionist models, they argue, is that there is no place for phonology. As MacWhinney and Leinbach (1991) have shown, however, morphological localism is not an inherent feature of connectionist models. By adding a set of input features which distinguish different morphological categories, e.g. past tense and past participle, they give their network the capacity to generalize from one form to ....

MacWhinney, B. & Leinbach, J. (1991). Implementations are not conceptualization: revising the verb learning model. Cognition, 40, 121--157.

....in a voiced phoneme (other than the hissing sounds) are pluralized by adding a [z] sound, and (3) Nouns ending in a voiceless consonant (other than the hissing sounds) are pluralized by adding a [s] sound. Our theory of acquisition differs significantly from those based on statistics (such as [16, 8]) It is incremental, greedy, and fast. It has almost no parameters to adjust. Our theory makes falsifiable claims about the learning of phonological constraints: 1) that learning requires very few examples tens of examples in a few steps as opposed to thousands of examples trained in thousands ....

....bit vector inputs for the performance model and learner. The output of the performance model and learner is a set of bit vectors that typically have a straightforward symbolic interpretation. To test the performance of the program on past tense inflection, we use the same dataset that MacWhinney [8], Ling [6] and Mooney and Califf [11] used. The dataset contains approximately 1400 stem past tense pairs. We randomly choose 500 verb pairs as the test set. The test set is disjoint from the training set. The program is trained on progressively larger samples from the training set. Starting from ....

[Article contains additional citation context not shown here]

Brian MacWhinney and Jared Leinbach. Implementations are not conceptualizations: Revising the verb learning model. Cognition, 40, 1991.

No context found.

MacWhinney, B., & Leinbach, J. (1991). Implementations are not conceptualizations: Revising the verb learning model. Cognition, 29, 121-157.

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