Christiansen, Morten, and Chater, Nick. Generalization and Connectionist Language Learning. In Mind and Language, 1994.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....such as Elman (1993) Elman s network learned a language containing only 23 words, and yet 50,000 sentences were used to train the net work. This means that every word could have been observed in every syntactic position may times over, greatly reducing the need to form generalizations. Christiansen and Chater (1994) investigated to what extent this kind of model was able to generalize to predict that a word observed in one syntactic position would also be grammatical in another position. In order to do this, they trained a similar connectionist network on a more complex language containing 34 words, again ....

Christiansen, M. H. & Chater, N. (1994). Generalization and Connectionist Language Learning. Mind and Language, 9, 273-287.

....these, owing to Phillips (1994) is restricted in scope to simple sentences (containing no relative clauses) Another (Niklasson and van Gelder, 1994) does not address language learning, but confines itself to a very narrow range of logic formulae with at most one level of embedding. By contrast, Christiansen and Chater (1994) have extended the work of Elman (1993) on predicting grammatical categories. They employ a (relevantly) restricted training corpus containing sentences generated from a moderately complex, context free grammar. Like Elman, they augment their network s memory during training by increasing the ....

Christiansen, M. H., & Chater, N. (1994). Generalization and connectionist language learning. Mind and Language, 9, 273--287.

....decade, simple recurrent networks (SRNs) have attracted considerable notice among connectionists involved in language acquisition research. In no small measure, this attention is due to the work of Elman (1990, 1993) and those who have adopted his temporal sequencing strategy (for examples, see Christiansen Chater, 1994; Christiansen, Chater, Seidenberg, 1999) However, despite a number of intriguing results, reservations about the applicability of SRNs, within the domain of language learning, persist. Some of these misgivings concern the range of syntactic patterns (Marcus, 1998; Rispoli, 1999) that SRN s can ....

....it will be helpful to consider some of the issues surrounding systematicity. Between 1990 and 1994, substantial research was directed at demonstrating the capacity of SRNs to predict the syntactic category of the next word in a sentence, given some current word of input (cf. Elman, 1990, 1993; Christiansen Chater, 1994). In addition, since about 1990, a number of publications have described networks which achieve some degree of syntactic or semantic systematicity (cf. Chalmers, 1990; Christiansen and Chater, 1994; Elman, 1998; Hadley, 1994a, 1994b; Hadley Hayward, 1997; Hadley Cardei, 1999; Niklasson and van ....

[Article contains additional citation context not shown here]

Christiansen, M.H. & Chater, N., 1994, Generalization and connectionist language learning. Mind and Language, 9: 273-287.

....if one can make inferences of the form Q P; Q P . Subsequently, Hadley (1994) characterized systematicity as generalization to novel syntactic position, based on a review of language learning. Researchers have demonstrated networks supporting this definition of systematicity to various degrees (Christiansen Chater, 1994; Hadley Hayward, 1994; Niklasson van Gelder, 1994; Phillips, 1994) However, others 2 question whether the empirical evidence supports this definition either way, given the difficulty of controlling subjects background knowledge and observing what knowledge they have acquired in the course ....

....a possible 104 patterns, the network correctly predicted the remaining 4 test patterns) and in light of evidence discussed here, different to that of humans. The limitation of the feedforward network on Klein 4group tasks also extends to the simple recurrent network (Elman, 1990) which others (Christiansen Chater, 1994; Niklasson van Gelder, 1994) have shown exhibits (to some degree) the definition of systematicity put forth by Hadley (1994) A simple recurrent network applied to the Klein 4 group task includes all the weights and units of the feedforward network, plus additional weights for mapping the ....

Christiansen, M. H., & Chater, N. (1994). Generalization and connectionist language learning. Mind and Language, 9 (3). Preprint.

....layer) which is Theta(k 3 ) 14 . Since the number of free parameters is greater, generalization across position is less likely given the same training set. So, a fully connected higher order network is unlikely to support strong systematicity. Simple recurrent network on word prediction Christiansen and Chater (1994) test for strong systematicity in the simple recurrent network on a word prediction task. They claim some limited supported for strong systematicity in one of two experiments, but analysis here shows no support for strong systematicity. A simple recurrent network was trained to predict words in ....

Christiansen, M. H., & Chater, N. (1994). Generalization and connectionist language learning. Mind and Language, 9 (3), 273--287.

....CA 90089 2520 morten gizmo.usc.edu mcm gizmo.usc.edu Abstract We present results from three psycholinguistic experiments which tested predictions from a connectionist model of recursive sentence processing. The model was originally developed to capture generalization using non local information (Christiansen, 1994; Christiansen Chater, 1994) From this model it was possible to derive novel empirical predictions concerning the processing of different kinds of recursive structure. We present behavioral results confirming network predictions about the acceptability of sentences involving multiple ....

....mcm gizmo.usc.edu Abstract We present results from three psycholinguistic experiments which tested predictions from a connectionist model of recursive sentence processing. The model was originally developed to capture generalization using non local information (Christiansen, 1994; Christiansen Chater, 1994). From this model it was possible to derive novel empirical predictions concerning the processing of different kinds of recursive structure. We present behavioral results confirming network predictions about the acceptability of sentences involving multiple right branching PPs (Experiment 1) ....

[Article contains additional citation context not shown here]

Christiansen, M.H. & Chater, N. (1994). Generalization and connectionist language learning.

.... arrows indicate trainable weights, whereas the thin arrow denotes the copy back weights (which are always 1) Shillcock, Lindsey, Levy Chater, 1991; Norris 1993) to the modeling of a mapping from meaning to sound (Cottrell Plunkett, 1991) to syntactic structure (Christiansen, in preparation; Christiansen Chater, 1994; Elman, 1991, 1993) Fairly extensive studies have also been conducted on their computational properties (Chater, 1989; Chater Conkey, 1992; Christiansen Chater, in submission; Cleeremans, Servan Schreiber McClelland, 1989; Cottrell Tsung, 1993; Maskara Noetzel, 1993; Servan Schreiber, ....

Christiansen, M. & Chater, N. (1994). Generalization and connectionist language learning.

....clauses (allowing multiple embeddings with complex long distance dependencies) These simulations suggested that an SRN can acquire some of the grammatical regularities underlying a grammar. In addition, Elman s SRN showed some similarities with human behavior on center embedded structures. Christiansen (1994, 1998) extended this work, using more complex grammars involving prenominal genitives, prepositional modifications of noun phrases, noun phrase conjunctions, and sentential complements, in addition to the grammatical features used by Elman. One of the grammars moreover incorporated ....

....and moreover that they exhibit the same qualitative processing difficulties as humans do on similar constructions. The nets also showed sophisticated generalization abilities, overriding local word co occurrence statistics while complying with structural constraints at the constituent level (Christiansen Chater, 1994). Current models of syntax typically use toy fragments of grammar and small vocabularies. Aside from raising the question of the viability of scaling up, this makes it difficult to provide detailed fits with empirical data. Nonetheless, some attempts have recently been made toward fitting ....

Christiansen, M.H. & Chater, N. (1994). Generalization and connectionist language learning.

No context found.

Christiansen, Morten, and Chater, Nick. Generalization and Connectionist Language Learning. In Mind and Language, 1994.

No context found.

Christiansen, Morten and Chater, Nick, 1994. "Generalization and Connectionist Language Learning." Mind and Language 9:273--287.

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