Title: Extending the Past-tense Debate: a Model of the German Plural - eScholarship
Abstract: Extending the Past-tense Debate: a Model of the German Plural Niels A. Taatgen ([email protected]) Artificial Intelligence, University of Groningen Grote Kruisstraat 2/1, 9712 TS Groningen, the Netherlands Abstract One of the phenomena that has been studied extensively in cognitive science is learning the English past tense. Many models have been made of the characteristic U-shape in per- formance on irregular verbs during development. An important test case for such models is whether they can be extended to other examples of inflection. A case that is often quoted as par- ticularly tough is the German plural. In the present study, an ACT-R model of the past tense is applied to the German plural. The model not only successfully learns the default rule, but also exhibits some other characteristics of the German plural. Introduction Learning the English past tense has been one of the central topics of debate in cognitive science since McClelland and Rumelhart published their original neural network model in 1986. The phenomenon is very simple. English verbs can be broken down into two categories: regular and irregular verbs. The past tense of a regular verb can be obtained by simply adding -ed to the stem. Irregular verbs on the other hand are unsystematic: each verb has a unique inflection. When chil- dren have to learn the inflection of the past tense, they go through three stages. In the first stage their use of the past tense is infrequent, but when they use the past tense they do so correctly. In the second stage they use the past tense more often, but they start overregularizing the irregular verbs. So instead of saying broke, they now say *breaked. On the other hand, inflection of regular verbs increases dramatically, indi- cating that the child has somehow learned the general regular pattern. In the third stage, they inflect irregular verbs cor- rectly again. This pattern of learning is often referred to as U-shaped learning. Although learning the past tense seems to be a rather sim- ple problem, it nevertheless encompasses a number of issues in language acquisition and learning in general. Apparently the past tense has two aspects: on the one hand there is a gen- eral rule, and on the other hand there is set of exceptions. Children are able to learn both aspects, and the phenomenon of U-shaped learning seems to implicate that children learn the general rule in stage 2. The important point McClelland and Rumelhart make is that this does not necessarily imply that this knowledge is actually represented as a rule in the cognitive system: their neural network model has no separate store for rules, but it nevertheless exhibits rule-like behavior in the form of U-shaped learning. Ever since their original model, the neural network approach has been challenged (e.g., Pinker & Prince, 1988), improved (e.g., Plunkett & Marchman, 1991), challenged again (e.g., Marcus, 1995) and improved again (e.g., Plunkett & Juola, 1999). I would like to highlight two unresolved issues in this debate, because they will be addressed here. The first issue is feedback. A well-known fact in language acquisition is that children do not rely on feedback on their own production of language (at least with respect to syntax), simply because they do not receive any (Pinker, 1984). Although this problem is addressed by some modelers (e.g., Plunkett & Juola, 1999), its resolution is not entirely satisfactory: the assumption is that learning takes place while children perceive past tenses, and not while they actually produce past tenses. This idea is at odds with the picture of skill acquisition in general, where practice is considered as a main means of learning. A second issue is the frequency of the regular cases. In English, most verbs are regular. This fact is essential for neural network models, as they need to be presented with regular cases at least 50% of the time (Marcus, 1995). This is already slightly problematic in English, as the token-frequency of regular verbs, how often a verb is actually used in language, is only around 30% (irregular verbs are just used much more often than regulars). Connectionist modelers have therefore intro- duced the input/uptake distinction: not every word that is perceived is presented to the network. This assumption becomes especially problematic if regular forms are much more rare. An example of inflection where the regular form is very rare is the German plural. The German Plural German has five different suffixes to mark plurality of a noun: zero (no suffix), -(e)n, -e, -er and -s. Moreover, the stem-vowel sometimes receives an Umlaut (¨), something we will ignore for the present. The plural is almost always indi- cated by suffixation: there are only a few exceptions, mainly words derived from Latin (e.g., Thema-Themen). Careful analysis of these suffixes has revealed that the -s suffix is actually the default rule (Marcus, Brinkmann, Clahsen, Wiese & Pinker, 1995). Interestingly enough, this suffix is also the least frequent of all five, both in type-frequency (how many words are there) and token-frequency (how often are they used). Marcus et al. estimate the type frequency of nouns ending in -s at 4%, and the token frequency at only 2%. It appears however that at least some of the other suf-
Publication Year: 2001
Publication Date: 2001-01-01
Language: en
Type: article
Access and Citation
AI Researcher Chatbot
Get quick answers to your questions about the article from our AI researcher chatbot