2.3 Diagnostic Devices

Two important diagnostic devices have been commonly employed for the processing of regular and irregular forms: frequency effects and phonological similarity effects.

2.3.1 Frequency Effects

Studies (e.g.,Rubenstein, Garfield & Milliken, 1970) show that the more often a given word is encountered for a learner, the better it will be remembered. That is, human memory traces generally become stronger with repeated exposure. The likelihood of the storage and retrieval of a word in an associative memory increases with its frequency. Thus, the high-frequency forms are likely to be remembered more firmly and retrieved more quickly, while the low-frequency forms are likely to be remembered more weakly and retrieved more slowly.

The single-mechanism approach posits that the human language is processed in a single route, and both regular and irregular forms are stored and retrieved by only an associative memory or a rule system. Therefore, the connectionist models assume that both regular and irregular forms show frequency effects, that is, both regulars and irregulars should be better remembered and more quickly retrieved if they are more encountered. By contrast, Generative Phonology assumes that neither regular nor irregular forms show frequency effects. On the other hand, if the dual-mechanism theories are correct and regular forms are computed by a rule system in real time while irregular forms are memorized in the lexicon and retrieved via an associative memory, then regular forms are free from the frequency effects but irregular forms are expected to be frequency sensitive, with the high-frequency forms being remembered better and retrieved more quickly than the low-frequency forms.

2.3.2 Phonological Similarity Effects

Similarity, also called neighborhood or family resemblance, in the current models accounting for the processing and representation of the human languages refers to the identicalness or similarity of some sounds between different words. Thus, it is generally a phonological phenomenon. Both regular and irregular forms show phonological similarity. For example, the irregular verbs such as cling, fling, sting, sling, swing and wring share the same sound[iη], and their past tense forms show the same pattern by changing i into u, thus resulting in similarity among these verbs. And the regular verbs like slap, clap, trap and flap share the same sound[æp], and their past tense forms are all generated by adding the suffix-ed, so these verbs also show phonological similarity.

The memory traces representing the phonological stem-past pairings shared among “friendly” neighboring irregulars or regulars should be strengthened by the learning of any of these pairings. For instance, hearing cling-clung should not only strengthen the memory trace of this pair, but also those with phonological similarity like fling-flung, sting-stung and sling-slung. However, each pattern with phonological similarity has some exceptions. These “enemy” neighbors have similar stems but different inflected forms and thus weaken each other.

The single-mechanism model assumes an associative memory system or a rule system for both regular and irregular forms. Therefore, the connectionist models assume that phonological similarity effects are expected for all the inflected forms, but Generative Phonology assumes that neither the regulars nor the irregulars show phonological similarity effects. By contrast, if the dual-mechanism models are on the right track and irregular forms are computed in the associative memory with distributed phonological representations, whereas regular forms are rule-products, then phonological similarity effects should be found only for the irregulars but not for the regulars.

2.3.3 Summary

To summarize, Generative Phonology assumes no frequency effects or phonological similarity effects for the regulars and irregulars, while the connectionist models assumes frequency effects and phonological similarity effects for both regulars and irregulars. Different from both Generative Phonology and the connectionist models, the dual-mechanism approach assumes frequency effects and phonological similarity effects only for the irregulars, but not for the regulars.