This experiment shows what regular past-tense rules are acquired from common verbs and their past-tense forms. Like the plurals, English past tense has regular affixes [t], [d], or [I.d] depending on the properties of the last phoneme of the verb stem. Verb stems that end in a voiced phoneme other than [d] receive [d], while those that end in an unvoiced phoneme other than [t] receive [t]. For verb stems ending in [d] or [t], the syllable [I.d] is added to the stem.
Unlike those in the plural rules, the exceptions in the regular past tense rules (for stems ending in [d] or [t]), do not form a natural class in terms of distinctive features. It is therefore interesting to see how the learner constructs disjunctive rules to capture the regularities in the input data.
The input consists of 21 verbs and their past-tense forms:
[t] [d] [I.d] irregular
------------------------------------------------------------
danced answered added draw/drew
dropped called needed sing/sang
fixed cried painted feed/fed
kissed hugged waited
laughed turned
liked
looked
touched
walked
The stem-past pairs are presented sequentially in a random order once. After the presentation of all the verb pairs, the learner has acquired six rule-classifiers and three exceptions (the irregulars). The phoneme bits of the classifiers are as follows:
1. [dc.dc.[+voice,+sonorant].d]
2. [dc.dc.[+voice,-coronal].d]
3. [dc.dc.[-low,-round,-tense,+continuant].d]
4. [dc.dc.[-voice,+strident].t]
5. [dc.dc.[-voice,-coronal,-continuant].t]
6. [dc.{d,t}.I.d]
Rule-classifiers 1, 2, and 3 together cover all the verb stems that end in a voiced phoneme other than [d]. These rule-classifiers overlap in the examples they cover. Rule-classifier 1 covers the majority of these cases (all the vowels, nasals, liquids, and glides). Rule-classifier 2 covers the voiced non-coronal stops ([b] and [g]) as well as some of the cases covered by rule-classifier 1, while rule-classifier 3 covers the voiced stridents. An over-general rule-classifier [dc.dc.[+voice].d] is acquired after the presentation of 11 stem-past pairs. The learner refines the over-general classifier to rule-classifier 1 on examining another 6 stem-past pairs (not all of which are relevant to rule-classifier 1).
Similarly, rule-classifiers 4 and 5 cover verb stems that end in an unvoiced phoneme other than [t]. Rule-classifier 4 covers stems ending in [k] or [p], while rule-classifier 5 covers the unvoiced stridents. Rule-classifier 4 is acquired after 8 stem-past pairs. Rule-classifier 5 is acquired after 11 pairs.
Rule-classifier 6 directly corresponds to the add [I.d] rule. This classifier is acquired after 18 stem-past pairs.
The experiment shows that even though word stems ending in [d] or [t] do not form a natural class, the learner can still acquire past-tense rule-classifiers (in the form of disjunctive rules) that lead to the correct behavior.