Empirical Studies On English Past Tense Verbs In Second Language Environment

Past tense in English verbs, as an important branch of morphology study, has received numerous and profound studies. Berko's Wug-test has demonstrated the overregularization phenomenon in child language acquisition. It shows that they are ready to add regular "-ed" suffix to novel verbs as the past tense. Besides this study, some field studies have shown the "U-shaped development profile" among children. At the beginning, the children speak out a verb like "went" correctly, a stage of error-free. Then after they have realized the vast majority of past tense has the "-ed" ending, they would apply it to all verb forms, like "goed", when the overregularization appears. Later they find "go" is an exception to the rule. The correct form reappears and keeps stable.Scholars make detailed studies of the second stage in child language acquisition. Two schools of models apply their theories in examining these errors. One of them is the Dual-mechanism Model argued by Pinker et al. There are two layers working between regular form and irregular inflection respectively. One is "symbolic rules", operating by adding an "-ed" suffix to a class of abstract mental words. Therefore it leads to overregularization. Children would be ready to apply the rules to all the verbs they meet. The second layer is "associative memory", corresponding irregular tenses to the basic forms. During the process, irregular inflected forms cannot coexist with their regular counterparts. When the regular form has dominated the only entry to the cell, say the lexical system, the irregular one is blocked. It remains stable unless continuous attack from the outside input shows it is the wrong one. It mainly depends on the frequent input to stimulate memory system in reaching the correct past tense. The sub-regularities mapping a cluster of phonologically similar verbs as "sing-sang, drink-drank" are partly associative. But meanwhile it leads to overregularization as "ding-clang" instead of the correct form as "ding-dung".The other theory extends from Parallel Distributed Processing, called as Single-connectionist Model. It is a neural network trained to map phonologicalrepresentations of stem forms into output representations of past tense forms. By adopting a general error correcting learning method, the statistical relations between the input stem phonemes and the output form phonemes are encoded in some links. These learned associations enable the model to generalize the regular past tense from novel verbs. Thus the network exhibits behavior that can be described as rule-governed even though no rules are involved in the actual generation. The whole process of learning mimics the U-shaped curve characteristics of children. Statistical learning is one feature that proves the importance of frequency. The errors in neighborhood structure, such as "glow-glowed" to "blow-blowed", are corrected, which adds the new links to the system.Considering the two theories, I design a questionnaire of 43 verbs, including 35 irregular and 8 regular. It is responded to by two classes of Grade 2 Chinese students. It is to test whether in the second language environment the errors would be like that among children as native English speakers, in aspects of inclination to "-ed" suffix, frequency and neighborhood structure. Meanwhile, it aims at investigating whether some new types of errors occur among these Chinese students.Some findings have been obtained as follows: 1) The overregularization, as inclination to -ed" suffix, is also common among the Chinese students. 2) Frequency plays an essential role in correctness. The more frequently used verbs get a lower percent of error numbers. 3) The neighborhood structure, as the sub-regularity functions, misleads the students to produce errors analogical to the findings from previous studies. Meanwhile these students give 10 percent of the errors in a new way. They would adopt the past tense form with letter "0", considering the stem with "i" and "e". It is more obvious than the errors submitted in native English speaker...