UC Berkeley

Different morphological theories assign different status to parts of words, roots and affixes. Models range from accepting both bound roots and affixes to only assigning unit status to standalone words. Some questions that interest researchers are 1) What are the smallest morphological units, words or word parts? 2) How does frequency affect morphological processing? 3) How do experimental results affect existing morphological theories? In this dissertation I attempt to investigate these questions in more detail using results from psycholinguistic experiments and integrating them into the Network theory of morphology.

I first consider what kinds of information a morphological model should take into account. It should reflect the intuition of speakers that words consist of different parts, while avoiding the difficulties of theories that propose that roots and affixes are separate units. In addition, it should incorporate the notion of paradigms and frequency information. I look at how various morphological theories account for Russian data, in particular verbal prefixes and words where they occur.

The status of roots is intensely debated in morphology, and views of the nature of roots range from being complete separate units on their own, to being considered epiphenomena over words, which are the only existing units. Additionally, it is frequently debated whether free and bound roots have the same status, or are different. In the dissertation I describe the results of a prefix separation experiment in Russian, where more than two types of roots can be identified. The results demonstrate that a productive prefix is separated with greatest easy from words with free roots and with greatest difficulty from words with completely bound roots. The remaining two root types are in between those extremes, according to the characteristics of the root types. I interpret these results as meaning that roots form a continuum based on the strength of connections between words. The stronger the form and meaning connections, the more easily identifiable a root. This interpretation fits well within the Network theory of morphology, where only words are units and there are connections of differing strength between them.

Continuing with the investigation of different units of morphological processing, I report the results of a complexity rating experiment in Russian where words with different prefixes were used. The results demonstrate that ratings of word complexity show a direct correlation with the level of transparency of prefixes in those words. Results of both of these experiments fit into the Network theory of morphology; however, two interpretations are possible with respect to prefixes. While I suggest that roots are not units, it is possible that prefixes, and affixes in general, can be represented as generalizations over many words. It is also possible that they are not units, and that strong connections are formed between words containing those prefixes and new words naturally fall into the pattern to form new words containing the prefixes.

In addition to investigating the unit or non-unit status of roots and prefixes, I study the influence of relative frequency in Russian morphological processing. Studying the data from the prefix separation experiment only with free roots, I compare the processing of words that are more frequent than their bases to processing of words that are less frequent than their bases. The difference between processing of those two sets of words demonstrates that relative frequency is an important factor in Russian morphological processing. These results are in accordance with previous studies in English, Italian and Tagalog, and are thus suggestive of a universal principle of morphological organization.

Finally, I incorporate these results into the Network theory of morphology, where units of morphological processing are words and connections between words. The stronger the connections between words, the more apparent are the different constituent parts. However, the roots and affixes within affixed words are not units of processing, but are epiphenomena over the words and connections. The status of prefixes is not completely clear: clearly, generalizations are formed over prefixes in prefixed words, however, it is not certain whether those generalizations are temporary or permanent. Overall, the Network theory of morphology is revised and updated according to the reported experimental results.