The Identification of Bases in Morphological Paradigms

Adam Albright
University of California, Los Angeles
June, 2002

A dissertation submitted in partial satisfaction
of the requirements for the degree
Doctor of Philosophy in Linguistics
UCLA Linguistics Department

Thesis committee:
Bruce Hayes, co-chair
Donca Steriade, co-chair
Carson Schütze
Brent Vine

[abstract] [text]


Many theories, in many domains of linguistics, assume that some members of morphological paradigms are more basic than others. Bases of paradigms are privileged in various ways: they may determine phonological properties of other forms, they may determine the direction of analogical changes, and so on. In this thesis, I propose that such effects are a result of the procedure by which learners seek to develop a grammar that allows them to project inflected forms as accurately and confidently as possible. I present a computationally implemented model of paradigm acquisition that attempts to use one form in the paradigm as the base to project the remaining forms, using stochastic morphological rules. I pursue two hypotheses about how this is done. The first is that learners are limited to selecting a single form as the base, and that the base must be a surface form from somewhere within the paradigm. Furthermore, the choice of base is global, meaning that the same slot must serve as the base for all lexical items. The second hypothesis is that learners select the base form that is maximally informative, in the sense that it preserves the most contrasts, and permits accurate productive generation of as many forms of as many words as possible.

As evidence for this approach, I analyze three cases in which an typologically marked form served as the base of a historical analogical change: Yiddish present tense paradigms (in which all forms were remodeled on the 1st sg), Latin noun paradigms (in which nominatives were remodeled on oblique forms), and Lakhota verbs (in which unsuffixed forms are being remodeled on suffixed forms). In each case, I show how the model correctly selects the base form, and also correctly predicts asymmetries in the direction of subsequent paradigmatic changes. I show that these asymmetries are not predicted by a more traditional model of underlying forms, in which learners compare all of the parts of the paradigm to construct abstract underlying representations that combine unpredictable information from multiple forms. Finally, I discuss possible extensions of this model to accommodate larger paradigms with multiple, local bases.

Text for download

  Front matter     124Kb
  Chapter 1 Introduction pp. 1-16 324Kb
  Chapter 2 Paradigm leveling in Yiddish pp. 17-34 308Kb
  Chapter 3 Identifying bases algorithmically pp. 35-52 284Kb
  Chapter 4 The Latin honor analogy pp. 53-74 380Kb
  Chapter 5 Extension of ablaut in Lakhota pp. 75-98 404Kb
  Chapter 6 Discussion pp. 99-128 668Kb
  Chapter 7 Conclusion pp. 129-132 104Kb
  Appendices Appendix A: Results for the Synthetic Neutralize I Language pp. 133-154 184Kb
Appendix B Metrics for base selection in Latin pp. 155-156
  References     128Kb