¹ This paper results mainly from reflections based on fieldwork on Choctaw (Mississippi) and Nunggubuyu (Arnhem Land, Australia). The fieldwork was supported by the American Philosophical Society and by the Australian Institute of Aboriginal Studies. Comments on Turkish, Arabic, and Basque derive from coursework at Harvard and Chicago; I also benefited from a Basque study program in 1972 sponsored by the University of Nevada. R. M. D. Dixon commented on an earlier draft of the paper.

Since the paper was written during a period of prolonged fieldwork which is still continuing, it has been impossible to do adequate library work to test the conclusions offered here on additional languages, and to find out to what extent previous scholars have arrived at conclusions similar to mine. I would, however, like to acknowledge a special (but indirect) debt to Martinet's functional studies of various phonological problems; I believe that his methodological principles will eventually be recognized as having even greater value in the context of morpho-syntax.

² It is of course not strictly true that all languages make these distinctions in their obligatory pronouns; e.g., Spanish neutralizes 1st and 3rd singular in its obligatory pronominals (subject-marking suffixes on verbs) in certain tense-mood paradigms. However, it is probable that all languages make all such distinctions at least in optional pronominals, and that such ‘optional’ pronominals are virtually obligatory when ambiguous obligatory pronominals are not disambiguated by context.

³ It should be mentioned that some Choctaw verbs show number suppletion for subject (intransitive) or object (transitive). Thus it is not quite true that there is no differentiation within the 3rd person category: there is some differentiation, but it is expressed in the verb stem rather than directly by pronominal elements. There is also a pre-verb oklah marking plurality of subject, and some dialects have a corresponding dual post-verb. However, these pre- and post-verbs are not obligatory, and suppletion affects only certain verbs; hence plurality is indicated only in an unreliable fashion.

⁴ This is not intended to be a general denunciation of cognitive analysis of morphological oppositions. This mode of analysis works perfectly well when we are dealing with such phenomena as tense and aspect. I am merely suggesting that cognitive analysis is not in itself adequate in treating pronominal systems.

⁵ The only important exception to the generalization that (nearly) all clause-sequences become strings of subordinated clauses followed by a matrix clause is direct quotative complements, which of course are not affected by any embedding transformations. There is also a minor exception which is more interesting. In 9a, the complement clause is in the ordinary present (unmarked) tense, without subordinating suffixes or particles. Furthermore, it cannot be taken as a direct quotative complement, although the complement of 9b is of this type. There is, however, a very good reason why 9a does not have a formally subordinated complement clause; namely, that the verb -banna- occurs only in type 9a and not in 9b. That is, the very fact that the verb is -banna- already indicates that the subjects of the matrix and complement clauses are coreferential. Since the function of subordinating suffixes and particles is to indicate whether the two subjects are coreferential, they would be redundant in this particular construction, and so are dispensed with.

In this discussion of the Choctaw data, I use the term ‘subject’ in a loose sense, covering not only the subject (nominative) case in nominal morphology, but also the syntactic notions of pivot and controller, all of which are determined by the same rules.

The term ‘switch-subject’ is an adaptation of the term ‘switch-reference’, first suggested by William Jacobsen.