Conspiracy Argument for OT from Emakhuwa Dialectology

Farida Cassimjee and Charles W. Kisseberth
Benedict College/ Tel Aviv University

Halle (1962), in trying to motivate Generative Phonology, spends what might appear to be an inordinate amount of space (for a relatively short article) demonstrating how a rule-based model such as GP provides significant insights into dialect variation. In developing Optimality Theory, Prince and Smolensky (1993) make little reference to dialectal variation, though they do place considerable emphasis on how the "factorial typology" provided by the theory predicts the cross-linguistic variation in certain sub-domains of phonology. In this paper, we direct attention back to the specifically dialectal implications of the theory.
We argue that (i) a rule-based model fails to characterize what the Pemba dialect shares tonally with dialects such as Eerati, Imithupi, Enlai; (ii) the rule model's failure is exactly akin to its failure to deal with examples like the syllable structure conspiracy in Yawelmani (discussed in Kisseberth (1970)); and (iii) OT provides a straightforward account of how Pemba is connected to the other dialects due to the very essence of the theory: by virtue of constraint interaction, a single constraint may both bar a phenomenon from occurring (do something except when) and demand that a phenomenon occur (do something only when). For several Emakhuwa dialects, there is clear evidence, in rule-based phonology, for rule (1). We call this rule High Tone Doubling.

(1) High Tone Doubling A High tone spreads onto the immediately following mora.

In most dialects, High Tone Doubling will fail to apply when (i) the second mora is IP-final, and (ii) the second mora is part of a long syllable that is IP-penult; in a less widely distributed number of dialects, High Tone Doubling will fail to apply (iii) when the second mora is in any IP-penult syllable. These restrictions can be built into the formulation of (1), or they could be gotten at through the formulation of rules that undo (1). The latter analysis is possible since (1) derives structures that do not exist in the language except as a consequence of (1). It is immaterial to this paper which type of analysis of these restrictions is given. In some dialects, there is a second rule, call it Delinking:

(2) Delinking Delink the left-branch of a multiply-linked H tone.

However, in the dialects containing rule (2), this rule never applies across-the-board. Specifically, it must be blocked from applying (i) to the first mora of a bimoraic syllable and (ii) to a mora that in turn is preceded by a H-toned mora. Dialects that have (2) as a categorial or a gradient rule include Eerati, Imithupi, and Enlai.

There is only one way to deal with the restrictions on (2). Delinking must be blocked by conditions on the Delinking rule. If one attempted to delink by one rule and then re-insert association lines by other rules into the relevant positions (first mora of a bimoraic syllable and between High tones), one cannot insert them correctly by any plausible rule. The argument that Delinking in these dialects must be blocked (by a preceding High tone and by virtue of being located on a bimoraic syllable) is crucial to our overall point.
There are a number of varieties of Emakhuwa (e.g. those in central region of Nampula province) do not have High Tone Doubling at all. In these dialects, underlying H tone specifications simply sit on a vowel and they is no further phonology of note. However, thereis one dialect (that spoken in Pemba town in Cabo Delgado) that we have come across that does not have High Tone Doubling in any general fashion, but where there are two instances of violations of tonal faithfulness: namely, there is spreading from the first mora to the second mora of a bimoraic syllable (subject to exactly the same constraints as in the other dialects) and there is spreading from one mora to another mora just in the case the second mora is itself followed by a High tone!
Rule-ordering describes the Pemba dialect as having a rule that spreads just in case the targetted mora is followed by a H; it describes Eerati/Enlai/Imitthupi as having a rule that delinks unless the preceding mora is H. These dialects simply have two different rules (spreading vs. delinking) with quite different looking conditions. From the point of view of rule-based phonology, these are simply arbitrary dialectal differences with no connection. There is no shared rule. Optimal Domains Theory [=ODT], a particular version of OT, permits an account of Emakhuwa dialectology where the relationship between Pemba and the other dialects is made transparent. The same well-motivated universal constraint, one we call Plateau (*H0H), serves to drive both the appearance of a H tone in Pemba and block the loss of a H tone in Eerati/Enlai/Imithupi. In other dialects (e.g. Ikorovere, Esaaka, Imeetto), the generality of the "doubling" phenomenon masks the effects of Plateau. Only in a range of dialects without doubling, and without any effects of Plateau, must Plateau be ranked below faithfulness (e.g. the dialects of central Nampula province).
We conclude there is a conspiracy argument for OT based on Emakhuwa dialectology. It is the consraint-based OT, and not the rule-based GP, that provides significant insight into the welter of dialectal variation.

< Back to List of Abstracts | Back to PLC23 Home Page >

About the PLC23 Committee
Previously held Penn Linguistics Colloquium: PLC22 (1998), PLC21 (1997)

Penn Department of Linguistics
University of Pennsylvania