Evaluating alternatives

In terms of simply generating the data, both the syncope and epenthesis analyses work. The question then becomes, is there a reason to choose one of these hypotheses over the other? It is entirely possible that we will not be able to come up with any compelling reasons for selecting one analysis over the other, in which case we must simply accept the fact that there are two equally plausible ways to account for the facts. As far as the simplicity, naturalness, and generality of the two analyses is concerned, neither theory is superior to the other. Processes inserting vowels to break up CCC clusters are very common, as are rules of syncope which delete short vowels in the context VC_CV.

We should also consider the factual predictions of the two analyses. The epenthesis analysis predicts that there should be no CCC sequences in the language, and this appears to be correct. On the other hand, the syncope analysis predicts that there should be no short vowels in the context VC_CV, which also appears to be correct. Interestingly, neither account actually makes the prediction of the competing analysis – so, the epenthesis analysis does not preclude the existence of short vowels in the VC_CV context, and the syncope analysis does not preclude the existence of CCC sequences. If it turns out that there are CCC sequences in the language, the epenthesis solution will probably have to be rejected; whereas if there are VCVCV sequences in the language, the syncope analysis will probably have to be rejected. This would motivate further research into the language, to determine if one of these analyses makes a bad prediction.

A related issue to consider is the question of “coincidence,” in terms of assumed underlying representations. In lieu of a specific rule which restricts the occurrence of phonemes in some environment, we expect phonemes to combine without any constraints. Clearly there must be some constraints on underlying representations in Yawelmani, since, for example, we do not find underlying representations such as /ioate/ with sequences of vowels. In this case, there is no motivation from phonological alternations to suspect that there might be underlying forms such as /ioate/. As far as logical possibilities in underlying forms are concerned for the issue at hand – epenthesis versus deletion – both analyses result in systematic gaps in the logically possible underlying forms. Under the epenthesis analysis, there are apparently no stems of the underlying form /CVCVC/, although there are stems of the form /CVCV:C/. Under the syncope analysis, we notice that all short second-syllable vowels in disyllabic stems are in fact /i/ (surface [u] in some cases, in accordance with vowel harmony).

At this point, it is impossible to give strong arguments in favor of one analysis over another, so we accept this indeterminacy for now. The fundamental point is that each analysis implies a set of predictions about possible and impossible forms in the language, and these predictions need to be tested against the available data. In this case, we have not been able to determine that one theory is clearly superior to the other. The main research problem which we face is that the corpus of data from Yawelmani available to us at this point is restricted, so we cannot know whether generalizations which we extract about the language based on this particular corpus are representative of the language as a whole. Even if we had access to a reference grammar for the language, there is some chance that our empirical generalizations based on the data from that grammar would not hold for the whole language, if the author of the grammar were not aware of all relevant types of examples.