|Title:||On the logic of conditional grounding|
|Comment:||Paper presented at CLS 37, 2001 (and to appear in the proceedings thereof)|
|Abstract:||In a language with vowel harmony, a certain class of vowels may systematically fail to harmonically alternate, resulting in predictably disharmonic forms. Such vowels are generally agreed to be incompatible with one of the harmonic feature values, preventing harmony from affecting them. For example, in many languages with [±atr] harmony, low vowels do not alternate; they are systematically disharmonic, always surfacing as [-atr] even in [+atr] harmonic contexts.
Systematic disharmony is commonly analyzed with implicational statements (called 'grounding conditions' by Archangeli & Pulleyblank (1994)) such as 'if [+lo], then [-atr].' Stating these conditions in implicational form suggests that the antecedent feature (in this case, [+lo]) is held constant while only the consequent feature ([-atr]) can be altered to satisfy the condition. However, there is another relevant pattern that cannot be accounted for this way.
In Diola, [±atr] harmony in forms with low vowels results not in systematic disharmony but in an alternation between a [+lo, -atr] vowel and a [-lo, +atr] vowel. This pattern clearly has the same underpinnings as systematic disharmony: a [±atr] alternation between low vowels would require a [+lo, +atr] vowel. However, it seems that this incompatibility cannot be due to 'if [+lo], then [-atr]' in Diola, because the antecedent feature is altered to satisfy the condition. The Diola pattern must instead be due to 'if [+atr], then [-lo].'
From a nonprocedural point of view such as that provided by OT, however, it becomes clear that these two conditions are logically equivalent, and so can be unified to account both for the Diola pattern and for cases in which low vowels are systematically disharmonic with respect to [±atr] harmony. The difference between the two patterns is shown to follow from different constraint rankings (rather than different constraints), and an analysis is presented of the conspiracy evident in two languages, Maasai and Turkana, which exhibit both patterns in different contexts.