No abstract

Patterns of vowel harmony are frequently interrupted by the presence of neutral segments, segments that are obligatorily realised with only one of the harmonic values. Peripherally, neutral segments appear in two patterns, referred to as relative and absolute alignment. Medially, such neutral segments may be opaque, interrupting the transmission of harmony, or they may be transparent, skipped over by harmony. It is argued that the properties of neutrality result from the interaction of three independently motivated families of constraints: faithfulness, alignment, and grounding. No process-specific constraints distinguishing between the types of alignment or between opacity and transparency are required under the proposed account.

The canonical image of vowel harmony is of a particular feature distributed throughout a word, leading to symmetric constraints like AGREE or SPREAD. Examination of the distribution of tongue-root advancement in Kinande demonstrates that harmonic feature distribution is asymmetric. The data argue that a formal (yet asymmetric) constraint (like ALIGN) is exactly half right : such a constraint correctly characterises the left edge of the harmonic domain. By contrast, the right edge is necessarily characterised by phonetically grounded restrictions on feature co-occurrence. Of further interest is the role of morphological domains : the interaction between domain restrictions on specific constraints and unrestricted constraints suggests a formal means of characterising the overwhelming similarity between constraint hierarchies at different morphological levels while at the same time characterising the distinctions between levels.

'Harmony' is a widely attested pattern in natural language, a configuration where within some domain all eligible anchors for some feature bear the same feature value. Typically, a harmony system exhibits a choice between two feature values. Either all anchors within some domain D bear the feature value F or all anchors within D bear the opposite value G. Depending on the theory of features and harmony, both harmonic values may be overtly specified or one may be indicated by the absence of featural specification. In exploring here the option of deriving harmony by prohibiting feature disharmony, the functional motivation is that the resetting of articulatory targets costs the grammar.

Transparency—in which a harmony effect passes over a segment without affecting it phonetically or phonologically—has been a controversial concept in previous literature on harmony systems. A typical case of so-called transparency involves cross-height vowel harmony in Kinande, a Bantu language (J.40). Previous accounts have analyzed low vowels in this system as being transparent to harmony [Schlindwein, NELS 17, 551–567 (1987)]. Further, some analysts have considered low vowels theoretically incapable of undergoing tongue root harmony. These claims were tested in a single-subject field study using ultrasound imaging to measure tongue root position in low vowels. Results indicate that (a) advanced versus retracted tongue root position (ATR) is a viable feature for describing the phonological distinction in the vowel system; (b) there is a phonetic difference between low vowels when adjacent to ATR triggering vowels; (c) this distinction in low vowels does not decrease with distance from trigger vowels, suggesting that these vowels are undergoing phonological harmony rather than phonetic assimilation; and finally, (d) the ATR distinction is phonetically categorical in high vowels, but shows crossover in mid and low vowels. Implications for phonological theory and phonetics-phonology interface will be discussed. [Work supported by NSERC and SSHRC.]