Assimilation to the Unmarked

Eric Bakovic
Harvard University

Assimilation processes are commonly triggered by positions of morphological or prosodic prominence; e.g., affixes typically alternate to agree with roots, and codas typically alternate to agree with onsets. Within Optimality Theory (Prince & Smolensky 1993), such positional effects have been fruitfully accounted for by positional faithfulness constraints (Beckman 1998): constraints demanding contrast maintenance in prominent positions in the face of contrast neutralization elsewhere. But in Swedish (Hellberg 1974, Lombardi 1997), coda obstruents do not necessarily agree with following onset obstruents; if either the coda or the onset is voiceless underlyingly, an obstruent cluster surfaces voiceless. Thus, we have hš[g] 'high' but hš[kt]id 'festival', with assimilation to the onset in voicelessness, and [d]ag 'day' but ti[st]ag 'Tuesday', with assimilation to the coda in voicelessness. This pattern is very similar to dominant-recessive vowel harmony systems like that of Kalenjin (Hall et al. 1974, Ringen 1988), in which affix vowels do not necessarily agree with root vowels; if either a root or an affix vowel in Kalenjin is [+atr], then all vowels in the word surface as [+atr]. Thus, we have /kI - A - ke:r - In/ -> [kiage:rin] 'I saw you', with assimilation to the root /ke:r/ 'see' in [+atr], and /kI - A - kEr - e/ -> [kiagere] 'I was shutting it', with assimilation to the noncompletive suffix /e/ in [+atr].
Lombardi's analysis of the Swedish facts involves the subordination of the positional faithfulness constraint IdentOnset[voi] to a markedness constraint against voiced obstruents, which is in turn dominated by the general faithfulness constraint Ident[voi]. An undominated constraint Agree[voi] demands that obstruent clusters agree in voicing; given a cluster consisting of a voiced obstruent and a voiceless one, regardless of which is in the onset and which is in the coda, Agree[voi] demands that one of the obstruents change to agree with the other. Whether the voiced obstruent changes to agree with the voiceless one or vice-versa, Ident[voi] is equally violated, and so the decision is passed on to the markedness constraint, deciding in favor of a voiceless cluster as the optimal candidate. An entirely parallel analysis can be applied to Kalenjin. The general faithfulness constraint Ident[atr] dominates a markedness constraint against [-atr] vowels, which itself dominates the positional faithfulness constraint IdentRoot[atr]. An undominated constraint Agree[atr] demands that all vowels in word agree in terms of [atr]; given a pair of vowels that disagree in terms of [atr], Agree[atr] demands that one of the vowels change to agree with the other. Whether the [-atr] vowel changes to agree with the [+atr] one or vice-versa, Ident[atr] is equally violated, and so the decision is passed on to the markedness constraint, deciding in favor of a pair of [+atr] vowels in the optimal candidate.
Lombardi notes that part of this account makes an odd prediction: in clusters of more than two obstruents, the high rank of Ident[voi] has a 'majority rule' effect. If there are more voiceless obstruents than voiced ones in the input, an all-voiceless cluster is optimal, whereas if there are more voiced obstruents than voiceless ones, an all-voiced cluster is optimal. This is because there is no tie on Ident[voi] in such situations; the more segments that are changed, the more this constraint is violated, and so the fewer segments that need to be changed to satisfy Agree[voi], the better. Such a predicted state of affairs would be surprising (as Lombardi notes), though the relevant examples in Swedish are not easily constructed; however, the parallel situation in terms of vowels and [atr] clearly does not hold in Kalenjin, as shown by the examples above: the presence of a single [+atr] vowel in a word causes as many vowels as necessary to become [+atr] as well.
I propose that the local conjunction (Smolensky 1993) of markedness and faithfulness (Lubowicz 1998) is necessary to remedy this problem. Specifically, the suboptimal all-voiced candidate from a mixed-voice input fatally violates a locally-conjoined constraint against voiced obstruents that are not voiced in the input; *[+voi] & Ident[voi]. Locally-conjoined constraints are universally higher-ranked than their conjuncts, and so *[+voi] & Ident[voi] rules out any candidate with a change from voiceless obstruent to voiced obstruent before Ident[voi] has a chance to make a decision of its own - regardless of the relative number of voiced and voiceless obstruents in the input - correctly and universally circumventing the 'majority rule' effect. This solution applies equally to Kalenjin, the relevant locally-conjoined constraint being *[-atr] & Ident[atr]. The advantages of this proposal over alternative solutions will be discussed at length in the talk.

Selected References

Beckman, Jill N. 1998. Positional Faithfulness. Doctoral dissertation, University of Massachusetts, Amherst.
Hall, Beatrice L., R. M. R. Hall, Martin D. Pam, Amy Myers, Stephen A. Antell, and Godfrey K. Cherono. 1974. African Vowel Harmony Systems from the Vantage Point of Kalenjin. Africa und †bersee 57, 241-267.
Hellberg, Steffan. 1974. Graphonomic Rules in Phonology: Studies in the Expression Component of Swedish. Acta Universitatis Gothoburgensis, Gšteborg.
Lombardi, Linda. 1997. Positional Faithfulness and Voicing Assimilation in Optimality Theory. To appear in NLLT.
Lubowicz, Anna. 1998. Derived Environment Effects in OT. ROA-239, Rutgers Optimality Archive, http://ruccs.rutgers.edu/roa.html.
Prince, Alan and Paul Smolensky. 1993. Optimality Theory: Constraint Interaction in Generative Grammar. RuCCS Tech Report #1, Rutgers Center for Cognitive Science, Piscataway, NJ. [To appear, MIT Press.]
Ringen, Catherine O. 1988. Underspecification Theory and Binary Features. In H. van der Hulst and N. Smith (eds.), Features, Segmental Structure and Harmony Processes (Part II), 145-160. Foris, Dordrecht.
Smolensky, Paul. 1993. Harmony, Markedness, and Phonological Activity. Paper presented at the Rutgers Optimality Workshop, Rutgers University. [ROA-87, Rutgers Optimality Archive, http://ruccs.rutgers.edu/roa.html.]


< 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