Donca Steriade
MIT/UCLA
Department of Linguistics and Philosophy

Maryland Linguistics Colloquium

Friday, October 26, 2001)
2:00 pm
Room 1304, Marie Mount Hall
University of Maryland, College Park


Context dependent similarity evaluations: the phonetics and phonology of half-rhymes

This study is a contribution to Correspondence Theory (McCarthy and Prince 1995): it seeks to provide empirical evidence for the proposition that correspondence constraints are context-dependent and that their rankings are partly predictable from perceived similarity relations among different string types (Steriade 2001, in Hume and Johnson eds.). For many phonological processes, the exact division of labor between markedness and correspondence constraints is uncertain: the same process can frequently be described as the interaction of context-free correspondence with context sensitive markedness constraints or, alternatively, as the interaction between context-free markedness with context sensitive correspondence.
One can make progress in resolving this type of ambiguity by focussing on phenomena where markedness plays no role and similarity alone determines the empirical outcome. With this in mind I study imperfect rimes (semi-rimes, SR): the starting assumption being that the poet's observed preference for one SR type over another reflects the poet's judgment that the better SR's involve more similar strings and thus a rhyming pair closer to perfect identity. Relative similarity will be reflected in correspondence rankings: if Corr1 >> Corr2 (where Corr1 and Corr2 are distinct correspondence constraints evaluating the identity between the rhyme domains of two lines) then a pair of lines violating Corr1 will be, all else equal, less harmonic and thus dispreferred to a pair violating Corr2.
The talk examines SR distributions in modern (19th-20th centuries) Romanian poetry. Romanian is selected because its frequent feminine rhymes allow us to examine the relative distribution of SR's based on the same feature in a wide variety of contexts. I review three sources of SR evidence (rhymed translations, native poetry, and a rhyme dictionary written by a poet for his own use) which converge on identifying those SR types that are both frequent in individual SR corpora and documented across all SR corpora. To evaluate the relative frequency of different SR types I compare, within a given corpus, the ratio of all rhymes containing a certain feature (say obstruent voicing) to SR's based on mismatches of this feature (e.g. SR's involving differences of obstruent voicing).
The immediate goals of the project are to determine whether changes of context induce different evaluations of phonological similarity and, if so, whether the different evaluations of similarity are tied to phonetic, context-dependent differences in the realization of the relevant features or to other factors. The interest in the possibility of context-dependent similarity is directly tied to the debate between the proponents of positional (i.e. context sensitive) markedness and those supporting positional correspondence. I argue that the latter conditions are necessary by observing that positional correspondence conditions must be operative in the analysis of rhyming.
The following is one example of the evidence I analyze. I use the formula D(x-y) to refer to the phonological distance or dissimilarity between two rhyme domains differentiated by the strings x and y and the formula D(x-y) < D (x-z) to abbreviate assertions of the form "the dissimilarity between x and y is less than that between x and z". The evidence for these assertions is the relative frequency of corresponding SR types in our corpora: our assumption is that more frequent SR's are those judged to involve more similar rhyme domains.

(Abbreviations: T: voiceless obstruent, D: voiced obstruent, N: nasal, V vowel, # end of word; rhyme domains are underlined in the examples.)
D (-NT#)-(ND#) < D (-NTV)-(NDV) < D (-VT#)-(VD#) < D (-VTV)-(VDV)

For instance, the SR type timp-skimb (based on the difference (-NT#)-(ND#)) is more frequent that the SR-type púnte-askúnde, which is in turn significantly less frequent than the type vød-føt. The type skáde-láte occurs but is least frequent. I interpret this result as indicating that voicing differences are perceived as least distinctive in post-nasal, word final stops, more so in prevocalic and postvocalic position. An acoustic study of Romanian stop voicing, carried out in collaboration with Jie Zhang, reveals the existence of gradient post-nasal voicing in Romanian, a possible source of the greater perceived similarity between T-D in the post-N context.


If you are interested in meeting with the speaker, please contact Scott Fults or Nina Kazanina.

To go back to the Colloquium schedule, click here.