Spreading, Harmony and Agreement
Lately I've been grappling with the differences between the theories of spreading, agreement and harmony. Essentially the three try to capture the regularities of what I'll call feature sharing wherein a segment is made to share a feature or set of features with another segment. What differs between them seems to be a trade off between what can intervene between matching segments versus what degree of similarity the segments must have to be able to match.
Spreading requires immediate adjacency between matching segments with adjacency being determined at the skeletal level. Segments of any type can agree as with CV feature spreading where the matching segments don't need to have anything in common. So with /ki/> [tshi], the palatal feature of the consonant is made to agree with the palatal feature of an adjacent vowel. The adjacency is immediate - the matching segments are completely different (prior to the spreading).
Harmony augments the basic notion of spreading to include adjacency at levels other than the base skeletal strucutre and will often make use of adjacency on the vowel tier (vowel harmony) or adjacency on the consonant tier (consonant harmony). The cost is that segments must be on the same tier to match, thus CV spreading can not be modeled in this framework.
This has been expanded to incorporate feature geometry, enhancing basic tiers to include complex hierarchical relationships wherein a given feature can be matched between two segments if it happens to be "lower" on the tier than the feature they share AND "lower" still than the features of the other segment which would otherwise intervene. Essentially this regulates the kind of tiers that can exist beyond the basic C and V tier.
Recently, agreement by correspondence (ABC Hansson 2001, Rose and Walker 2004) has been used to explain how consonants could match over long distances while skipping intervening segments that couldn't otherwise be accounted for in other theories. Essentially, it seems to allow the establishment of a higher threshhold of similarity (in the guise of correspondance) than basic tiers (which could only do Cs or Vs). What isn't clear to me is how feature geometry couldn't account for the same phenomena. Thus, ABC may establish a correspondence between all corronals so all segments in correspondence (here, corronals) may be made to match w/rt voicing. Thus /t/>[d], /s/>[z] etc. Feature geometry I think could capture the same thing if the voicing node happened to be lower than the corronal node (which it is) and if none of the intervening segments had nodes lower than voicing...
Regardless, it's just a riff on a similar theme where the degree of similarity between matching segments is sacrificed for what is allowed to intervene between matching segments.
So, it seems that ABC simply fiddles with the single parameter to allow for the idea of similarity to be used perhaps defined by typological research where simple or complex tiers used external structures to establish what could match and what could intervene. Ultimately, it seems that ABC will win out because the typological research (see Gunnar Hansson's dissertation in particular) seems to indicate that feature geometry can not capture the agreement found in the world's languages but it's not clear what principles dictate how similarity should be constrained.
Spreading requires immediate adjacency between matching segments with adjacency being determined at the skeletal level. Segments of any type can agree as with CV feature spreading where the matching segments don't need to have anything in common. So with /ki/> [tshi], the palatal feature of the consonant is made to agree with the palatal feature of an adjacent vowel. The adjacency is immediate - the matching segments are completely different (prior to the spreading).
Harmony augments the basic notion of spreading to include adjacency at levels other than the base skeletal strucutre and will often make use of adjacency on the vowel tier (vowel harmony) or adjacency on the consonant tier (consonant harmony). The cost is that segments must be on the same tier to match, thus CV spreading can not be modeled in this framework.
This has been expanded to incorporate feature geometry, enhancing basic tiers to include complex hierarchical relationships wherein a given feature can be matched between two segments if it happens to be "lower" on the tier than the feature they share AND "lower" still than the features of the other segment which would otherwise intervene. Essentially this regulates the kind of tiers that can exist beyond the basic C and V tier.
Recently, agreement by correspondence (ABC Hansson 2001, Rose and Walker 2004) has been used to explain how consonants could match over long distances while skipping intervening segments that couldn't otherwise be accounted for in other theories. Essentially, it seems to allow the establishment of a higher threshhold of similarity (in the guise of correspondance) than basic tiers (which could only do Cs or Vs). What isn't clear to me is how feature geometry couldn't account for the same phenomena. Thus, ABC may establish a correspondence between all corronals so all segments in correspondence (here, corronals) may be made to match w/rt voicing. Thus /t/>[d], /s/>[z] etc. Feature geometry I think could capture the same thing if the voicing node happened to be lower than the corronal node (which it is) and if none of the intervening segments had nodes lower than voicing...
Regardless, it's just a riff on a similar theme where the degree of similarity between matching segments is sacrificed for what is allowed to intervene between matching segments.
| Matching Type | What Can Match | What Can Intervene |
| Spreading/Stricit Locality: | Anything | Nothing |
| Tiers | Segments on the same tier | Segs on diff tiers |
| Feature Geom | Anything | Segments with nodes higher in the hierarchy than the matching segments |
| Agreement by Corr | Similar Segments | Anything (not similar) |
So, it seems that ABC simply fiddles with the single parameter to allow for the idea of similarity to be used perhaps defined by typological research where simple or complex tiers used external structures to establish what could match and what could intervene. Ultimately, it seems that ABC will win out because the typological research (see Gunnar Hansson's dissertation in particular) seems to indicate that feature geometry can not capture the agreement found in the world's languages but it's not clear what principles dictate how similarity should be constrained.
0 Comments:
Post a Comment
<< Home