Please post comments and questions on the Week 2 readings (Pater/Lombardi), and on any aspect of class discussion in Week 2 here.
Please post comments and questions on the Week 2 readings (Pater/Lombardi), and on any aspect of class discussion in Week 2 here.
1. Concerning your Austronesian paper — It seems I don’t understand the definition of some of the constraints.
1.1 I am afraid I don’t understand why segmental fusion does not violate MAX.
Why can we not proclaim any case of segment deletion a manifestation of fusion? Indeed, with due ingenuity, one can always find a matching feature between the deleted segment and a surviving one.
2.1 Why nasal substitution does not violate IdentI->O[Nas]? The obstruent that is getting replaced by the nasal lacked [Nas], so there seems to be a mismatch in features between the input and the output.
2 Concerning Lombardi’s paper (these are rather comments than questions):
2.1 She claims that long-distance assimilation doesn’t exist, but aren’t Lyman’s Law effects in rendaku exactly a phenomenon of this type?
2.2 I don’t quite understand how her account matches with the fact that geminated voiced obstruents often become devoiced. It is easy to take care of this positing an appropriate highly-ranked constraint, but this move seems to be merely descriptive.
Many thanks,
Best,
David
Question (for the reading Pater 2004)
I could not understand the formulation (and the application) of the linearity constraint (S1 reflects the precedence structure of S2, and vice versa).
I don’t understand why in (8) both losing candidates are not ruled out by ROOT LINEARITY that is ranked above *NC(O).
I would expect that a deletion of a fragment would create a change the linear order of the elements (those elements that you marked in (7) as 1 and 2). An insertion of an element would be ruled out too if “precedence†means “immediate precedenceâ€.
It is also not entirely clear to me how to distinguish the result of fusion from the result of segmental deletion (the obstruent deletion) in your example in (7).
Thanks,
Katya
I wanted to clarify the definition of Linearity. With respect to example 6, does the winner [memilih] violate linearity once or twice? In addition to the single violation discussed (the nasal N not preceding [p]), there is also [p] which does not precede [i] for this candidate.
I dont follow the ramifications of the string versus segment based formulation of Rootlin. (Section 2.2, p. 275)
Could we also discuss more examples of IO versus OI constraints. I am not sure I follow the advantages of recourse to the IdentO->I[Nas] for Konjo (Section 3.1, p. 279)
Thanks,
Sakshi
With respect to to the Lombardi paper:
– While Harms’ Generalization is a reasonable empirical statement, I’m having trouble seeing how to implement it as a constraint – it seems to be inherently non-local in the sense of needing handle a (theoretically arbitrary) number of segments and make a statement about their total configuration. Are there better ways to derive the generalization from simpler constraints?
– Relatedly, the definition of AGREE here does something similar – how do we formalize what qualifies as an obstruent cluster (or, equivalently, what qualifies as an intervener in such a sequence)?
– Lombardi notes the case of Swedish, in which IDLar >> IDOnsLar. What other good cases are there of general >> specific rankings? How do we account for the apparent rarity of this type?
– Lombardi discusses several aspects of the theory that must be modified at the post-lexical level; what are good tests for whether a process is post-lexical?
Regarding the Pater:
– Do we have any cases in which both IdentI->O(F) and Ident(F) (for the same feature) are active in such a way that we can distinguish their effects? What would that even look like?
Some questions about your 2004 paper:
1) What is sub-segmental linearity? If it pertains to features, which features?
2) Given the definition of Linearity, why is it that deletion and epenthesis do not violate it, even though they do change the precedence structure?
3) p. 278
Two Correspondence Theory approaches in McCarthy and Prince (1994a, 1995) are given as follows:
(i) require mappings between instances of features in the Input and Output.
(ii) set of identity requirements between segmental correspondents.
What is the difference between these two? They look the same.
Also, if Ident[F] already makes reference to the specific position of the correspondents (as opposed to just counting the number of segments in the Output which have changed from the Input), why do we still need Linearity? Will a violation of Lin not be entailed by violation of Ident[F]?
(4) What exactly is meant by the symmetrical nature of Ident?
(5) Why was Ident[Nas] picked, and not, say, Ident[Obst]? Is there some obvious reason for this that I am missing? Am I right in assuming that Ident[Obst] must exist in the language, but it is just not relevant here?
Thanks.
I have a few comments about the Lombardi paper:
Is there any difference in referring to coda position vs. syllable final position? I wonder this whenever I see “syllable final” in a situation where “coda” would be equally felicitous. Is there something theoretical at sake here or is it simply author’s term choice?
On the second page she says some languages do not have devoicing alternations because there are no underlying forms which end in voiced obstruents. She notes this is important and should be accounted for because this is still a distributional pattern which matches the languages with syllable final devoicing. It seems that it is not simply chance that languages exist considering the prevalence of obstruent final devoicing. Do we see in the history of these languages that there may have previously existed final obstruent devoicing and then this rule became integrated into the underlying forms?
On page 287, Lombardi mentions markedness of rankings. She suggests a theory of ranking markedness might be required. Does such a thing exist? How would we be able to justify it? I have wondered about this previously with factorial typology. Typologies give us the possible languages but not which ones are more common or likely to occur. Some rankings are more unmarked in the sense that they are more common. What type of theory can account for this?