From Joe Pater (pater@linguist.umass.edu)
I’d like to ask whether anyone knows of examples of two apparently unattested phonological patterns, and also raise the issue of what their unattestedness means for accounts of unbounded feature spreading.
The better-known of the two is the Sour Grapes pattern, a gap brought to our attention by Eric Bakovic, John McCarthy and Colin Wilson. It has the following description: spread a feature unboundedly, unless there is a blocking segment in the potential span, in which case don’t spread at all. I had never seen a case like this, until I read Adam Jardine’s recent paper which points to Bickmore and Kula’s work on Copperbelt Bemba, which seems to display a Sour Grapes effect (in tonal spreading).
The less well known gap is what Kevin Mullin and I are calling the “Use it or Lose it” pattern: spread a marked feature if there is an available target, but if there is not, delink it. Spreading into a licensing position for a marked feature, like a stressed syllable, is attested at least for some features and some positions, and so are the related neutralization patterns (see Rachel Walker’s book), but the pattern we’re pointing to doesn’t require that the target position for spreading be a licensing position. Kevin and Colin Wilson seemed to have independently noticed that this pattern is generated by McCarthy’s (2011) Share account when a markedness constraint is included in the constraint set. We talk about the pattern, and propose a means of avoiding it, in this handout from MFM 2015.
Assuming that these two gaps are real, for at least non-tonal features, they seem to point in different directions for our understanding of harmony, at least in OT. Sour Grapes seems to indicate that there needs to be a positive benefit for each instance of spreading. Use it or Lose it seems to indicate that this benefit cannot override a compulsion to delete a marked feature (as Kevin and I note, it seems to be a general problem for positive spreading constraints, including also Align and reward-assigning Spread). Our solution is to make the choice to parse a feature the first step in spreading it, but I’m sure there are other ways of getting a system that simultaneously avoids both kinds of overgeneration (Wilson suggests that Targeted Constraints can do the job, for instance).
For “use it or lose it”, you should look at Hyman’s 1998 paper on height harmony in Yaka. If Yaka is like other progressive height harmony patterns in Bantu, then Yaka shows evidence for sour grapes. Hyman argues against this position, claiming that harmony is regressive in Yaka. One result of this is that final /e/ is raised to [i] if it cannot spread it’s [-high] feature, which looks like “use it or lose it.” It’s interesting to consider the effect of one’s analytical choices on pathological predictions here.
Second, there are two Ghana-Togo Mountain languages of southeastern Ghana that show what looks like “sour grapes.” One of them is Tafi, and is described in Bobuafor (2013). Bobuafor does not address the issue, but it is clear that when first person pronominal prefixes co-occur with [-high] prefixes, that leftward [+ATR] spreading from the root is blocked. This harmony is blocked regardless of what intervenes between the conditional blockers and the [+ATR] root. To save you a little time searching through her description, consider the following examples (tone has been omitted). [+hi] subject prefixes regularly alternate for ATR, as in (1).
(1) a. i-di-bhui ‘1S-NEG2-cut’ (p. 330)
b. ɪ-dɪ-tsÉ” ‘1S-NEG2-do.early’ (p. 341)
In (2), we see that FUT also regularly alternates for ATR.
(2) a. e-be-dzu ‘3S-FUT-put-down’ (p. 380)
b. a-ba-kÉ” ‘3S-FUT-give’ (p. 376)
However, when a [+hi] subject prefix and a [-hi] TAM prefix, like FUT, co-occur, [+ATR] does not spread leftward from the root.
(3) a. ɪ-ba-dzu ‘1S-FUT-build’ (p. 89)
b. lɪ-ba-zi ‘3S.DEP-FUT-be’ (p. 264)
c. kɪ-ba-dzi ‘3S-FUT-become’ (p. 273)
d. kɪ-vla-bhui ‘3S-again-cut’ (p. 346)
Note that we can’t call [-hi] vowels blockers because [-hi] regularly undergoes and spreads [+ATR] to other vowels, even to [+hi] vowels, so long as they aren’t subject pronominals. Look at (4) to see that cross-height harmony is readily attested in the language.
(4) a. e-ti-be-teÅ‹u ‘3S-NEG1-FUT-can’ (p. 229)
b. e-ti-be-bi ‘3S-NEG1-FUT-be.cooked’ (p. 280)
There is one instance in Bobuafor’s grammar with partial spreading, rather than the failed spreading we see in the examples above, which is shown in (5).
(5) ɪ-ka-vle-hu ‘1S-PERSIST-again-hit’
I think there is more going on in the language, but from these examples, as well as from a larger collection of data from neighboring Tutrugbu that James Essegbey and I are working on, there is something akin to sour grapes in these two languages.