Bob is absolutely right — I grossly mischaracterized his Lexicon Underoptimization idea. (Bob concludes that my mischaracterization was “derived ultimately from [his] earlier failure to elucidate [his idea] adequately”, but I think he’s just being kind here; given that there was not a single mention of markedness in Bob’s original post, I had little if any justification for linking his thoughts about underspecification to any notion of markedness.)
Bob also makes very clear the issue with Lexicon Optimization that we’ve been discussing. When unleashed on morphemes with nonalternating [ɾ], Lexicon Optimization selects /ɾ/ and not /t/ or /d/ as the underlying representation; assuming that Sally Thomason’s experimental results tell us that the underlying representation of such forms actually has /t/, then (obviously) Lexicon Optimization makes the wrong choice. Bob’s Lexicon Underoptimization idea is meant to address exactly this problem: “when positing underlying representations, remove any feature specification that is not needed to generate the proper output”.
With all that cleared up, I still have questions — and I hope I don’t just mischaracterize Bob’s position again.
It’ll help to do a thought experiment first. In Sally Thomason’s study, the finding was that kids under 6 pronounced [ɾ] as [t] when asked to repeat a set of relevant words “very, very slowly”. Note that if the kids had pronounced [ɾ], we might not even be talking about this (though we should probably still find it curious, given that taps are very quick, ballistic gestures that are most comfortable when between vowels). If the kids had pronounced [d], we might have concluded that [d] sounds most like [ɾ] (a fact I invoked here) — but we would not be tempted to invoke underspecification. Why not? Bob’s reasoning is that there’s something [d] has that [t] lacks: voicing. As well-accepted as this reasoning is, I find it hard to pin down its justification.
First of all, the idea that voicing is “something” and that voicelessness is a lack of that “something” is not as obvious as it may seem; both results involve laryngeal gestures (stiff vs. slack vocal folds) that could, for all we know, be “specified” on both segments. (This, by the way, is why I mistakenly latched on to the markedness thing before — it was the only possible justification I could see for assuming that voicing is something that voicelessness is not.)
Second, as we all know, the /d/~/t/ contrast in English is not usually realized as a contrast in voicing; rather, there’s a difference in voice onset time. If we say that /t/ is generally realized with a delay in voice onset time that /d/ generally lacks, does that justify the assumption that /d/ is underspecified relative to /t/? I don’t think so, but there’s no obvious reason why voicing should trump aspiration in this case. (Note that Geoff Nathan’s post is relevant to this point: Geoff reports that David Stampe observed kids who had “slightly aspirated [t]’s” in the relevant contexts — in fact, this is a more accurate representation of Helen’s speech, and I’m sure also of Sally Thomason’s subjects.)
Geoff’s post highlights the final point I’d like to make. The kids that David Stampe observed were not replacing [ɾ] with (slightly aspirated) [t]; their parents had glottal stops in the flapping context. Can we confidently assert that [t] is somehow less specified than glottal stop? And for that matter, what does [ɾ] have that [t] lacks, exactly? I would think that, if anything, each has at least one “thing” that the other doesn’t; how would Lexicon Underoptimization decide which specifications are removeable and which aren’t? I just don’t see how it would work, aside from just stipulating it or (as I mistakenly did) linking it to something like markedness.
I raise these issues not because I think underspecification is simply wrong, but rather because I don’t think it’s fun. I know first-hand what a hard business underspecification is, full of contradictions and blind alleys and difficult assumptions about what phonological representations consist of and what their relationship to phonetics is. This [t]-for-[ɾ] discussion has been interesting (to me, at least) in large part because it brings out all of those problems. If there’s anyone out there who’d like to suggest solutions, feel free.