Flapping without alternations

Over on Language Log yesterday, Mark Liberman noted an interesting eggcorn: deep-seeded for deep-seated. Today, Sally Thomason follows up with a post asking a question that is very appropriate for phonoloblog: in the case of taps in the flapping environment within morphemes (e.g., metal, meddle, mettle, medal) what is underlying — a tap, a /d/, or a /t/?

Of course, different approaches to phonology offer different answers to this kind of question. Somewhere in K&K’79 (I don’t have it to hand at the moment, but I’ll supply the exact quote and page ref. later See update below), the authors write that the underlying representation should be identical to the surface (their “phonetic”) representation unless there is evidence to the contrary.

Update: The K&K’79 page number is 141, and the actual claim is: “Unless there is evidence to the contrary, the UR of a morpheme is assumed to be identical with the phonetic representation.” This appears in the “Corpus-Internal Evidence” section of Chapter 5, “Evidence and Motivation”. Two related claims are also made and discussed in this section: “Each morpheme is assumed to have a unique UR unless there is evidence to the contrary.” (p. 140) and “All other things being equal, a phonological solution is preferred over a solution that divides the lexicon into arbitrary classes (a lexical solution) or over a solution that lists the morphological/syntactic contexts in which a rule applies (a grammatical solution).” (p. 142).

(This basic heuristic was enshrined under Lexicon Optimization in OT, but let’s not go there right now.) Since the relevant parts of these morphemes the morphemes mettle and meddle never alternate (cf. metal ~ metallic, medal, medallion), the only “evidence to the contrary” is that the distribution of taps is entirely predictable and, in morphemes that do exhibit the requisite alternation, the relevant phonemes surface as either [t] or [d] (modulo aspiration, release, preglottalization, etc.), which are (uncontroversially) the basic allophones of their respective phonemes.

But these considerations do not settle the question in this case. Most folks would probably not posit /mɛɾəl/ for any of the aforementioned morphemes (though this is what Lexicon Optimization would demand, if you take it seriously), but then again there is some ambiguity about whether the tap in each case is basically a /t/ or a /d/. Is it /mɛt&#601l/ or /mɛt&#601l/? Does it even matter?

Sally tells the story about how she once hypothesized that speakers probably base their underlying representations on the orthography: /t/ for metal and mettle, /d/ for medal and meddle. This seems reasonable to me (though again, does it even matter?). Then she thought: wouldn’t it be great to see what illiterate English speakers do with such words? So she did a little experiment with some kids under six, having them slowly repeat words of the relevant type. Sally’s the first to admit that her experimental methods were not beyond reproach, but the result was still pretty interesting: all of the kids pronounced all of the words with an unmistakeable [t].

Sally concludes: “What does this mean? I haven’t the faintest idea.” Neither do I, but it’s interesting. I’m generally skeptical of the emergent-phoneme-in-hyper-slow-and-careful-speech business, but still, there’s something here and it’s interesting. One can probably take a guess that it has something to do with the relative unmarkedness of voiceless stops vs. voiced stops, etc., but does that really explain it?

And yet again: does it even matter? Maybe to our mental representations, but to phonological theory? I don’t think so, but I’m ready and willing to be proven wrong on this point.