I had lunch with some non-linguists today, and the conversation turned to calling people by their initials. Some interesting intuitions show up which appear to be linguistic in nature, though somewhat gradient. Here’s the deal: we know we can assign initials-based referring expressions using the first letters of the referent’s first and middle or first and last name. But there appears to be some limits on what constitutes an allowable set of initials. The example at lunch was, MJ is an allowable form, but MN and ML are not. I have some ideas about why, but it’s not so simple.
For now, let’s put aside written usage, such as email sign-offs or psych journal citations, since presumably writing allows any combination of 2 letters. Speech is different, since we pronounce the letters by their names: [ei, bi:, si:, di:, i:, ɛf, dʒi:] and so on. For this discussion, I’m thinking of the kinds of initials you could use over the phone, as in “Is MJ there?”.
My first thought was that you want to avoid initials if the spoken form of any of the letters is VC, as in [ɛm], [ɛl], or [ar]. Obviously the acceptability of MJ thwarts this, so maybe it only applies to the second letter. But JR is out there, so the story needs to be more complicated.
The second hypothesis is that initials are forbidden if both letters’ names are VC. In other words, initials can have no more than one of the letters AEFILMNORSX (and I’m gonna go right out and say H and W just can’t be used at all – can you imagine saying “is HG there?” over the phone?). A discussion of the formal analysis is below, but first, I need to admit some additional considerations.
First, I invite comments about these intuitions. Basically, the prediction is that initials like FT, MT, ST, TF, TQ, TX, and so on are acceptable as over-the-phone usages. FL, MR, MF, and so on are not.
Second, it shouldn’t take long to find counterexamples in which two VC letters do sound alright next to each other. FM, XM, MS, and MX are some that do sound OK to me. I suspect this is the case because each is an abbreviation for some other non-onomastic use; FM and XM for radio, MS for citations and Microsoft, and MX for the legendary Reagan-era nuclear missile.
Let’s then agree that these exceptions are frequency-driven; they are common enough to drown any objections we have to them as novel initials. Now we can stick with the no-more-than-one-VC-component hypothesis.
OK, so to the analysis. In OT terms, this looks like a weird cumulative effect, where one instance of whatever’s wrong with [ɛm] or other VC letters (be it violations of NoCoda or Onset) is not enough to sink the form, but two instances are. In the case of two violations, the null parse is the output. But strict dominance isn’t supposed to let that happen: If Faithfulness outranks NoCoda and Onset for one VC, it does for the other too. So there is a problem somewhere in the analysis.
Strict dominance, an inherent aspect of constraint ranking and hence a lynchpin of OT, is too restrictive. We can follow this to its logical conclusion, that OT can’t handle this system.
I identified the wrong output constraints as players.
Possibility 1 might be tempting, but I think Possibility 2 is the story here. This was an easy trap to fall into, given the ease with which we can divide names of letters of the alphabet into VC and CV categories. Phonetically, ML would be something like [ɛmɛl], but the syllabification is relevant. Let’s allow either [ɛ.mɛl] or an ambisyllabic [ɛm.mɛl]. Next, let’s call each ‘letter’ a separate morpheme in the construction. Then the boundary between the two morphemes in ML is within a syllable: [ɛm.m+ɛl]. The new analytical hypothesis is that ML is a bad pair of initials because the letter-names and syllable boundaries are misaligned.
In contrast, MJ has a morpheme boundary occurring at a syllable boundary, as in [ɛm+.dʒei]; any other combo of VCCV should follow suit. Likewise, maybe JR also has its morpheme boundary aligned with the syllable boundary, as in [dzei+.ar]. Any other combo of CVVC should follow suit. Problem solved, strict ranking intact.
I’m going to call it the moral of the story that as a generality, Possibility 2 (you got the wrong constraints) should be pursued exhaustively before accepting Possibility 1 (the data confound the model), whether the model is OT or any framework. I’m also going to follow the moral from EB’s post and refrain from attempting to quantify how many people accept Possibility 1 without exploring Possibility 2.