Floating gallows

[Emma May Smith]

What I was hoping to dig into was the question of whether any discernable "rules" of gallows insertion would emerge out of these statistics.  Maybe [cheey] is unusually flexible.  But [chedy], for example, is less so -- we can find [kchedy], [cKhedy], and [chkedy] (and [t] equivalents), but inserting a gallows to either side of the [d] seems dispreferred, except for a single token of [chedty].  The words [chey] and [chdy] follow a similar pattern: for [chey] we have [kchey], [cKhey], [chkey], [cheky] (and [t] equivalents); but for [chdy] we have only [kchdy] and [cKhdy] (and [t] equivalents).  When the combination [oe] appears in a word, it looks as though a gallows might be drawn preferentially to the spot between them, and also to be especially likely to appear: hence, possibly, the preference for [cheokey] and [chotey], and also for [qokeedy] (291 tokens) as opposed to [qoekedy] (2 tokens) or [qoeedy] (18 tokens).  Whether this kind of analysis will turn out to have any advantage over other ones, I don't think I'm yet in any position to say, but I thought it could be worth a try.

I quite understand. I suppose we have two possibilities for discerning rules:
1. That the forms themselves show a clear preference/dispreference for place of gallows insertion.
2. We analogise from non-ambiguous situations.

My guess is that we normally do 2), assuming that analogies hold and we can create rules from strong prohibitions in some circumstances and transfer them to other environments. So we see that [lchey] exists (and is very common) but [clhey], [chley], and [chely] don't exist and [cheyl] has one token, so assume that the form [Xchey] is not the same as [cXhey] (or other). Now, [l] isn't a gallows, I know, but it forms part of the information we use to understand word structure. We would otherwise be stating that the [chey] in [lchey] is not the same as the [chey] in [kchey]. But is that any better (or worse) than saying that the [chey] in [Xchey] is not the same as [cXhey]?

The question is whether these assumptions about analogising word structure are correct and whether they've been adequately tested. I think your hypothesis is a good challenge and, while I believe we have the word structure broadly correct, I'm not sure we can prove it in every instance. Can we break down the question about word structure and gallows insertion into smaller parts?