Opaque feeding interactions

Just a day after I posted the answer to the “A counterbleeds B” conundrum, Elliott Moreton — not having read the answer yet — wrote to say:

My money is on the later rule’s counterbleeding the earlier one.

Casual followers of the discussion thus far will have figured out that Elliott doubled his money. More careful readers will also note that I more recently cited a 2002 paper co-authored by Elliott and Paul Smolensky in which the relevant phrase “A counterbleeds B” is used incorrectly in two different ways.

  1. A precedes B, so B should counterbleed A. Elliott’s e-mail to me clarifies that he understands that the later rule (B) counterbleeds the earlier rule (A), so it’s somewhat surprising that he got this wrong in the paper.

  2. The rules A and B cited by Moreton & Smolensky are not even in a counterbleeding relationship, as it turns out. If anything, A feeds B (as I clarified before and am about to clarify again).

The relevant example is from Turkish. Moreton & Smolensky cite Sprouse (1997), who writes (p. 3):

In Turkish an epenthetic vowel is required to break up certain disallowed coda consonants clusters. Yet this vowel sometimes triggers the deletion of one of the offending consonants.

Here are some relevant facts presented by Sprouse. The 2nd person singular possessive suffix has a postvocalic allomorph [n] (araba ‘car’, araba-n ‘your car’) and a postconsonantal allomorph [Vn], where V is a high vowel that harmonizes for backness and rounding (temel ‘foundation’, temel-in ‘your foundation’). The vowel is due to epenthesis, because underlyingly high vowel initial suffixes undergo glide insertion postvocalically (araba-yım ‘I am a car’, temel-im ‘I am a foundation’). However, intervocalic stem-final /k/ deletes in Turkish, and this deletion is triggered by epenthesis: bebek ‘baby’, bebe-in ‘your baby’.

So when Moreton & Smolensky (2002:10) state that “epenthesis counterbleeds a consonant-deletion rule in Turkish”, this is not accurate: epenthesis feeds deletion, because the application of deletion depends on the prior application of epenthesis (in this context); if deletion applied first, it wouldn’t bleed epenthesis because deletion simply wouldn’t apply.

However, I think the incorrect branding of this as a case of counterbleeding has some justification. First, feeding and counterbleeding form a natural class (made most clear by Kiparsky (1968), which I discussed here): these are the two types of rule interaction in which both rules get to apply to forms on which their structural descriptions overlap. (With bleeding, the second rule is blocked by the application of the first rule; with counterfeeding, the second rule creates the right conditions for the first rule, but too late.) Second, feeding and bleeding are generally associated with transparency of rule interaction, whereas counterbleeding and counterfeeding are generally associated with opacity. The striking property of the epenthesis + deletion case in Turkish is precisely that it’s an opaque feeding interaction: the application of deletion destroys the context that was necessary for the prior application of epenthesis. So, the reasoning Moreton & Smolensky followed was probably something like this:

  1. Both rules apply, so it’s either feeding or counterbleeding.

  2. The interaction is opaque, so it’s counterbleeding or counterfeeding.

  3. Therefore, it must be counterbleeding.

So why are opaque feeding interactions unfamiliar? I think it’s because typical examples of feeding interactions are between rules that are in what I will call a same-segment-feeding relation: the first rule changes X to Y in some set of contexts, and the second rule changes Y to Z in some overlapping set of contexts, so the feeding interaction can be fully described as the derivational trajectory taken by a single segment: X → Y → Z.

In the Turkish example, the rules are not in a same-segment-feeding relation; rather, the change made by the first rule (epenthesis) simply creates an appropriate context for the change to be made by the second rule (deletion), which in turn happens to destroy the context that motivated the first rule in the first place. These types of feeding interactions (whether or not they are also opaque) can only be described by taking the context into account — e.g., PXR → PYR → QYR; and whether or not the interaction is opaque depends on whether or not P was a conditioning factor in the change of X to Y.

Note that the only way that a same-segment-feeding relation X → Y → Z can be opaque is if X = Z, in the strictly relevant sense that Z readopts the feature(s) of X that conditioned the change to Y — in other words, only if you have a Duke-of-York derivation. (The link just there is not to John McCarthy’s well-known work on this topic or to Geoff Pullum’s original paper but rather to a 2003 University of Essex dissertation by Russell Norton, which I have to admit I haven’t yet read — but given how interested I’ve gotten in this topic, I plan to.)

Update 1, 6/6/2005:

Talk about a coincidence. A week or so after posting this, I’m making photocopies for class in my departmental copy room and I spy a copy of Phonology 19, Number 3 (2002) in the middle a stack of other papers and books that looks like it’s been sitting there since … well, 2002. This is the issue with, among other good articles, Chris Potts and Geoff Pullum‘s article on model theory and OT constraints, which — esp. given the coincidence — I decide to look at first.

In their discussion of sympathy constraints in section 7.2 (p. 384ff), Potts & Pullum bring up the same Turkish epenthesis and deletion example discussed above. They draw the example from René Kager‘s (1999) OT textbook, and describe the relationship between the rules this way (pp. 384-385):

Turkish exhibits a classic counterbleeding relationship between epenthesis and deletion.

So not only is it counterbleeding, it’s classic counterbleeding? How can this be? I don’t think I’m wrong in assuming that a truly “classic” example of counterbleeding is something like the typical rule-based analysis of Austronesian nasal substitution:

  1. Nasal Place Assimilation: [+nas] → [x-place] / __ [x-place]
  2. Voiceless Stop Deletion: [-voi] → ø / [+nas] __

This is “classic counterbleeding” because if the rules are reversed, then Voiceless Stop Deletion would bleed Nasal Place Assimilation. This is not the case in the Turkish example, as I’ve already explained (note that these rules are rough approximations):

  1. Epenthesis: ø → [+high] / C __ C
  2. Deletion: /k/ → ø / V __ V

If the rules are reversed, then Deletion simply wouldn’t apply in the relevant set of contexts — much less bleed Epenthesis, which would apply unimpeded. True, if the naked result of Deletion (meaning: ignoring the fact that it can only apply intervocalically) were passed on to Epenthesis, this latter rule would be bled, but we’re not dealing with naked results when we’re talking about rule interactions. This is not a classic case of counterbleeding in any sense; if anything, it’s an opaque feeding interaction. I’m surprised that this point escaped Geoff’s usually overly-scrutinous attention.

Update 2, 3/2/2006:

Another interesting example of this type is found in a Tagalog problem set that I often use in introductory phonology classes. This time, it’s an epenthesis rule that feeds a deletion rule (the reverse of Turkish), and Deletion renders Epenthesis opaque.

  1. Epenthesis: Ø → h / V __ V
  2. Deletion: V → Ø / VC __ CV

So, /polo+in/ → |polohin| (by 1) → [polhin] (by 2). (Gloss: ‘ask for trifles’)

Even more interesting, perhaps, is the Tiberian Hebrew example made infamous by John McCarthy:

  1. Epenthesis: Ø → e / C __ C#
  2. Deletion: ʔ → Ø / __ {C, #}

So, /deʃʔ/ → |deʃeʔ| (by 1) → [deʃe] (by 2). (Gloss: ‘tender grass’)

The reason I think this case is perhaps more interesting is because, as stated, Deletion counterbleeds Epenthesis. However, there can never be any evidence that Deletion would otherwise apply in the C __ # context, since Epenthesis always applies there. In other words, if Deletion were stated such that it applies in codas (roughly, V __ {C, #}), then Epenthesis actually feeds Deletion! (And of course, either way, Epenthesis is rendered opaque by Deletion.)

(Note that McCarthy appears to side-step the issue (p. 3 of the preprint, p. 333 of the article in Phonology) by saying that Deletion applies “outside onsets” rather than “in codas”. I wonder if that was purposeful.)

3 thoughts on “Opaque feeding interactions

  1. Pingback: phonoloblog»Blog Archive » Talk about a coincidence

  2. Charles Reiss

    I was wondering about Eric’s solution for Tagalog summarized here:
    “Another interesting example of this type is found in a Tagalog problem set that I often use in introductory phonology classes. This time, it’s an epenthesis rule that feeds a deletion rule (the reverse of Turkish), and Deletion renders Epenthesis opaque.

    1. Epenthesis: Ø → h / V __ V
    2. Deletion: V → Ø / VC __ CV

    So, /polo+in/ → |polohin| (by 1) → [polhin] (by 2). (Gloss: ‘ask for trifles’)”

    I always teach the Tagalog problem as involving word-final deletion of [h], not insertion. Is there a consensus on this? Any good reasons? I have a vague recollection that justified the analysis with MSCs: if you assume final /h/ in the relevant roots, and you assume glottal stops before all the apparently V-initial ones, you get all roots to have the shape CVCVC-. I am not saying that this is a good argument for h-deletion, but I think it is what I was taught at some point.

  3. Eric Bakovic Post author

    Good point, Charles — I left that alternative out here. I usually either assign this problem set to my undergraduates or provide it as a practice problem set with a sample solution, including some discussion of both analyses. The latest version of the practice problem set + solution handout is here, if you’re interested.

Leave a Reply

Your email address will not be published. Required fields are marked *