Pitfalls of rule-based phonology, and surface constraints as alternative.

A recap

Pieces on the board

(1) Units of phonological structure:

(2) An example tree:

Rules of the game

• We assume there are two components to the sound of an utterance:

  1. What is arbitrary: what makes the words in the utterance those words, as opposed to any other; the semantic/pragmatic categories reflected in the prosody.

  2. What is predictable: the generalisations that are true over all environments that have x, y and z phonological units in common. Examples: AmEng /t,d/-flapping, absence of [ŋ] from English syllable onsets.

• We assume that it’s only the arbitrary part that a speaker has to remember. The phonology part of the language faculty can fill in all the predictable bits. The abstract, arbitrary part stored in the mental lexicon is the underlying representation (UR).

  The principle that underlying representations are minimal has been challenged by those who note that speakers seem to have access to phonetically detailed memory of a word as they have previously heard it uttered. On this basis, exemplar-based models of phonology have been proposed in which the lexical representation of a word consists of a “cloud” of previously heard or uttered tokens (see Johnson 1997; Pierrehumbert 2001, 2002).

• The output of a phonology is the surface representation (SR). A good SR should contain everything a speaker needs to know to produce an impeccably native-sounding utterance of the UR it is derived from.

  Many analyses in phonology use the kinds of units in (1) for both UR and SR, and leave the job of translating the SR into articulatory gestures to a separate phonetic implementation module. For a radically different approach, you might read about Articulatory Phonology, which argues that phonological representations (at all levels) should be defined in terms of articulatory gestures.

  For a summary of Articulatory Phonology written by its partisans, see Browman and Goldstein (1992). The responses to the article in the rest of Phonetica vol. 49 are also worth reading. More references can be found at http://www.haskins.yale.edu/research/gestural.html.

• The predictable part of the SR is baked into it by phonological rules. Examples:

  (3)

  	consonantal
  	obstruent		→	[+spread glottis]	/	#
  −	voice

  (see Lecture 2)

  (4)

  	obstruent		→	∅	/		cons			[cons]
  +	voice						nasal

  (see Lecture 3)

  General form: A → B / C ＿ D, where:

  • A, B are phonological units specified in the terms of (1)

  • C, D are sufficient to describe the environment of the generalisation.

So what’s wrong with that?

If your goal is to describe, for example, The Sound Pattern of English, then you can use these principles to do it (Chomsky and Halle 1968 certainly did).

On the other hand, if your goal is to produce a theoretical model of how the brain decides what noises the articulators to make to express the meaning encoded in a given string of words, or produce a theoretical model that will do that job as well as a brain would—in other words, if you are trying to explain phonology, then the rule-based generative model has some problems:

• The general problem with transformational grammar applies: rules are too powerful. You can in principle turn any UR into any SR, and have unnatural rules like [voc] → [nas] / [strid] ＿ [plos]. (In words: “nasalise a vowel between a strident fricative and a stop.”)

  Rule-based phonology also over-predicts with respect to the ordering of rules. We will have more to say about the ways phonological generalisations can interact with one another over the next few weeks, but for now let me note that an overview of a classic example of the sort of controversies that can arise in rule ordering is given by Kaye (1990). By the standards of linguistics papers, it’s quite an entertaining read, too!

• In a rule-based derivation, there are a number of intermediate representations for each UR on its way to becoming an SR. Recall this example from Exercise 2:

  (4)

  UR	/met+u/	/pek+l/	/greb+l/	/krad+la/
  [cor, plos] → ∅ / ＿ [lat]	―	―	―	krala
  [lat] → ∅ / C＿#	―	pek	greb	―
  [obs, +voi] → [−voi] / ＿#	―	―	grep	―
  SR	[metu]	[pek]	[grep]	[krala]

  There is little evidence for the psychological reality of intermediate representations such as {greb} above.

• On the other hand, there is good evidence that phonological rules concern themselves with what the SR ends up looking like. For example, consider the following two rules from the Bantu language OshiKwanyama:

  (5) Post-nasal voicing rule: [obs, −voi] → [+voi] / [nas] ＿

  e.g. in loanwords from English and other languages: stamp → [sitamba], ink → [oinga].

  (6) Root-initial nasal substitution: [nas] [obs, −voi, αART] → [nas, αART]

  e.g. /eːN-pati/ → [eːmati] ‘ribs’; /oN-tana/ → [onana] ‘calf’.

  It is also generally true that sequences of nasal + voiceless obstruent do not occur within the URs of morphemes in OshiKwanyama.

  (Kager 1999: 83)

  Rule-based phonology requires separate rules for post-nasal voicing vs. nasal substitution, plus perhaps a Morpheme Structure Constraint for the prohibition on NC̥ generally. The rules, however, are all serving the same function: making sure that NC̥ sequences never appear in surface representations. This form of interaction between rules is known as conspiracy.

  The classic paper on conspiracies is Kisseberth (1970).

  Note also that this intuition, that phonological processes primarily constrain surface forms, has made it all the way into the conservative cloisters of comparative philology. Ringe (2006:120) argues explicitly that Sievers' Law is not a prompt sound change on the Neogrammarian model, that simply re-organises the sound of the language as it exists in one point in time, but rather a “surface filter”, that acts to make sure that particular structures are never actually pronounced.

So what’s the alternative?

We can implement phonological generalisations of the sort described by rules in terms of universal violable constraints.

(7) Example constraint: *NC̥

Assess a violation for every sequence of nasal + voiceless consonant that appears in the surface representation.

Defining constraints this way has certain consequences:

• Because constraints are axiomatically universal, to propose a new one means stating what its effects are on every language in the world.

• Constraints are violable: under certain circumstances, you can have an SR that the constraint penalises. Nevertheless, this is not the same as having no constraint at all, even in those languages that permit violations of the constraint more or less anywhere:

  • Positing a constraint means positing an implicational universal: we expect that structures not penalised will be tolerated in more contexts than structures that are. So, for example, we expect sequences like [nd] to be more frequent typologically than [nt] if *NC̥ is a real constraint.

  • We also predict that marked structures will be acquired later than unmarked ones by children. So, we predict that a child with brothers named Andrew and Anthony will be able to call brother Andrew by his right name first!

How would that work?

(8)

Input: /atʃowowo/			Non-Fi­na­li­ty	GrWd­=­PrWd	Ft­Bin	WSP	Un­even­Iamb	Parse­Syll	All­Ft­Left	Dep-µ-IO
a.	☞	(a.tʃóː).wo.wo						**		*
b.		a.(tʃo.wóː).wo						**	*!	*
c.		(a.tʃóː).(wóː).wo					*!	*	**	**
d.		(a.tʃó).wo.wo					*!	**
e.		(a.tʃóː).(wo.wóː)	*!							**

Optimality Theory (Prince and Smolensky 1993) replaces the ordered-rules component of generative phonology with a transitively ranked hierarchy of universal violable constraints. What that means and how it works in practice is the subject of next week’s lecture.

Homework

Our practical class this week will be more theoretical than usual. Read Archangeli (1997) and come prepared to discuss the following points:

1. If phonology is all about satisfying constraints on surface structure, why is it possible to produce any utterance other than whatever is the most harmonic one; something like [ba] or [ʔiʔi]?

2. How many faithfulness constraints are there? How many faithfulness constraint schemas do we need?

3. Where does the surface representation we're whipping into shape using constraints come from? Is its relation to the underlying form only enforced through faithfulness constraints, or does the UR guide the search for the optimal SR in some other way as well (or instead)?

If you want to write up answers to any or all of these questions and hand them in, you’re welcome to, but the main thing is to be ready to discuss them in class. I hope to do as little of the talking as possible!

References

Archangeli, Diana (1997) “Optimality Theory: an introduction to linguistics in the 1990s” in Archangeli, Diana and D. Terence Langendoen, eds. Optimality Theory: an overview pp. 1‒32. Oxford: Blackwell.

Browman, Catherine P. and Goldstein, Louis (1992) “Articulatory Phonology: an overview” Phonetica 49:155-180.

Chomsky, Noam and Halle, Morris (1968) The sound pattern of English. New York: Harper and Row.

Johnson, Keith (1997) “Speech perception without speaker normalization: an exemplar model” in Johnson, Keith and John W. Mullennix, eds. Talker variability in speech processing pp. 145‒166. San Diego: Academic Press.

Kager, René (1999) Optimality Theory. Cambridge University Press.

Kaye, Jonathan (1990) “Whatever happened to Dialect B?” in Mascaró, Joan and Marina Nespor, eds. Grammar in progress: GLOW essays for Henk van Riemsdijk pp. 259‒264. Dordrecht: Foris.

Kisseberth, Charles (1970) “On the functional unity of phonological rules.” Linguistic Inquiry 1:291‒306.

Pierrehumbert, Janet (2001) “Exemplar dynamics: word frequency, lenition and contrast” in Bybee, Joan and Paul J. Hopper eds. Frequency and the emergence of linguistic structure pp. 137‒157. Amsterdam: John Benjamins.

Pierrehumbert, Janet (2002) “Word-specific phonetics” in Gussenhoven, Carlos and Natasha Warner, eds. Laboratory Phonology VII pp. 101‒139. Berlin: Mouton de Gruyter.

Prince, Alan S. and Smolensky, Paul (1993) [2004] Optimality Theory: constraint interaction in generative grammar. Oxford: Blackwell.

Ringe, Donald A. (2006) From Proto-Indo-European to Proto-Germanic. Oxford University Press