AUTHORS: Mark Hale and Charles Reiss TITLE: The Phonological Enterprise PUBLISHER: Oxford University Press YEAR: 2008
Michael Cahill, SIL International
SUMMARY In some ways, this is a ''philosophy of phonology'' book, aimed at shaking up phonologists who have not examined the roots of their discipline. The authors, Mark Hale and Charles Reiss (H&R), have for years offered challenges to Optimality Theory (OT); this volume expands on their previous work, examining the presuppositions of phonology.
H&R aim this book at phonologists who share a broadly generative outlook, but, they assert, ''often offer explanations and propose models which are largely *inconsistent* [authors' emphasis] with these assumptions'' (p24). They regard phonology as the study of mental representations, and thus an abstract system, separate from phonetics.
Chapter 1 Introduction H&R state that it is uncontroversial that humans have a language faculty, which as a whole system is distinctly human. It follows that there must be some initial state of this faculty, before any experience with an actual language, and this initial state, S0, is what they label Universal Grammar (UG, defined somewhat differently than others). Similarly, they adopt the notion of a “Language Acquisition Device” (LAD) proposed by Chomsky and others as a biological mechanism inherent to humans. They introduce two simple data sets that they refer to later: ''Georgian lateral fronting'', and a Catalan alternation, both discussed below.
Though H&R believe in synchronic phonology, they also maintain that many synchronic patterns are solely the result of regular diachronic change. Child language acquisition is crucial in this, to the point that ''conspiracies'' in phonology can be explained by diachronic change and the fact that languages tend to have transparent processes. They propose that phonology is the study of possible languages, not just tendencies and statistical universals, which may be a result of historical factors or sampling.
Chapter 2 The Subset Principle in phonology This chapter, say H&R, is an extended application of the quote ''In any computational theory, 'learning' can consist only of creating novel combinations of primitives already innately available'' (Jackendoff 1990:40). They show by means of model languages how this principle (termed the ''Innateness of Primitive Principle, IofPP) is logically necessary, but that much work on acquisition of phonological inventories is inconsistent with this. Children learn a particular language not by having it described to them, nor on the basis of negative reinforcement, but by induction -- generalizing from spoken input. The Subset Principle (SP) of language acquisition, as commonly articulated, says that a child's initial hypothesis of the language must be a subset of possible later hypotheses, and so assumes an initial language state that is quite impoverished.
H&R illustrate this with an extended illustration of language using a deck of cards, in what they concede is ''potentially excruciating detail''. Various ''UG's'' can be constructed with different posited features and operators. Thus examining what ''real grammars of cards'' allow will permit deduction of what is a possible UG. The IofPP says that all grammars must have a set of primitives, or nothing could be distinguished from anything else, and consequently be learned. It is apparent that this set of primitives must ultimately be an initial state of the grammar. Applied to phonology, H&R conclude that all humans have access to the set of phonological features used in all languages of the world.
In contrast to this, the Subset Principle traditionally plays out quite differently. Imagine a child at a three vowel stage, knowing /i, u, a/, but learning a language which includes seven vowels, including /ɪ/. When the child hears [ɪ], she must either parse it, that is, assign it a linguistic representation, or fail. Since the only representations available are those which produce /i, u, a/, the [ɪ] must be parsed as one of those, likely /i/, or fail to be parsed. Critically, the traditional model provides no mechanism for expanding the vowel inventory. If the child distinguishes [ɪ, i], then that entails having access to features that distinguish them, which according to the model of impoverished beginnings, the child does not have.
Therefore, H&R take the Subset Principle to be a restriction that is relaxed on the basis of language input. A small vowel inventory as above might be interpreted as a small number of elements. However, this view would assume the segment as a primitive. If one posits the phonological feature as the primitive, then a small segmental inventory entails a large number of primitive features that are available to restrict the class. H&R thus claim that children ''are born with the representational apparatus to parse all vowel distinctions'' (p48). Studies show infants are sensitive to phonetic distinctions that are crucial in languages of the world, but by 10 months lose some of this. Also, we know that children's mental representations are more detailed than their speech output, since they react negatively to mispronunciations, even when they also mispronounce.
The Innateness of Primitives Principle, in which infants have access to all the representational apparatus of UG, is in their view a logical necessity.
Chapter 3 Competence and performance in phonological acquisition In this chapter, H&R argue that children's speech differs from adults not in terms of grammar, but because of a range of cognitive and physiological factors -- performance.
Studies of the intoxicated speech of Captain Hazelwood of the Exxon Valdez show an ''emergence of the unmarked'' by misarticulation of [r, l], devoicing of final consonants, and deaffrication. There are two options here. First, alcohol impaired the captain's performance system, or second, alcohol caused constraint re-ranking or changes in rules. The first is obviously preferable, and shows that performance is available as a possible explanation. There is high variability in children's speech, but there is also high variability in other muscular control systems at that age. This casts doubt on the assertion that children have a distinct phonology (mental representation).
H&R show that the oft-cited Smith (1973) study of Amahl can be interpreted in terms of performance, not competence. H&R assert that increasingly accuracy in pronouncing an /s/, for example, is due not to any increase in representational sophistication, but to increasing accurate motor skill in saying an /s/.
They discuss Smolensky (1996), especially ''chain-shift'' in children's speech, where /θ/ is said as [f] (''fick'' rather than ''thick''), yet /s/ is said as [θ] (''thick'' rather than ''sick''). Smolensky says this shows there is a grammar problem, since theta is physically pronounceable by the child. However, H&R interpret this as: each target segment has its own physiological trajectory, and so each mispronunciation is a unique miss of a target, and the performance account is feasible.
Children often pronounce both ''shoe'' /šu/ and ''Sue'' /su/ as [su]. Two hypotheses arise. First, the grammar changes the UR so the output of the phonology is /su/, and it is pronounced [su]. Second, the grammar retains /šu/ as the output of the phonology, and performance factors produce [su.] Children clearly distinguish adult pronunciations of [šu] and [su] (they reject a rendering of ''shoe'' as [su]), showing that their grammar distinguishes these, and that their parser does not make reference to their own phonetic output. In other words, the second hypothesis is supported.
Chapter 4 The Georgian problem revisited This brief chapter applies the conclusions of previous chapters to the Georgian data, where velarized ''L'' occurs before /a, o, u/ and the plain 'l' occurs before /i, e/. H&R propose that an initial state of a rule L --> l before /i, e/ would start as two separate rules, one for /e/ and one for /i/, and these rules would be fully specified for the relevant features of /i/ and /e/. The learner eventually collapses these into one rule, which omits the feature [high] that distinguishes /e, i/.
H&R's position is thus ''The correct statement of a rule arrived at by the LAD is the *most highly specified* representation that subsumes all positive instances of the rule, and subsumes no negative instances of the rule [their emphasis].'' This of course contradicts the common practice of proposing the most economical rule.
Chapter 5 Isolability and idealization In this chapter, H&R examine what phonology actually should include, arguing for phonology as an isolable discipline. They demonstrate by means of a diagram illustrating the process of someone uttering ''cat'' and a hearer understanding it. By doing this, they can focus on the factors and issues that are properly linguistic, and can be isolated and idealized for study.
A key concept here is the difference between computation and transduction. Computation is some process (deleting, rearranging, etc.) that uses the same ''representational alphabet'', such as features, throughout the process. Transduction is a process in which the input form is changed to a different system altogether, such as when differences in air pressure are converted to electric signals in a microphone. An ''articulatory transducer,'' for example, takes a phonological output representation and changes it to an articulatory score. So, the proper study of phonology, in their view, does not relate to articulatory gestures, or the waveform of the sound. These are all extralinguistic ''substances.'' It does include phonemic representations, the phonological computation system (whether rules or constraints), and the phonetic output representation.
Chapter 6 Against Articulatory Grounding Much work in the last decade and a half or so has gone into the connections between phonology and phonetics, often but not always in the context of OT. In this chapter, H&R argue that the ''phoneticist'' approach to OT lacks coherence, and that exploration in this area does not belong to phonology at all, but to historical linguistics. Phoneticists propose phonological featural representation mapping to gestural or acoustic representation. H&R note that this is a transduction, since it involves changing the elements of the representation. The incoherence of the phonetic grounding approach to OT, in H&R's view, comes from a conflict involving two incompatible but simultaneous ideas. In OT, candidates are evaluated based on a representation involving structure and features. If constraints are universal, they are a part of UG, and in some way, then, must be part of humans' genetic inheritance. (This connects to H&R's point above that phonological features are inherent to humans.) On the other hand, the phonetically-based OT approach appeals to phonetic grounding (e.g. ''Voiced Obstruent Prohibition'' because of articulatory difficulty factors) as a reason for proposing constraints. So, does the Voiced Obstruent Prohibition constraint exist because of genetics or because of practical physiological limitations?
A first response might be that a constraint is universal because of universal human anatomy and physiology, and invoking genetics is unnecessary. However, H&R would argue that phonological *features* are real (as in Chapter 2), and that they are the primitives of phonology which must be dealt with.
Chapter 7 Against Typological Grounding H&R here argue that arguments from markedness hold no water, and cite three cases of what they term ''substance abuse.'' The first case is the well-known positional faithfulness constraints of Beckman (1997), who concludes that psycholinguistically prominent syllables are more faithful to underlying forms than others. H&R argue that the observed patterns can be explained by acquisition factors -- learners will more easily learn contrasts which are in positions of greater intensity and duration. If acquisition is outside of the core of phonology (Chapter 3), then positional faithfulness should be also. The other cases cited are McCarthy's (1993) /r/-insertion, which must include an arbitrary component (not universal), and the parameterized constraints in McCarthy (1996), which have no phonetic grounding.
H&R write that phonetic ''substance'' (waveforms, duration, etc.) are not what is manipulated by the computational system of phonology, but that abstract features are. Markedness is therefore not a property of phonology, but of the intersection of acoustic salience and language change. Markedness is an emergent property, not one built into the phonological system. H&R do have a place for phonetic substance, but in a separate module from the mental representation they say is the proper domain of phonology. A transduction relates the two.
H&R also dismiss functionalistic approaches as arbitrary and providing no insight into the nature of grammar. For example, a ''dysfunctional'' approach could be proposed, e.g. with a constraint OBFUSCATE (''merge contrasts''). The data behind this is no less common than behind the proposed functionalist constraints MAINTAIN CONTRAST, and both have to do largely with the way humans either intentionally or unintentionally process speech.
Chapter 8 Against Constraints In this chapter H&R argue against the position that constraints are part of UG, and sketch a learning path that does not depend on these. If constraints express what is NOT, then there are necessarily an infinite number of these constraints.
H&R note that constraints even in a rule-based system are based on circular reasoning. Constraints express the avoidance of disfavored structure, e.g. ''A rule is blocked by a constraint because to apply the rule would lead to an ill-formed structure.'' But we know that the structure is ill-formed is because the rule is blocked. Any constraint could be omitted if either a more fully specified initial rule is proposed, or another rule is proposed which repairs the structure in question.
One explicit goal is to remove the notion of ''ill-formedness'' from the formal terminology of the phonology. The very notion of constraints presupposes the idea of ill-formedness. If ill-formedness is not a coherent notion, then constraints are unnecessary. They illustrate this by examining the Obligatory Contour Principle (OCP), reiterating some points made in Odden (1988), who shows that various languages have rules with opposite results, such as:
Insert a vowel between consonants unless those consonants are identical. Insert a vowel between consonants ONLY IF those consonants are identical.
As Odden and H&R note, the OCP is invoked ''rather opportunistically.'' It is just as possible for a rule to create OCP violations as it is to repair them. The widespread existence of identity-effect phenomena that the OCP is invoked to explain is, in H&R's view, motivation to more deeply understand the nature of what are possible rules of phonology.
Chapter 9 Against Output-Output Correspondence In this chapter, H&R specifically target the Output-Output Correspondence (OOC) constraints of OT. In the literature, they find cases of opportunism, misanalysis, and problematic predictions. They include quite a detailed discussion of McCarthy's (2000) Rotuman analysis, significant because it provided a spur for later work on OOC. Also, they show that much of the data for the alleged OOC is the result of the historical process of analogy. Finally, they discuss Kenstowicz's (1996) analysis of Korean and Spanish, declaring the OOC to be unmotivated in these.
Chapter 10 A principled solution to Catalan We must wait 14 pages into this chapter for Catalan, while H&R present their own view of phonology. One key postulate is the Unified Interpretive Procedure. This says that a rule must be either feature-changing or feature-filling, but not apply vacuously. Underspecification is thus a formal category.
The Catalan data in question is: masc. fem. Gloss sɛk sɛkə 'dry' sek seɣə 'blind'
H&R list seven possible solutions, rejecting four of them quickly, and leaving possibilities for 'blind' of underlying /g/, /ɣ/, or the underspecified /G/. They themselves have taught that the best solution is /g/, for several reasons, including that /g/ is less marked than /ɣ/. However, in this volume they have argued that markedness is formally incoherent, as are segment inventory symmetry considerations. Similarly, they reject the other reasons as inadequate to force a conclusion of /g/. The /ɣ/ has the advantage of being a surface manifestation, but this also is not compelling. H&R combine their previous discussion of language acquisition to say the acquired rule should be the intersections of the individual instances. The most highly specified representation that subsumes both /k/ and /ɣ/ is in fact /G/. The Unified Interpretive Procedure discussed above prevents a rule applying to both /G/ and /k/; these are distinct targets of a rule.
EVALUATION Readers of this book will likely find it maddening or refreshing, or perhaps both. A strength of the book is their almost philosophical approach to phonology. They return to first principles to attempt a logically coherent approach. This goes beyond the pragmatic approach that characterizes an average evaluation of a theory.
However, the notion of ''economy'' is applied unevenly. H&R appeal to criteria like ''economy'' (Ch 8, p200) in support of their hypothesis about UG. But earlier (Ch 4) they proposed a learning algorithm with a less economical representation than a traditional one. Economy is a factor, but certainly not the most crucial one, as Occam's Razor says (and which they often invoke).
For example, as previously stated, ''The correct statement of a rule arrived at by the LAD is the *most highly specified* representation that subsumes all positive instances of the rule, and subsumes no negative instances of the rule.'' This will not be the most economical statement. One wonders what an entire formal phonology of a language would look like in this system.
I find chapters 6-7, on articulatory and typological grounding, the most intuitively unappealing, especially H&R's rejection of an explicit influence of phonetics into formal phonological systems -- surely phonology relates somehow to actual sounds. H&R's position is somewhat reminiscent of Fudge (1967), who concluded that phonologists ''ought to burn their phonetic boats and turn to a genuinely abstract framework''. However, step by step, they build a case for their position, and it is incumbent on ''phoneticists'' to argue their case as explicitly as H&R do.
There are a few issues H&R have neglected, such as the connection between syntax and phonology that arises in some languages. Indeed, they explicitly say that ''phonological computation has access ONLY to phonological structures and processes…'' (p224). It is possible that they may include these types of phenomena in the seemingly arbitrary phonological processes that have their root in historical developments, but they do not say so.
Finally, Chapter 9, against output-output correspondences, seems a bit out of place, as it targets one very specific issue within OT, and is much less general than the other chapters.
The refreshing part of the book comes in H&R's reasoning about every single topic. They are consistently asking the big ''why'' questions, and demanding precision of definition from those who have not always supplied it. To this reader's mind at least, the argumentation is mostly well-grounded in the reality of language and current data on language learning and processing. They raise issues that must be addressed by phonologists if we are to have any claim to consistency and coherence.
Fudge, E.C. 1967. The Nature of Phonological Primes. Journal of Linguistics 3:1-36.
Jackendoff, Ray. 1990. Semantic Structures. MIT Press: Cambridge, MA.
Kenstowicz, Michael. 1996. Base identity and uniform exponence: alternatives to cyclicity. In Current Trends in Phonology: Models and Methods (J. Durand and B. Laks, eds.), pp. 363-393. CNRS, Paris X, and University of Salford.
McCarthy, John. 1993. A case of surface rule inversion. Canadian Journal of Linguistics 38:169-195.
McCarthy, John. 1996. Remarks on phonological opacity in optimality theory. In Studies in Afroasiatic Grammar: Papers from the Second Conference on Afroasiatic Linguistics, Sophia Antipolis, 1994 (ed. J. Lecarme, J. Lowenstamm, and U. Shlonsky). Pp. 215-243. Holland Academic Graphics, The Hague.
McCarthy. John. 2000. The prosody of phase in Rotuman. Natural Language and Linguistic Theory 18:147-197.
Odden, David 1988. Antiantigemination and the OCP. Linguistic Inquiry 19:451-475.
Smith, N.V. 1973. The Acquisition of Phonology: A Case Study. Cambridge University Press.
Smolensky, Paul. 1996. On the comprehension/production dilemma in child language. Linguistic Inquiry 27:720-731.
ABOUT THE REVIEWER
ABOUT THE REVIEWER:
Michael Cahill (Ph.D. Ohio State University) was involved with the Koma
Language Project in Ghana for several years before serving as SIL's
International Linguistics Coordinator from 1999-2010. He is now
Editor-in-Chief of the Content Services part of SIL Publishing. His main
linguistic interests are phonological, especially tone systems and the
phonology, phonetics, and historical development of labial-velar
obstruents, as well as African languages in general.