Date: Tue, 12 Jul 2005 13:22:48 -0700 (PDT) From: Lev A Blumenfeld Subject: Phonological Augmentation in Prominent Positions
AUTHOR: Smith, Jennifer TITLE: Phonological Augmentation in Prominent Positions SERIES: Outstanding Dissertations in Linguistics PUBLISHER: Routledge YEAR: 2004
Lev A. Blumenfeld, Stanford University
Phonological Augmentation in Prominent Positions, a slightly revised version of Jennifer Smith's 2002 UMass dissertation, addresses the role of substantive pressures on formal phonological systems, focusing on the behavior of prominent positions. The book's central claim is that phonological grammars contain both an abstract, formal mechanism that generates OT constraints from a limited set of primitive objects, and a set of conditions, or filters, that arise from phonetic or psycholinguistic factors. The job of these substantive filters is to determine which constraints among all the formally possible ones may be members of the universal constraint set, CON. Under this Schema/Filter model of CON, substantive pressures affect grammars only indirectly, by determining which constraints can in principle be active in a grammar, and not by influencing actual phonologies of actual languages. Smith's illustration of this proposal comes from a detailed investigation of phonological augmentation in prominent positions.
In Chapter 1 Smith outlines the central issue addressed in the book, "the question of how the phonology, a formal system that manipulates abstract, formal objects, can nevertheless be shaped by substantive considerations" (p.4). The chapter serves as an introduction of the Schema/Filter model of OT phonology and the empirical domain of the book. Smith argues for dividing prominent or strong positions into two types: phonetically and psycholinguistically strong positions (henceforth ph-strong and ps- strong). Augmentation of ph-strong positions serves to increase their inherent perceptual salience, while augmentation of ps-strong positions caters to processing pressures such as early-stage word recognition and demarcation of word boundaries. Ph-strong positions discussed are the stressed syllable, onset (or released) consonants, and long vowels. Ps- strong positions are the initial syllable and the morphological root. Markedness constraints relativized to both types of positions are subject to a general substantive filter, the Prominence Condition; another filter, the Segmental Contrast Condition, applies to constraints on ps-strong positions.
The Prominence Condition requires augmentation constraints referring to strong positions to call for perceptually prominent properties to be present in those positions. The Segmental Contrast Condition further restricts augmentation constraints referring to ps-strong positions by prohibiting them from altering features that are important in early-stage word recognition, unless the constraint's effect is to demarcate the left edge of a word.
Smith presents phonological evidence for the notion 'strong position' and compares two kinds of constraint schemas that can be used to analyze augmentation effects: faithfulness constraints referring to weak positions (F/wk) vs. markedness constraints referring to strong positions (M/str), arguing for the latter on the grounds that strong, not weak positions, tend to be more easily definable.
Chapter 2 presents the theoretical framework of the analysis of augmentation processes. Building on the previous work of Eisner and Hayes, Smith lays out the Schema/Filter model of CON. The theory contains a set of primitives (features, correspondence relations, strong positions), a set of constraint schemas (e.g. Ident, Align, the augmentation schema C/str), and substantive filters on potential constraints.
Discussing the nature of ph-strong positions, Smith argues that the notion is an abstract one, only indirectly related to phonetic detail such as the presence of cues to contrasts. The argument for this claim, often repeated elsewhere in the book, is that M/str augmentation constraints may have "no relationship to the featural contrast for which the [strong] position has special salient cues. Thus, the status of the stressed syllable as a strong position is more abstract and general than the phonetic origin of that privileged status" (p.32).
The notion of a ps-strong position is also abstract in some sense: while psycholinguistic importance of phonological material tapers off gradually from the initial syllable onward, phonologically it is only the initial syllable that is privileged with respect to the others, without any evidence of gradience.
The remainder of the chapter contains a detailed illustration of how the Schema/Filter model applies to phonological augmentation. The Prominence Condition requires that markedness constraints relativized to prominent positions be augmentation constraints, i.e. call for the presence of perceptually prominent properties. In other words, markedness constraints can be relativized to prominent positions if they would make that position even more prominent "by association with some perceptually salient property" (p.44). So a constraint forcing stressed syllables to be heavy passes this test, while a constraint banning mid vowels in the stressed syllable does not. Smith's way of testing whether a given constraint counts as an augmentation constraint relies on the substantive notion of "perceptual prominence": if a candidate that satisfies the constraint is more prominent than a similar one that does not, then the constraint is an augmentation constraint. Perceptual prominence itself is grounded in extragrammatical factors: "it may be appropriate to categorize one stimulus as more perceptually prominent than another if the first stimulus elicits a neural response of greatera magnitude than that elicited by the second" (p.45).
Smith then discusses several constraints that qualify as augmentation constraint by her criteria: HeavySyll, *Peak/X, Onset, *Onset/X, HavePlace, H(igh)Tone, and HaveStress. Cross-classifying with the five strong positions, Smith arrives at a list of constraints predicted to be potentially active in a language.
Smith then moves on to ps-strong positions and the constraint filter specific to them, the Segmental Contrast Condition. This filter prevents augmentation constraints referring to strong positions from altering features that are distinguished in early-stage word recognition. In practice, this means that constraints neutralizing segmental contrasts are not allowed to be relativized to psycholinguistically strong positions, since it is segmental and not prosodic features that play a role in early- stage word recognition. Interestingly, syllable quantity is 'segmental enough' to be controlled by this condition, but stress is not. The Segmental Contrast Condition has an important codicil: constraints are exempt from it if they serve to demarcate the left edge of a constituent. This clause allows Smith to account for constraints that call for word- initial onsets.
The chapter concludes with a brief discussion of Steriade's Licensing By Cue theory (LBC) as a filter on positional faithfulness constraints. LBC theory claims that the presence of cues to a feature in a given position is responsible for the existence of a faithfulness constraint protecting that feature in that position. Smith suggests that one can think of the information about cues to contrast serves as a substantive filter on positional faithfulness constraints. Smith argues that while it is the case that positional faithfulness constraints relativized to ph-strong positions are subject to LBC -- in other words, those constraints only protect those features for which the position has special cues -- this is not the case for ps-strong positions. For example, stressed syllable faithfulness constraints are limited to vowel features, while initial syllable faithfulness includes consonantal features as well. Smith suggests that this difference can be accounted for by a substantive filter on a free constraint construction mechanism.
Having laid the groundwork of the theory in the two opening chapters, Smith moves onto a detailed investigation of augmentation phenomena. Chapter 3 covers ps-strong positions and Chapter 4 ps-strong ones. Both chapters compare the predicted set of augmentation constraints with typological observations.
Chapter 3 opens with a discussion of augmentation constraint referring to stressed syllables. Smith considers five augmentation constraints: HeavySyll, HTone, *Peak/X (calling for high-sonority nuclei), Onset, and *Onset/X (calling for low-sonority onsets). Relativized to stressed syllables, each of these constraints, Smith claims, produces two kinds of phonological patterns: augmentation of stressed syllables, and attraction of stress to the augmented property in question. All ten patterns are illustrated with actual examples.
Smith also looks at positional augmentation of long vowels and syllable onsets.
Chapter 4 moves on to ps-strong positions, the initial syllable and the root, exemplifying the interactions with the constraints Onset and *Onset/X. Relativized to the initial syllable, these constraints call for the presence of word-initial onsets and low-sonority onsets, respectively. Both patterns are attested. An interesting case is presented by Mongolian and some other languages, where word-initial glides but not liquids are allowed, in apparent contradiction to the predicted patterns. This problem prompts a discussion of the differences between true onset glides and nuclear onglides, which, Smith proposes, accounts for the aberrant behavior of Mongolian.
Augmentation of the strong position 'Root' is exemplified with the constraint HaveStressRoot. Smith also suggests that root minimality effects may also be due to augmentation constraints, but leaves the specifics for future research.
Chapter 4 moves on to a lengthy discussion of psycholinguistic evidence behind the Segmental Contrast Condition, the filter specific to ps-strong positions. Smith offers evidence showing a distinction between early- and later-stage word recognition, and evidences supporting the relatively greater importance for early-stage word recognition of segmental material vs. prosodic information, roots vs. affixes, and initial vs. medial segments. Smith concludes the chapter with an argument that what she has been calling the 'initial syllable' is the first syllable of the morphological rather than the prosodic word.
The final chapter of the book is dedicated to the relationship between positional augmentation and positional neutralization. After presenting several OT analyses of neutralization phenomena, Smith argues that they are independent of the augmentation processes that had been the focus of the preceding chapters. Two separate constraint families are required for these two types of processes.
_Phonological augmentation_ lives up to its goal as an important contribution to one of the most important recent debates in phonology on the role of functional pressures in phonological systems. In part functionalist and in part formalist, the book offers a well-presented, thoroughly argued, and thought-provoking proposal that will prove to be an important step toward an answer to this central question in phonological theory.
Smith's exposition is detailed and precise. Perhaps, to make the argument even more clear, the material on psycholinguistic evidence consigned to the end of Chapter 4 could have been placed earlier in the volume. Otherwise, the book is readable and easy to follow. Important points are reinforced frequently throughout, the predictions of the theory are clearly laid out, and the author's goals are always in the reader's sight.
I will now raise one substantive issue having to do with the role of functional pressures in phonology, and two objections to Smith's proposal.
The central and most interesting aspect of Smith's proposal is the indirect way in which substantive pressures shape phonological grammars. Smith offers at least two explicit arguments for her position. Augmentation of both ph- and ps-strong positions shows properties not easily explicable in a more directly functionalist theory. First, the stressed syllable, a ph-strong position, undergoes augmentation of features not directly related to its phonetic prominence, showing that the notion of 'prominent' is more abstract than the phonetic facts themselves and that functional pressures must apply at a more abstract level. Second, the initial syllable, a ps-strong position, is phonologically categorically stronger than the rest of the word, even though psycholinguistic importance tapers off gradually as one goes farther away from the beginning of the word.
In addition to these two arguments, the typological survey in Chs. 3 and 4 gives an empirical argument, even if left implicit, that the substantive pressures act at the level of constraint construction in CON rather than at the level of individual grammars. Consider the Segmental Contrast Condition: it rules out augmentation constraints relativized to ps-strong positions that alter segmental features. This explains the absence of such constraints as HeavySyllable and *Peak/X (calling for high-sonority nuclei) relativized to the initial syllable. This predicts that there should not be vowel lengthening or lowering in word-initial syllables independently of stress. The functional explanation behind this fact is that segmental contrasts, unlike prosodic ones, are important in early- stage word recognition, and therefore neutralizing them would be dysfunctional.
However, if this functional pressure were to exert is influence at the level of individual phonologies rather than as a constraint filter, then augmentation constraints relativized to ps-strong positions would be unable to affect only those segmental features that are CONTRASTIVE in the language. Allophonic features do not distinguish words from each other and therefore are at least not as important as contrastive features in word recognition. In other words, a more directly functional theory of augmentation in ps-strong positions would predict that languages should allow COMPLEMENTARY distribution of segmental features between ps-strong and ps-weak positions. Such a language would have, for example, long vowels in the initial syllable and short vowels elsewhere, due to the constraint HeavySyllable/S1. Likewise, the constraints in the family *Peak/X/S1, calling for high-sonority nuclei, could have their effect in a language without contrastive mid vowels. Such a language would have mid and high vowels in complementary distribution: the former in initial syllables, the latter elsewhere.
Smith lists no such languages, and I also am not aware of examples of such a distribution. If this pattern does not exist, then its absence lends empirical support to Smith's hypothesis that functional pressures are moderated by an abstract component of phonology such as the constraint construction mechanism, rather than directly affecting actual phonologies.
Next I will point out what seem to be to be two shortcomings of Smith's book. First, some of the proposals are vague enough that the reader is unsure how to apply the reasoning to new cases. The notion that does much explanatory work in Smith's theory is "perceptual salience". Knowing which of two forms is more salient than the other allows one to decide what counts as an augmentation constraint. This in turn is crucially important for the Prominence Condition, the constraint filter that allows only augmentation constraints to refer to prominent positions.
Smith grounds the notion of salience in neurophysiology: one stimulus is more salient than another if it "elicits a neural response of greater magnitude" (p.45). Smith admits that this proposal is a preliminary step toward a more explicit and fleshed out criterion. As it stands, however, it is not always clear how to apply the Prominence Condition to new cases.
Smith does not offer neurological or psycholinguistic evidence for the categories she considers prominent, viz. the heavy syllable, high tone, or relative position on the sonority scale (do liquid syllable nuclei elicit a greater neural response than nasal ones?). She does provide a psycholinguistic argument for Onset and *Onset/X constraints: interspersing low-sonority consonants and high-sonority vowels allows the perceptual system to more efficiently recover from adaptation, making both the consonants and the vowels more perceptible. This argument, however, could be used just as easily to support the constraint Coda, because the perceptual system cares simply about alternating dissimilar segments, not about syllable structure.
The fit between Smith's model and the observed data is very good: those and only those features claimed to be salient turn out to participate in augmentation processes. Given the looseness of the criterion for selecting salient features, however, it would have been helpful to more systematically consider features that are NOT salient and thus do not gravitate toward prominent positions. In other words, the argument could have been made stronger if Smith considered not only the positive side (salient features that participate in augmentation), but also the negative side (non-salient features that do not). Smith does mention some examples, e.g. the absence of a constraint banning mid vowels in stressed syllables. There are, though, less clear-cut cases which could have been discussed in the book to help make the proposal more clear. What about aspiration and glottalization? Is an aspirated consonant more perceptually salient than a plain one? If not, what is behind the recurring pattern of attraction of laryngeal features like aspiration to the onsets of stressed syllables?
Once again, these difficulties with Smith's proposal arise due to its admittedly preliminary status.
I will raise one other substantive point. Smith shows that the general classes of constraints predicted by her system are all attested in the world's languages. It seems, however, that the typology is not as evenly populated when we look at the constraints within each class. The *Peak/X class, for example, calls for high-sonority syllable nuclei. Relativized to the stressed syllable, it generates the familiar pattern of sonority- driven stress. The same constraint relativized to the strong position Long Vowel predicts a pattern of lowering of long vowels. Smith gives one example (Yokuts) where long high vowels fall; she could have added here languages like Nalik that have only one long vowel, and it is low. However, far more frequent is the opposite pattern, often seen in chain shifts: long vowels rise (cf. the English Great Vowel shift). Perhaps, there is another explanation for long vowels rising, but if there are separate functional accounts for two opposite patterns, what explanatory power does either of them have?
Next, consider the Onset family of constraints. There are many examples of languages where stress is repelled from onsetless syllables; these stress systems are analyzed with the simple constraint Onset relativized to the stressed syllable. The constraint family *Onset/X relativized to the stressed syllable, however, is not very well supported typologically. There is one example (Niuafo'ou) where stress is repelled from syllables that begin with glides. In Pirahã, the voicing of the onset plays a role. There are no examples where the cutoff on the sonority scale between acceptable and unacceptable stressed syllable onsets occurs elsewhere, e.g. between liquids and nasals, or between nasals and obstruents. This typological sparseness of stress effects due to *Onset/X is not necessarily an argument against Smith's proposal, but it appears to be a significant fact in search of an explanation.
I hasten to add that despite the two objections that I raised, Smith's book offers a stimulating argument for a particular view of phonological theory, and will surely provoke much debate in the near future. It is obligatory reading for phonologists interested in the interplay of formal and functional explanations. The few and minor shortcomings of the book, no doubt, are a sign of exciting future research.
A final note on the presentation of the book: The editors did not correct egregious pagination problems around large tables, which often mislead the reader that a chapter has ended (e.g. most of p. 77 is blank). Also, shading in tableaux looks like second-generation photocopy. This is not serious and would not have been worth mentioning except that the volume costs $90.
ABOUT THE REVIEWER:
ABOUT THE REVIEWER
Lev Blumenfeld is a fourth-year graduate student in linguistics at Stanford, specializing in word-level prosody and in metrics. He is currently working on the interaction of prosodic structure and segmental processes, and on the role of stress in Latin quantitative meters.