By Sari Pietikäinen, FinlandAlexandra Jaffe, Long BeachHelen Kelly-Holmes, and Nikolas Coupland
Sociolinguistics from the Periphery "presents a fascinating book about change: shifting political, economic and cultural conditions; ephemeral, sometimes even seasonal, multilingualism; and altered imaginaries for minority and indigenous languages and their users."
Date: Thu, 16 Dec 2004 20:18:32 +0100 (MEZ) From: Markus Poechtrager Subject: Living on the Edge: 28 Papers in Honour of Jonathan Kaye
EDITOR: Stefan Ploch TITLE: Living on the Edge SUBTITLE: 28 Papers in Honour of Jonathan Kaye SERIES: Studies in Generative Grammar 62 PUBLISHER: Mouton de Gruyter YEAR: 2003
Markus A. Pöchtrager, University of Vienna.
This book is a collection of 28 papers to honour Jonathan Kaye's contribution to phonology. Jonathan Kaye is one of the founding fathers of Government Phonology, and accordingly, the majority of the papers deal with issues pertinent to that particular framework. That is why the book will be an indispensable collection of material for anyone with a serious interest in state of the art Government Phonology. The ideas discussed tackle basically every aspect of the theory: government and licensing relationships, empty nuclei, constituent structure, element theory, interfaces with other components etc. The individual articles cover a wide range of languages and offer interesting solutions to many issues that have been problematic so far.
The book is introduced by a very personal section, where three of the contributors recount their own experiences with Jonathan Kaye, both as a researcher and as a person.
The remainder of the book is divided into three parts: 1. General issues (Acquisition, Computation, The organisation of grammar, Philosophy of science and metatheory), 2. Elements: segmental structure and processes, 3. Structure (Branching onsets, "Codas", Empty categories, "Syllabic consonants", Templates and morphology, Metrical structure).
COMMENTS ON THE PAPERS
"Meno's paradox and the acquisition of grammar" B. Elan Dresher (pp 7--27) This article provides a concise introduction to the problems of language acquisition. Plato's problem (i.e. the paradox that linguistic competence of humans by far exceeds what they possibly could have extracted from the data they have been exposed to) serves as the starting point, followed by a discussion of the principles and parameters model and the difficulties associated with parameter setting: Far from being a trivial issue, the relation between a certain parameter setting and the effects it has might be rather indirect, due to the fact that parameters interact with each other. Thus, a learner with an incorrectly set parameter might know that there is something wrong, but s/he might not be able to identify the source of the problem, i.e. which parameter out of quite a number is to be changed (credit problem). Furthermore, parameters are usually stated in abstract terms and involve abstract concepts such as heads, anaphors etc. A learner might initially not be able to identify these categories in the input, which makes the correct setting a non-trivial task (epistemological problem).
Dresher then goes on to discuss three competing learning models for parameter setting. Model #1 is cue-based learning (Dresher & Kaye 1990). Children not only have an innate knowledge of UG principles and paramters, but also a kind of road map to guide them. The proposal is that each parameter has a reliable cue associated with it, which helps the learner in making the right choice. In addition to that, learning proceeds along a certain pre-specified path, i.e. parameter #2 can only be set after parameter #1 etc. Dresher exemplifies this by sketching out a possible learning path for the acquisition of stress, along with the relevant parameters and their cues. Opposed to this is model #2, the triggering learning algorithm (Gibson & Wexler 1994). This is basically a trial-and- error model. The learner tries to parse incoming data; if s/he succeeds, the parameter settings stay unchanged. If the data cannot be parsed, one parameter setting will be arbitrarily changed and the sentence reprocessed. Upon success, the change will be adopted; upon failure, the original setting is retained. Dresher points out that such a model suffers from serious deficiencies: learners might get stuck without a chance of escape, they might thrash around indefinitely, the sequence of learning is entirely accidental, at an early stage of acquisition it might be impossible to match the input perfectly and the learning process could not get off ground. Model #3 does not fare any better: the genetic algorithm model (Clark & Roberts 1993) suggests that learners simultaneously hold a number of competing hypothesis, each one of which is exposed to data and assessed for how well they did. The fittest hypothesis is retained, the others dropped. This, however, requires both an accurate fitness measure and that the relation between parameter and effect be relatively direct --- and none of the two requirements can be fulfilled.
To sum up, model #1 seems by far the most promising candidate to explain language acquisition.
"On the logical order of development in acquiring prosodic structure" Nancy A. Ritter (pp 29--54) Research into language acquisition is usually driven bei either one of the following hypotheses: (1) children have an innate disposition to acquire language, they have a UG which is a tailor-made acquisition device for language. (2) Children acquire language with the help of more general cognitive abilities. Ritter argues that both of these hypotheses have to be adopted. Certain general types of principles such as salience/dominance or locality have a role to play in UG, but they also occur in different instantiations in other components. It is claimed that children at an early stage do not base their production on correct adult forms which they have somehow acquired, but rather that children actively construct phonological representations, which come closer to the adult form as the childrens' phonological competence evolves. The prosodic development, which can already be seen to arise at the prelinguistic stage, is the rudimentary construction of phonological competence.
The article is couched within Head-Driven Phonology (van der Hulst & Ritter 1998), whose central tenets are the head/dependency principle (saliency is translated into the notion "head") and the binarity principle (heads and the material around them are grouped in a binary fashion); along with three kinds of licensing relations: structural licensing (responsible for hierarchical structure), paradigmatic licensing (determining the contrasts allowed for in a certain position) and syntagmatic-content licensing (the material in a certain position influences the material in an adjacent position).
After having sketched the central assumptions of the theory, Ritter walks us through the various stages of prosodic development and demonstrates how they follow from the evolution of the computational system. The first step is the recognition of a head in the shape of a highly salient part of the utterance, typically the (stressed) vowel, which falls under the paradigmatic licensing relation. The next step, viz. the emergence of CV syllables, is a manifestation of a syntagmatic-content licensing relation: the vocalic head requires that the object preceding it contain stricture. Structural licensing, the third kind of licensing countenanced by the theory, can be seen at work at the next stage, i.e. the reduplication phase: This reflects binary grouping of vocalic units (and their consonantal dependents) into binary feet. From there, various others phases can be accounted for by repetition of the relevant types of licensing at various higher levels.
At a certain age (around 3;0) children become aware of onsets as units autonomous from nuclear heads. In other words, after the possibilities of licensing become exhausted at the vocalic plane, the child moves on to the consonantal plane: by paradigmatic licensing onsets can be recognised as independent units and further application of the other types of licensing yields branching onsets as well as coda-onset structures.
To sum up, the prosodic development can satisfyingly be shown to follow from the three different kinds of licensing.
"On the computability of certain derivations in Government Phonology" Geoff Williams (pp 55--74) Williams' article is about the search for a model of phonology which both makes strong empirical claims and is adequately formally constrained. Government Phonology is characterised by its avoidance of arbitrariness in analysis and by its restrictiveness. However, much of the theory is expressed in relatively informal terms, which has also provoked the criticism from proponents of the declarative approach.
By stating grammatical regularities in terms of well-formedness conditions on surface forms, the declarative approach is guaranteed to do without derivations and to be computationally tractable. Yet, the empirical content of this theory, particularly as compared to Government Phonology, is rather poor. Williams sets out to show that for Government Phonology, too, a formal basis for tractability can be established. In order to do this, the basics of complexity theory and also its application to linguistic theories are sketched out. Ristad (1990) and Berwick (1991) performed a complexity analysis of the autosegmental framework whose results were rather daunting: The autosegmental approach is computationally intractable. However, this is due to theoretical mechanisms such as lexical underspecification and the lack of constituent structure being encoded in lexical entries. Crucially, Government Phonology is different in that respect (there is no underspecification and constituent structure is encoded with lexical entries) and thus tractable. Williams demonstrates this with the example of Turkish vowel harmony and comes to the conclusion that "certain substantive constraints and principles of Government Phonology do indeed have formal constraining power".
"Structure paradoxes in phonology" Harry van der Hulst (pp 75--93) This article argues for the coexistence of two kinds of phonological structure, a lexical one and a post-lexical one. While it is usually assumed that one is built on top of the other, van der Hulst proposes that both of them should be allowed for simultaneously in order to cope with phonological structure paradoxes. He calls this the Duality Hypothesis. Evidence for such a distinction comes from various phonological phenomena: onset/rhyme theories of the syllable argue that the syllable node (if assumed at all) branches into an onset and a rhyme (nucleus plus coda) as immediate constituents. Moraic models, however, put onset and nucleus under one mora and the coda under another one. These two tree structures are incompatible, which ceases to be a problem as soon as parallel structures are allowed for. A similar argument can be made for stress facts. Evidence from Dutch word stress shows that ambisyllabic consonants make the preceding syllable heavy, i.e. lexically they should be represented as geminates or codas only. Postlexically, however, all we see is single consonants which belong to both syllables. This discrepancy can be accommodated if the structures are allowed to coexist. Another point can be made for French, which lacks lexical accent but has phrasal, i.e. postlexical accentuation. On yet a higher level, we can distinguish between a lexical phonotactic word (phonotactically unanalysable) and a postlexical prosodic word (including for example level II affixes).
Without allowing for two parallel and coexisting structures numerous adjustments and destructive operations are required for phonological structure. It is an advantage of van der Hulst's approach that this can be avoided. However, it remains to be seen if the loss of restrictiveness such a proposal causes is counterbalanced by an increase of insight into phonological structure, as the author himself notes.
"An x-bar theory of Government Phonology" John R. Rennison and Friedrich Neubarth (pp 95--130) This article is an outline of a heavily modified version of Government Phonology. Like with many other "schools" of that theory, the CV pair is stipulated to be the sole building block of syllable structure. In stark contrast to strict CV (Lowenstamm 1996), however, the labels C and V are only derived: a skeletal point (the head or "V") combines with another skeletal to its left ("C") and by projection of the head they both form a branching structure termed "syll". It is at this projectional level that sylls are linked to form a phonological string. Along with syllable structure, also the theory of elements is redefined; the authors assume I, U, R, H, L and a functional element F which assumes various roles (of "old" elements) according to its position in a phonological expression and the association to the head or non-head portion of a syll. The authors further discuss so-called "lazy" elements, i.e. they propose that a phonological expression can contain elements that lag behind in their phonetic realisation, thus creating the effects of both traditional countour segments as well as branching onsets. A corollary of this rather major change in the theory of melody is the split between the internal make-up of phonological expressions and their "strength" in governing relationships. While in Standard Government Phonology the governing ability can be directly calculated from the number of elements, Rennison & Neubarth attribute a certain value to each element and thus a look-up table has to be consulted in order to verify whether governing can obtain between two sylls. The paper concludes with some exemplary representations of German and English word forms and also hints at a possible extension of the theory in order to include higher prosodic structures.
"Meta-phonological speculations" Sean Jensen (pp 131--148) This article questions two commonly held assumptions of phonological theory, viz. the Grammaticality Hypothesis and the principle of Attestation (any word can become attested for a speaker, i.e. "mean something"). Jensen's claim is that any theory which countenances both of these assumptions is untestable and therefore problematic. He demonstrates that the theory of attestation and the theory of grammaticality are in fact the same, and thus we are faced with the dilemma that statements of grammaticality are both false and untestable. The only way out, so Jensen, would seem to be denying the possibility of letting new phonological forms into our grammar. But this amounts to saying that there could be no such thing as loan words; certain forms in certain languages could never become part of the system.
We are thus forced to find a more interesting theory of grammar that can mimick the effects of grammaticality judgements. Jensen introduces affiliation as the central concept, i.e. intuitions whether certain forms belong to a language, as information about forms instead of a system defining licit and illicit structures. He goes on to sketch out a possible model of evaluating distances between phonological structures, by way of which average phonological forms can be calculated, i.e. barycentres. The "mass" an individual phonological form has is defined via the morphosyntactic meaning assigned to it, and from this one can calculate the exact location of the barycentre. The distance between a given phonological form and the barycentre is then a means to establish how "typical" (or: how close to grammaticality) the phonological form is in a particular linguistic system.
"Metatheoretical problems in phonology with Occam's Razor and non-ad-hoc- ness" Stefan Ploch (pp 149--201) Ploch's contribution to the volume is an elaborate argument against certain popular but flawed and untestable versions of simplicity and non- arbitrariness. He suggests to let go of both of them and to exclusively use Popper's criterion of testability instead, from which more scientific versions of simplicity and non-arbitrariness will follow automatically.
In order to demonstrate his point, Ploch compares a number of phonological theories and evaluates how they fare with respect to scientificness. The comparison between Standard Government Phonology (which allows for branching constituents) and Strict CV Phonology (which does not) starts off with the representation of a word such as English "brand". While the Standard Theory has one empty nucleus (i.e. domain-finally), Strict CV has to posit three of them (finally, between "b" and "r" and between "n" and "d"). In this way, Strict CV can get rid of branching, but only at the cost of stipulating additional empty categories which cannot be tested. The only way to know that they are where they are is because the theory stipulates them. In other words, such an empty nucleus cannot be observed and is thus unfalsifiable. Ploch goes to great length to demonstrate that any additional argument to defend the proliferation of empty nuclei runs into similar problems, i.e. it cannot be tested. While this leads him to discard Strict CV as less scientific, he takes the argument even further and proposes that the Standard Theory, too, get rid of any empty nuclei it allows for. The comparison Strict CV/Standard Theory does not end here, though. In addition to the problem of empty nuclei, the author also discusses whether Strict CV really has gotten rid of branching, as is usually claimed. As he points out, branching is not basic but really an instance of licensing, which Strict CV also has to allow for. In other words, the alleged simplification, as Ploch claims, is only imaginary. This also drives one crucial point home: representations can only be evaluated in their propositional content, not their definitions.
Next, Ploch turns to arbitrariness, for which the Phonetic Hypothesis is a classic example, i.e. the assumption that phonology is grounded in, or driven by, phonetics. Consider a change such as kt > tt (e.g. Latin doctor > Italian dottore). If one wants to claim that this is due to ease of articulation, then why did Classical Arabic kataba go to ktIb in Moroccan Arabic? Clearly, one needs an opposing force to ease of articulation in order to accomplish this. However, as soon as we allow for opposing forces that can be applied rather flexibly (whenever there seems to be need for them), we lose any hope of testability. Similar points can be made for Lexical Phonology or Optimality Theory: In a Lexical Phonology analysis of English, one could say that velar softening applies in electric/electricity, but not in kick/kicking (*kissing) because the marker -ing is added at a later stratum than -ity. If we restrict velar softening to the earlier stratum, it will no longer be active when -ing is added and the correct result can be derived. However, there is no way of falsifying this: we can conveniently place velar softening and certain suffixes in any stratum we like, depending on whether /k/ goes to /s/ or not. Similar problems hold for Optimality Theory, where the very innovation of violable constraints guarantees untestability. Constraints can be ranked in any order, so if we do not see any effects of a certain constraint, we can just assume it is ranked very low and thus has no consequences. (Not to mention that there is no limit to what qualifies as an OT constraint -- since any constraint can be violated, there seems to be no a priori reason to disallow any kind of constraint.)
To sum up, Ploch's article is a vigorous plea for more testability and scientificness in the formulation of linguistic theories. Phonology, as the author claims, "could finally leave its pre- or pseudoscientific period and enter an age of enlightenment long overdue."
"Eerati tone: towards a tonal dialectology of Emakhuwa" Farida Cassimjee and Charles W. Kisseberth (pp 203--222) This article is a sketch of the tonal system of Eerati, a dialect of the Bantu language Emakhuwa. Emakhuwa has a highly predictable tonal system where there is no distinction in the tonal patterns of stems, but there exists grammatical tone. The dialects fall into two main groups, the doubling and the non-doubling group. A non-doubling dialect is one where a tone is realised on the mora it is underlyingly specified for. In a doubling dialect, on the other hand, a tone is realised not only on its "own" mora, the so-called sponsor, but also on the one following it. Eerati belongs to this latter group.
Cassimjee & Kisseberth offer an account of the facts of Eerati tone in their own version of Optimality Theory: Optimal Domains Theory. Whilst standard OT believes in the appropriateness of autosegmental representations, which to the authors is "intimately tied up to the notions of both underspecification and derivation", Optimal Domains Theory takes a different angle: a feature specification F is only realised iff it is located on a segment in an F-domain. Such F-domains are construed as entirely parallel to other units of phonological structure. After laying out some basic constraints responsible for tonal patterns and discussing an another doubling dialect, Ikorovere, the stage is set for Eerati. Ikorovere was a "well-behaved" doubling dialect, where the high tone is realised on its sponsor and the following mora. Eerati patterns similarly in many cases, but there remains a substantive amount of forms where the tone is realised exclusively on the mora following the sponsor. Cassimjee & Kisseberth propose a constraint ranking to account for these facts and discuss some additional constraints needed. The data from Eerati might make the dialect look like a tone shifting one, but Optimal Domains Theory allows for working out the parallels with Ikorovere and for classifying Eerati as doubling, too.
"Government Phonology and the vowel harmonies of Natal Portuguese and Yoruba" Margaret Cobb (pp 223--242) Cobb's article provides a reanalysis of vowel harmony in Natal (Brazilian) Portuguese and Yoruba. Both systems have been dealt with in the framework of Government Phonology before, but since the theory of elements has been changed massively in the meantime, a fresh look at the facts is called for. Both languages have a seven-vowel system (i e E a O o u) and in both cases it is E and O which undergo vowel harmony.
Earlier analyses argued for ATR-spreading across an A-bridge, i.e. the element ATR would spread provided that both the source and the target contained an A. This begs the obvious question why there should be such an intimate relationship between ATR and A, i.e. why does the spreading potential of ATR depend on A? Furthermore, in revised Government Phonology the element ATR has been done away with, i.e. the proposal is not expressible in those terms any longer. The property of ATR is nowadays represented by the presence or absence of a head in a phonological expression. Headedness, however, is an attribute which can be transmitted to a preceding phonological expression in a principled fashion (head- licensing). Cobb shows how the facts of Natal follow from this mechanism of head-licensing in a very simple and elegant way. In Natal, for example, E/O are harmonised to e/o before e/o because the triggers are headed and pass on this property to E/O which thus go to e/o. However, the correct characterisation of the phenomenon requires a certain fine-tuning of head- licensing. Cobb demonstrates that, like other types of government, head- licensing can parametrically be subject to the Complexity Condition. The effects of this vary: in Natal, the head-licensor has to be more complex than the licensee, "superfluous" elements in the licensee are delinked. In Yoruba, on the other hand, only complex expressions are possible governors. Analysing vowel harmony in terms of head-licensing predicts that there are in fact two kinds of [e] and [o]. One kind is lexically generated and the other one derived from E/O via head-licensing. In other words, we should expect to find e/o before headless expressions, too. As Cobb shows, this prediction is borne out by the facts.
"Palatalisation in Brazilian Portuguese" Thais da Cristófaro-Silva (pp 243--257) This article discusses various aspects of the palatalisation of /t/ and /d/ in a number of Brazilian Portuguese dialects. Palatalisation is seen as spreading of the I-element. The author shows convincingly how spreading occurs within indepently defined licensing domains only. Dialectal variation follows from minimally different conditions on the kind of licensing domain involved: In certain dialects spreading of the I- element can only occur with nuclear licensor, in other dialects any licensor will do. The paper also shows how palatalisation interacts with other processes of the language such as onset simplification, which accounts for the apparent "blocking" of spreading.
"Two notes on laryngeal licensing" Michael Kenstowicz, Mahasen Abu- Mansour and Miklós Törkenczy (pp 259--282) The authors build upon Lombardi's typology of voicing assimilation and neutralisation, which emerges as a consequence of the ranking of several constraints: Positional Faithfulness (identity of a tautosyllabic presonorant obstruent in input and output), Context-free Faithfulness ([voice] remains the same in input and output), Markedness (*[voice]in obstruents) and Uniformity (obstruent clusters agree in voicing). This also allows for a differentiation between final devoicing and regressive assimilation.
However, this approach runs into problems with Hungarian. Hungarian has no final devoicing but regressive voicing assimilation in clusters, both medially and finally. Since word-final consonants exclusively go into the coda, the special status of onset positions (i.e. that they license a voicing distinction) cannot be made use of and the constraint *[voice] should determine the outcome of word-final clusters, i.e. we should expect only voiceless obstruents clusters finally. This is contrary to the facts.
The authors propose that Lombardi's Onset Licensing rather be seen as subcase of a more general notion of phonological/phonetic salience of which the privileged status of onsets ist just one manifestation. Hungarian stops in prepausal position are saliently released, which is an important parsing cue. Hungarian voicing can thus be better analysed by the phonetically motivated Laryngeal Licensing Constraint "The feature [voice] is licensed in contexts of salient release." In other words, prepausal position can license voicing contrasts in the same way as onsets can. As a further positive consequence, this constraint also turns out to be useful in the analysis of word-final geminates in colloquial Hungarian: final sonorants are degeminated, obstruents are not. This only makes sense under the assumption that prepausal obstruents have salient release.
Another important consequence is the independence of syllabic affiliation. The proposed constraint is stated in segmental terms, which becomes crucial in a number of Arabic dialects the authors turn to next. In Daragözü, for example, the voiced pharyngeal in a coda position does not devoice before sonorants. The Laryngeal Licensing Constraint allows for a characterisation of this without making reference to syllabic affiliation. The authors come to the conclusion that features like release, which are often regarded as insignificant, do have an impact on phonological structure.
"On spirantisation and affricates" Tobias Scheer (pp 283--301) This paper deals with (a certain type of)spirantisation and the absence of stops for certain places of articulation and tries to show that these two seemingly independent phenomena share a common cause.
Scheer distinguishes two kinds of spirantisation. The first type targets aspirated stops only (i.e. in some sense aspiration triggers the change) and involves a change in the place of articulation. The second type is triggered by a sonorous context and no shift in the place of articulation is to be observed. While the second type somehow seems to be expected (a segment becomes more sonorous in a sonorous environment), the first type is not so clear: What should be the relationship between aspiration and the shift of place of articulation?
Scheer offers a solution to the puzzle by proposing that aspirated stops are in fact contour segments (two melodic expressions linked to one point) with the element A as its second expression, and that this A is interpreted as aspiration. What happens if the two expressions have to be collapsed into one? Since A is also a place definer, we predict a change in the place of articulation when the merger applies, and this is in fact what happens. However, this is not the end of the story. As A combines with the rest of the melody, spirantisation occurs. In other words, the stop element ? seems to get lost. Scheer's claim here is that this is in fact the result of the incompatibility of the two elements A and ?. Whenever the two are forced to combine due to some phonological process, a sort of repair strategy has to be applied, with two possible outcomees. In one case, the stop-element is lost, yielding a fricative -- this is what we see as "spirantisation". The other possibility is to split up the phonological expression in two and to get yet another kind of a contour segment, an affricate: the first part carries the stop-element and the second part the A-element. The two questions posed at the beginning thus receive a simple answer: In cases of spirantisation, where A becomes integrated in a segment, the element responsible for occlusion is lost. On the other hand, places of articulation that are (even if only in part) characterised by A, could not have corresponding stops, only fricatives -- this is due to the inability of A and ? to combine. What is possible, however, is having affricates instead.
"Branching onsets in Polish" Eugeniusz Cyran (pp 303--320) The clear distinction between "true" branching onsets and onset-onset clusters which just happen to be similar to branching onsets is a central feature of Standard Government Phonology. In his article, Cyran critically reviews the evidence for branching onsets in Polish, and suggests that there is no real need to distinguish between those two types and also often no means to tell them apart.
For example, branching onsets usually block Proper Government from applying across them, yet Polish has triconsonantal clusters that seem to contradict this and call for a parameter to control this. In other words, the branching onset analysis is available, but so is one involving a sequence of onsets, were it not for certain distributional restrictions holding between the second and the third member of the cluster, which is typical of branching onsets. Another diagnostic, Government Licensing, is equally unrevealing. Usually empty nuclei repel Proper Government and phonetic non-interpretation in order to government license a preceding branching onset. In Polish, however, empty nuclei are silenced by Proper Government. An analysis making use of onset-onset clusters is again problematic, but possible. The really troublesome cases come from the vocalisation of yers in Polish, particularly in verbal prefixes. In monomorphemic words, certain sequences of three consonants (where the second and third consonant could or could not form a branching onset) are realised without any empty nuclei popping up. This generalisation does not hold when the first consonant is part of a prefix. Morphological bracketing might be a viable solution, but then one would have to say that bracketing depends on the syllabic structure: No brackets preceding branching onsets and vice versa. Even if that were true, certain triconsonantal clusters where C2 and C3 could not possibly qualify as branching onsets still behave as if there was no bracketing. In other words, the crucial distinction is not branching onset vs. onset-onset cluster but rather whether the stem-yer ever shows up. The word-final context is yet another problem case. While at first glance Polish seems to have final branching onsets, the idyllic situation is marred when one takes derived forms into account (where the branching onsets are broken up, which argues against their being branching onsets at all)or other clusters that could not possibly form branching onsets and yet behave like the ones that can. To sum up, the arguments for branching onsets in Polish are weak, to put it mildly -- which leads Cyran to the more general question how we can know whether a language has branching onsets in the first place.
"Are there branching onsets in Modern Icelandic?" Edmund Gussmann (pp 321- -337) Taking one of the central axioms of Government Phonology as a starting point, viz. that constituents are maximally binary branching, this article sets out to have a closer look at Icelandic which on the surface has clusters of up to four consonants, e.g. strjúka 'stroke'. Following the arguments from Kaye (1996), /sC/ can never form a constituent, thus the four-member cluster immediately reduces to a three-member one. Gussmann observes that all the remaining cases end in /j/ which can be argued to occupy an onset of its own. In order to test the status of the remaining clusters, i.e. whether they qualify as branching onsets or not, vowel length is called upon. Icelandic is well-known for having a reliable system of tonic lengthening, i.e. in stressed syllables a non-branching nucleus is followed by a coda-onset cluster, and a branching nucleus is followed by an onset, branching or non-branching. The syllabification model endorsed by Government Phonology, i.e. analysing word-final consonants as onsets, remedies some of the shortcomings of traditional approaches and makes sure that vowel length provides the appropriate testing environment for the onsethood of certain clusters. The results of Gussmann's scrutiny are surprising: Traditional analyses of Icelandic take as branching onsets only the combinations of /ph, th, kh, s/ with /j, v, r/, to which test set Gussmann adds the combination neutral stop plus /j, v, r/ as well as obstruent plus /l, m, n/; but only very few of these candidates actually pass the onsethood test: All we are left with are obstruents followed by /r/ and possibly /v/.
"Remarks on mutae cum liquida and branching onsets" Jean Lowenstamm (pp 339--363) This article argues that branching onsets (mutae cum liquida) of the Indo- European kind are in fact illusionary. What sets English and Arabic apart in that respect is that the former, but not the latter, contains segments with liquids as a sort of secondary articulation. In other words, branching onsets are in reality monosegmental, but complex. Lowenstamm starts off with evidence from Chaha, where based on the templatic character of the morphology both muta cum liquida sequences as well as monosegmental muta cum liquida segments have to be countenanced. Further support comes from metrical theory: if stress systems are so sensitive to the branching vs. non-branching character of the rhyme, then how come the branching of onsets never plays any role? Facts from reduplication are also brought to bear on this issue: Ilokano and Ancient Greek are contrasted and the differences between the two systems are reduced to a simple parameter which controls whether features of secondary articulation are (Ilokano) or are not (Greek) reduplicated. This very parameter also accounts for why aspiration, another feature of secondary articulation, is not copied in Greek. The final set of data comes from Czech where vowel- zero alternations in prefixes can be handled quite easily once the monosegmental status of mutae cum liquida is accepted. As an additional benefit, alternations between mutae cum liquida and muta plus syllabic sonorant fall out naturally as well.
"Defective syllables: the other story of Italian sC(C)-sequences" Emannuel Nikièma (pp 365--383) This article tackles the distribution of the Italian definite masculine article, /il/ vs. /lo/. Nikièma shows how the various triggering contexts can be neatly captured in two groups, one containing an initial defective syllable (where either the onset or the nucleus is empty) and the other one starting with a non-defective syllable. The sC(C)-clusters are analysed as s plus empty nucleus plus C(C), contra Kaye (1992), and thus are comparable to vowel-initial words: while the former structure is defective due to the empty nucleus, the latter is defective because of the empty onset. Accordingly, both of them take /lo/. Non-defective initial syllables, however, opt for /il/. In other words, when the initial syllable is defective, the determiner is not and vice versa. The analysis can be extended to both rarer clusters and other alternations in determiners. Also the facts from raddoppiamento sintattico find a natural explanation in Nikièma's account.
"Remarks on prenominal liaison consonants in French" Yves Charles Morin (pp 387--400) This article questions two aspects of conventional wisdom that are widely held as regards French liaison with adjectives: (1) the adjectives in question end in a latent consonant which is realised in the appropriate context and (2) the feminine form is identical to the underlying form, i.e. the one including the latent consonant. Thus, Morin tests native speakers to see whether the liaison consonant can be elicited automatically once the appropriate context is provided and whether the liaison-consonant really is the same as in the feminine. Both assumptions are disconfirmed.
Morin proposes that liaison after prenominal adjectives should rather be seen as a form of adjectival declension, similar to what we find in most Germanic languages. The only obstacle seems to be the fact that the inflectional ending appears as the initial consonant of the following noun. This can be remedied by reanalysing liaison consonants as part of the following noun; in other words, Morin suggests a kind of status constructus solution, where the head is marked depending on the presence or absence of a (certain) complement. The kind of liaison consonant defines morphological class membership; however, class change is predicted in such a model and this is indeed what we find in an "embryonic" state.
"The phonotactics of a 'Prince' language: a case study" Glyne L. Piggott (pp 401--425) This paper provides an encompassing analysis of Selayarese, a typical 'Prince' language (named after Alan Prince), i.e. a language with exclusively CVC-syllables where the coda cannot license distinctive material. Piggott's analysis is basically couched in Government Phonology, with certain extensions like a particular view on licensing as well as the assumption that there is a coda and a syllable node.
After having gone through an outline of Selayarese syllable structure, Piggott proposes a "Prince Coda Condition", i.e. a definition of what is characteristic of a 'Prince' language: the coda is not a feature licenser, any material hosted by it must receive its license from somewhere else. In the case of geminates this is easy: the material is provided by the following onset. Alongside this, non-existent clusters like *st, *fp or *rt can be readily excluded. Nevertheless, additional adjustments have to be made: Selayarese does not allow for voiced obstruents to geminate (*bb etc.) and Piggott proposes a constraint to exclude them. Sonorants, however, are to be found in the coda position and it is suggested that they are licensed "by default". The same goes for the glottal stop and we derive clusters such as /?l/ or /?b/. Here we notice a certain gap that interestingly enough is a mirror image of the voiced obstruents: while ?b for example does exist (to the virtual exclusion of *bb), *pp is ungrammatical and instead we find ?p. This can be observed particularly well in the case of epenthetic material that shows up between a root and certain suffixes. If the stressed syllable is headed by an epenthetic vowel, it has to be closed by gemination or a ?C-cluster. Tonic lengthening, though existent in the language, is not an option. Epenthetic vowels, Piggott claims, stand in a particular relationship with the preceding nuclear position from which they gain their melodic material. The epenthetic vowel is the dependant part in this relationship and can thus not be longer than its head. As a result of this, the required heaviness of the stressed syllable has to be created some other way, and accordingly we observe either gemination of the following consonant or, if that is not a viable solution, insertion of a glottal stop in the coda position.
Piggott further shows that Selayarese also has initial and final codas. While the former is not so unexpected in Government Phonology, where initial empty nuclei are allowed for under certain conditions, the latter claim is rather unorthodox: Coda Licensing requires that every coda be licensed by a following onset and thus effectively excludes domain-final codas. Piggott argues against this by pointing to the fact that in final position we find the same consonants which in internal position can be shown to be independent of following material, i.e. the velar nasal and ?. However, since this material has to be licensed in some way, it is proposed that the syllable node functions as the formal licenser of material in syllable-final (and also word-final) position.
"On the syllabification of right-edge consonants --- evidence from Athna (Athapaskan)" Keren Rice (pp 427--448) This article argues that right-edge consonants cannot be universally syllabified as either onsets or codas (or as a third possibility: extraprosodically), but that both options have to be available in one and the same language. By bringing evidence from Athna to bear on this issue, Rice demonstrates the different properties that set the two structures apart in that language. For example, while voicing and spirantisation do not affect coda consonants, onsets do get targeted by these processes. The very same effects can be shown to be at work not only at the right edge, but also elsewhere in the domain, thus supporting Rice's proposal.
Rice further proposes that in order to arrive at a correct characterisation of nasalisation in Mentasta, one of the four major dialects of Athna, one has to assume that there are two kinds of empty nuclei: One which is present for purely structural reasons (and causes nasalisation) and the other one has a morphological source, as a consequence of which nasalisation fails to apply.
Both innovations seem to beg the question how the proposed structures can be learnt by the child. Rice sketches out acquisition strategies that guarantee for the child to come up with the correct representations.
"Licensing constraint to let" Yuko Yoshida (pp 449--464) Yoshida argues that the licensing constraints of Ancient Chinese still have a reflex in the phonology of loan words in modern-day Japanese.
The structure of Chinese words is determined by a four-positional template consisting of two ON-pairs. The first nucleus assumes headship for the entire domain and thus has to be filled. The second nucleus on the other hand can either be filled (in which case its onset has to be empty) or empty (and the onset has to be filled). In other words, in the second ON- pair one and only one position must be filled and, vice versa, one and only one position must be empty.
A reflex of this can be found in Japanese pitch accent. While native words with two nuclei are rather unpredictable in their stress patterns (initially stressed, finally stressed or unstressed), loans from Chinese have initial stress in the majority of the cases. (Those that do not seem to belong to a particular morphosyntactic class like numerals.) The fact that the first nucleus was the head of the domain in Chinese is in some sense retained in Japanese where that position functions as the metrical head and is assigned pitch accent.
Related to that issue is the emergence of the syllabic nasal in Japanese, /N/, which is in complementary distribution with /nu/. Yoshida argues that the phonological representation is identical for both cases: /N/ is never accented and never appears in word-initial position, while /nu/ usually bears stress.
"Empty and pseudo-empty categories" Monik Charette (pp 465--479) Charette's article investigates the properties of pseudo-empty categories, which had been introduced to account for vowel-glide alternations in French by Haworth (1994). A pseudo-empty category is a position which shares all its segmental material with an adjacent position. If such a category is followed by a possible proper governor, p-licensing will occur. As a consequence, the material in the nuclear position is suppressed phonetically and realised in the adjacent consonantal position only. Similar to other cases involving empty categories, p-licensing will be blocked if the governee has to license a preceding branching onset. Charette now takes a wider context into account. What happens to empty positions that are followed by a pseudo-empty position? Here, two types of structure emerge: in the first one, Continental French allows for p- licensing of the empty position, while Québec French varies. In the second one, p-licensing of the empty position is blocked in both Continental and Québec French. The two types can be characterised by the kinds of consonants flanking the empty nucleus. If the two consonants can enter into a governing relationship, p-licensing of the straddled empty nucleus is possible. Otherwise, the empty nucleus has to be realised. Charette claims that the governing relationship between the two consonants is not enough to silence the empty nucleus, an external governor is needed. The pseudo-empty category immediately following is itself licensed and thus cannot assume that role. The only eligible governor is the filled position following the pseudo-empty position. In other words, a full vowel can govern twice: one time locally (the preceding pseudo-empty position) and another time non-locally (the empty nucleus preceding the pseudo-empty position), provided the consonants flanking the empty nucleus can enter into a governing relationship.
"Unlicensed domain-final empty nuclei in Korean" Yong Heo (pp 481--495) This article explores the reason for why certain empty nuclei at the right edge of a domain in Korean have to be realised despite the fact that the language allows final nuclei to remain empty. In particular, Heo discusses the behaviour of the derivational suffix /p_/, whose final nucleus is realised if the suffix follows a consonant-final stem but remains empty otherwise. By investigating the alternations of the suffix-initial /p/, one can show that in the case of consonant-final stems a government relationship between the stem-final consonant and the /p/ is created. Fur such a relationship to hold, however, the governor /p/ needs to be licensed by the following (domain-final) nucleus, which thus cannot remain silent. No such government obtains in the case of vowel-final stems, since there is no adjacent consonant /p/ could govern. Accordingly, the final- nucleus can remain /silent.
In those cases, the final nucleus gives up its licensed status when it is called upon to do some licensing itself. Another conceivable possibility would be for the nucleus to remain silent, which means something has to happen to the preceding onset. Heo makes use of this second option in his analysis of the peculiar behaviour of stems ending in /rr/. Here, depending on the variety, the following nucleus can remain silent, which leads to the simplification of /rr/ to /r/.
"Unreleasing: the case of neutralisation in Korean" Seon-Jung Kim (pp 497-- 510) This article investigates various neutralisation patterns of final consonants in Korean: Aspirated and tensed plosives become neutral, fricatives and affricates turn into a plosive and /r/ is realised as [l]. Kim demonstrates how all these alternations can be accounted for neatly by assuming that (1) final empty nuclei cannot license the element H and that (2) final consonants must contain the stop-element ?. In other words, the right edge position imposes serious conditions not on the number, but on the kind of elements that can be hosted.
"/r/ syllabicity: Polish versus Bulgarian and Serbo-Croatian" Grazyna Rowicka (pp 511--526) Drawing on data from Polish, Bulgarian and Serbo-Croatian, Rowicka puts forth her proposal that there are two criteria for syllabicity: a metrical one and a phonotactic one which need not necessarily overlap. Both Polish and Serbo-Croatian have syllabic consonants (from a phonotactic point of view), yet what sets Serbo-Croatian apart from Polish is that syllabic consonants count for the metrical structure (e.g. in their ability to bear stress) in the former, but not in the latter. Bulgarian, on the other hand, does not have syllabic consonants of either kind. Rowicka then goes on to show that certain distributional facts about complex clusters in Polish can be captured relatively easily once we assume the existence of syllabic sonorants in the language, thus making the facts of Polish relatively similar to those of Serbo-Croatian. The analysis is done in Rowicka's framework of left-headed (trochaic) proper government with a strict CV skeleton and a set of OT constraints. In addition to that, considerations on the simplicity of the system lead to the conclusion that the proper representation of a syllabic consonant is one where it is linked to an onset and spreads to the following empty nucleus.
"The syllabic nasal in Japanese" Shohei Yoshida (pp 527--542) This article has two objectives: Firstly, to provide an accurate description of the allophonic realisations of the Japanese syllabic nasal /N/ and the typical mistakes language learners make when studying Japanese. Secondly, to provide an appropriate representation of /N/ within Government Phonology and to show how the allophonic variation can be derived. Yoshida proposes that the syllabic nasal is an ON-pair together with a floating N-element which is linked to the skeleton as needed. For example, in word-final position the N is linked to the nuclear slot in order to fulfill the requirement that in Japanese no domain may end in an empty nucleus. Additional spreading of the melodic material associated to the preceding nuclear position accounts for why in casual pronunciation the syllabic nasal is realised as a nasalised copy of the preceding vowel. In careful pronunciation, however, word-final /N/ is realised as a uvular or a velar nasal, despite the fact that the N-element is linked to a nuclear position. Yoshida's claim is that this is nothing but a phonetic effect, due to the well-known fact that in Japanese prepausal words are frequently followed by a glottal stop. The sequence of /N/ plus glottal stop is interpreted as a nasal stop. This is a word-internal phenomenon only, no such effects can be observed word-internally. Accordingly, word- internal N is realised as a nasalised high, back, unrounded vowel in careful pronunciation or as a nasalised copy of the preceding vowel or as the nasalised vocalic version of the following glide in casual pronunciation. Last but not least, the representation for sequences of /N/ plus non-continuant is slightly different. Here, the floating N is linked to the onset position of the ON-pair and homorganicity is guaranteed by acquiring the relevant melodic material responsible for occlusion and place of articulation via spreading from the following onset.
"Template and morphology in Khalkha Mongolian --- and beyond?" Ann Denwood (pp 543--562) This article is an attempt to apply a tool which has proved useful in the analysis of Chinese (Goh 1996), viz. a four-positional template, to Khalkha Mongolian. Denwoood shows that the conditions on which positions of such a template are realised are in fact virtually identical. Since Khalkha Mongolian has a richer inventory of syllabic structures than Chinese, Denwood proposes that combinations of such four-positional templates will allow to build up the required representations and account for phonotactics in a straightforward way. By this it becomes evident that the different morphological status of stem vs. affix is reflected in restrictions on where melodic material can associate to the template. The article finishes off with some tentative suggestions on how the model of a four-positional templatic could be extended to other languages. Whilst English remains a challenge (despite some promising similarities), Turkish seems to be a more successful candidate. This also sheds some light on possible parametric variation in the templatic structure.
"A non-derivational analyysis of the so-called 'diminutive retroflex suffixation'" Yeng-Seng Goh (pp 563--580) This article challenges some traditional assumptions about the representation of the so-called diminutive retroflex suffix [@r] which derives etymologically from the independent word for 'child' or 'smallness'. Goh puts into doubt the status of [@r] as a suffix. In his view, the suffixed forms comply with the minimal phonological string; the "suffixed" forms are thus unanalysable minimal words.
The minimal domain in Beijing Mandarin consists of four positions, viz. two ON-pairs where a special condition holds for the second pair. Either O2 or N2 has to be empty, no domain can have both of them filled simultaneously. Goh shows that simple addition of the "suffix" [@r] to this template yields incorrect results; instead, the correct analysis is one where a simple retroflex [r] occupies 02 of the template. This causes all kinds of adaptations in the four-position template: shortening of long vowels (which in the "un-suffixed" form span two nuclear positions), complete loss of final consonantal material or "evacuation" of part of this material into preceding positions. While the resulting forms comply with the phonological patterns of Beijing Mandarin, the changes themselves are not phonological in nature. This is strong evidence to show that diminutives of this kind are in fact unanalysable words.
"Why Arabic guttural assimilation is not a phonological process" M. Masten Guerssel (pp 581--598) This article argues against the traditional analysis of guttural assimilation in Arabic as a phonological process. Guerssel proposes that the phenomenon in question is dictated by the principles of non- concatenative morphology.
Guttural assimilation is the term applied to the process of /i/ becoming /a/ when standing next to a guttural (uvulars, pharyngeals, glottals). Oddly enough, this process fails to apply in the majority of cases. It is observed in the imperfect active of Form I exclusively, i.e. in one out of 60 possible patterns. The phonological status of this phenomenon is thus highly questionable. After a short sketch of the general theory of apophony, an apophonic path taking zero to /i/, /i/ to /a/, /a/ to /u/ and /u/ to /u/, as laid out in Guerssel & Lowenstamm (1996), the basic mechanisms of constructing Arabic verbs are discussed. Each triliteral root belongs to one out of four classes, characterised by an (unassociated) thematic vowel: i, a, u or zero. The derivation of the categories of voice and aspect proceeds along a kind of flow-chart, where the triliteral root with an /a/ in the first nuclear position (e.g. /fat_q/) serves as the input. Simple association of the thematic vowel to the second nuclear position yields the active perfect ([fataq]), while the passive perfect is derived by applying apophony the the input ([futiq]). By prefixation and further application of apophony to the thematic vowel the corresponding imperfect forms can be created: [ja-ftuq] for the active, [ju-ftaq] for the passive. Verbs where the thematic vowel is zero are special, however. Since there is no thematic vowel to be associated to the root, a phonological process of a-insertion takes place to make sure the output is well-formed. This gives us an active perfect such as [kasar] (from the morphological output /kas_r/ which is the same as the input). The passive perfect is derived in the normal way, i.e. apophony takes /kas_r/ to [kusir]. When the second or third literal is a guttural, however, the output is different: An input such as /xad_9/ [9 = voiced pharyngeal, M.A.P.] surfaces as [xada9] in the perfect and [ja- xda9] in the imperfect. Guerssel's elegant proposal is that all gutturals contain the element A in their internal make-up and that this A spreads to the adjacent empty position. In other words, the second [a] in [xada9] is not due to the phonological process of a-insertion but to a morphological operation whereby an root is associated with its thematic vowel. Since the root xd9 lacks a thematic vowel, morphology makes use of the A-element contained in the guttural. Since the result is a well-formed string ([xada9]), it can surface as such and also serve as the input for the imperfect formation. Since the second [a] of [xada9] is linked to both a vocalic and a consonantal slot, it is immune to apophony and we thus derive [ja-xda9].
"On a certain notion of 'occurrence': the source of metrical structure, and of much more" Jean-Roger Vergnaud (pp 599--632) Vergnaud's article demonstrates that the correspondence between metrical grids and beats is inherent in the formal definition of the metrical grid. His formal account involves a model of metrical structure which relies upon the notion of "occurrences" of units, in much the same sense as in syntax. In other words, Vergnaud attempts to sketch out a strong parallel between the behaviour of phonological and syntactic units. The notion of occurrence calls for a distinction between two kinds of properties of entities: those properties that identify the type of an entity and those that identify an instance of that type. The properties defining an object are intrinsic to that object, while those that define the instance are not and can be freely associated with all objects. The properties defining the instance in a sense function as a contextual feature that represents the object's ability to combine with other objects. As a consequence of that, the phonological string is split up, it is really a pairing of two strings. An immediate consequence of this is the recursivity of linguistic expressions.
The definition of stress we end up with is the following: A stressed unit is one that cannot be interpreted as a context for another unit. This is the traditional interpretation of primary stress as a mark of juncture. The occurrence feature of a stressed position is left "dangling". Vergnaud suggests that this might lead to an anticipatory gesture. Such anticipation is an important component of phonological strings. This approach to the properties of metrical theory has far-reaching consequences for syntax, too, Vergnaud concludes. More generally, it may lead to an understanding of the general principles that govern the ordering of categories across linguistic levels.
GENERAL COMMENTS & SUMMARY
As said before, this volume will become indispensable for anyone working in the framework of Government Phonology. The contributions are interesting, stimulating to read and highly relevant for current issues in the theory. Basically any topic that has ever been important in the theory is addressed in at least one of the articles. Many ideas that have been around for quite a while are extended, improved or formulated more precisely. The more than 600 pages of text also pave the way for interesting developments in the future: loads of interesting questions are raised and intriguing solutions have come into sight. It is well to be expected that this book will spawn a flurry of further research.
To take but one example, the discussion on Standard Government Phonology vs. Strict CV has been going on for a number of years. While the contribution by Ploch questions the scientific status of Strict CV (and also major parts of the Standard Theory), the article by Rennison & Neubarth and the one by Lowenstamm attempt to find a solution for one of the major challenges for Strict CV, namely branching nuclei. This adds new spice to the discussion and will certainly be of importance for a number of years to come.
The list of contributors is basically a "who is who in Government Phonology" -- nearly all of the articles are written by major proponents of the theory, spanning Europe, Asia, the Americas and Africa.
The book should also be accessible to people whose background in Government Phonology is not that strong. The articles are well written, fully indexed and have an extremely large reference section.
To sum up in a nutshell, this book it a cutting-edge contribution to the theory of Government Phonology, or rather to phonological theory in general. Due to the enormous number of current issues covered and the innovative approaches, it should hold pride of place in any phonologist's library.
Berwick, Robert. 1991. Computational complexity theory and natural language; a paradox resolved. Theoretical Linguistics 17: 123--157.
Clark, Robin and Ian Roberts. 1993. A computational model of language learnability and language change. Linguistic Inquiry 24: 299--345.
Dresher, B. Elan & Jonathan Kaye. 1990. A computational model for metrical phonology. Cognition 34. 137--195.
Gibson, Edward & Kenneth Wexler. 1994. Triggers. Linguistic Inquiry 25: 407--454.
Guerssel, Mohand & Jean Lowenstamm. 1996. Ablaut in Classical Arabic measure I active verbal forms. In: Jacqueline Lecarme, Jean Lowenstamm & Ur Shlonsky (eds). Studies in Afroasiatic Grammar. The Hague: Holland Academic Graphics, 123--134.
Haworth, Emmanuelle. 1994. The trouble with French glides. SOAS Working Papers in Linguistics and Phonetics 4: 53--70.
Kaye, Jonathan. 1992. Do you believe in magic? The story of s+C-sequences. SOAS Working Papers in Linguistics and Phonetics 2: 293--313.
Kaye, Jonathan. 1996. Do you believe in magic? The story of s+C-sequences. In: Henryk Kardela & Bogdan Szymanek (eds). A Festschrift for Edmund Gussmann from his Friends and Colleagues. Lublin: The University Press of the Catholic University of Lublin. 155--176.
Lowenstamm, Jean. 1996. CV as the only syllable type. In: Jacques Durand & Bernard Laks (eds). Current Trends in Phonology. Models and Methods, volume 2. European Studies Research Institute/University of Salford. 419-- 443.
Ristad, Eric S. 1990. A constructive complexity thesis for natural language. PhD dissertation, Massachusetts Institute of Technology.
van der Hulst, Harry & Nancy Ritter. 1998. Kammu minor syllables in Head- driven Phonology. In: Eugeniusz Cyran (ed). Structure and Interpretation. Studies in Phonology. (PASE Studies and Monographs 4.) Lublin: Folium, 163- -182.
ABOUT THE REVIEWER:
ABOUT THE REVIEWER
Markus A. Pöchtrager is a PhD student of general linguistics at the University of Vienna. His main interests are phonological theory, in particular Government Phonology (and variants/predecessors thereof). At the moment he is finishing up his doctoral dissertation on the structure of phonological length in Finnish, Estonian and German; this also involves a simplification of the element calculus of Government Phonology, in particular getting rid of the so-called stop element.