Synchronic functional grounding: Evidence from positional augmentation

Transcription

1 : Evidence from positional augmentation Jennifer L. Smith UNC Chapel Hill Manuscript date: May 2004 Comments welcome Abstract Phonological patterns are often phonetically natural, but the reason for this observation is under debate. Two opposing accounts of the influence of phonetics on phonology are the synchronic grammar-internal functional grounding model, which holds that phonetic factors directly restrict the contents of the formal phonological grammar, and the diachronic misperception and re-phonologization model, which holds that phonetic factors influence phonology only by determining the course of diachronic change. This paper shows that the functional restrictions on positional augmentation constraints (markedness constraints on phonologically strong positions) are incompatible with the diachronic misperception model, providing support for the inclusion of synchronic functional grounding in the phonological grammar. 1. Introduction One of the fundamental questions that drives research in phonological theory and the phonology-phonetics interface is this: Why do phonological patterns so often involve wellformedness conditions that make sense phonetically? Some researchers propose that the phonetic naturalness of phonological patterns comes about because phonetic factors have effects internal to the phonological grammar; that is, the formal grammar is functionally grounded (Archangeli and Pulleyblank 1994). Recent proponents of this position include Flemming (1995), Boersma (1998), Kirchner (1998), Hayes (1999), and Steriade (2001). Others take the position that phonetic factors provide influence external to the phonology, by way of phonetic effects on speech perception, language acquisition, and diachronic change. In particular, it has been proposed that the phonetic naturalness of phonological patterns is a direct consequence of diachronic change through listeners' misperception of an ambiguous acoustic signal, followed by a reanalysis of the phonological grammar on the basis of the new interpretation of the signal's phonological structure (Ohala 1981, 1993; Blevins, to appear; Blevins and Garrett 1998, 2004; Hyman 2001; Kavitskaya 2001; J.A. Barnes 2002). On this second view, the formal phonology is itself phonetics-free. Phonological grammars tend to be phonetically natural only because phonetic patterns determine what kinds of misperception can arise, and therefore what kinds of diachronic reanalyses can take place. This paper presents a case of phonetically influenced phonological patterning that is not compatible with an explanation based on diachronic change through misperception and reanalysis. Markedness constraints on phonologically prominent positions (de Lacy 2001; Parker 2001; Smith 2000, 2002), also called positional augmentation constraints, are restricted by functional factors such constraints are attested only if the demand they make of their prominent position is one involving the enhancement of perceptual salience (Smith 2000, 1

2 2002). It is argued here that the functional restrictions on these constraints cannot all be reanalyzed as the effects of diachronic misperception and re-phonologization processes, but can be accounted for in a model that incorporates functional grounding internal to the formal grammar. First, background discussion of the debate over how phonetics constrains phonology is given in 2. Next, 3 presents an overview of positional augmentation constraints and the functional restrictions on such constraints. Case studies involving positional augmentation constraints are discussed in 4, to exemplify the kinds of phonological patterns that these constraints are responsible for. Then, 5 examines the problems presented by the case studies in 4 for the view that diachronic misperception and re-phonologization processes are responsible for all apparent phonetic restrictions on possible phonological patterns. This section also outlines a synchronic model of grammar-internal functional grounding that accounts for the patterns observed. Conclusions and implications are discussed in The functional grounding debate This section summarizes two opposing positions concerning the nature of phonetic influence on phonological systems: the grammar-internal functional grounding model ( 2.1) and the diachronic misperception and re-phonologization model ( 2.2). In 2.3, the assumptions of the two approaches are exemplified and contrasted in a discussion of how each would account for the fact that unstressed vowels, but not stressed vowels, are particularly susceptible to reduction to [ ]. Finally, 2.4 discusses the conditions that the diachronic misperception and re-phonologization account would need to meet if it is to extend to all cases where phonological patterning is restricted by phonetic naturalness. 2.1 Functional grounding as a restriction on formal grammars Formal phonological grammars allow for the expression of both 'natural' and 'unnatural' rules/constraints/processes (Chomsky and Halle 1968; Eisner 1997). The problem is that the phonological elements, constituents, or operations that are needed to characterize natural (phonetically motivated) phonological processes can be formally recombined in ways that are not phonetically motivated, as shown in (1) It may be the case that a certain number of 'unnatural,' phonetically unmotivated processes are in fact involved in natural-language phonology (e.g., Bach and Harms 1972; Anderson 1981; Hyman 2001). The question addressed by this paper is, when there are phonetically motivated restrictions on phonological patterns, how those restrictions should be modeled. 2

3 (1) Formal phonological models can express unnatural processes a. Unnatural rule [+nas] > [-voi] / [+lab] 'Nasals become voiceless before labials' Formally similar rules that are natural [+nas] > [+lab] / [+lab] 'Nasals become labial before labials' [-son] > [-voi] / # 'Obstruents become voiceless when final' b. Unnatural constraint *[+NAS, +VOI] 'Segments are not both nasal and voiced' Formally similar constraints that are natural *[+NAS, -VOI] 'Segments are not both nasal and voiceless' *[-SON, +VOI] 'Segments are not both obstruent and voiced' One strategy for addressing the problem of formal overgeneration is to appeal to functional grounding (term due to Archangeli and Pulleyblank 1994), the proposal that phonological entities or processes are based on, determined by, or restricted by functional (phonetic, psycholinguistic) factors. The functional factors themselves are usually assumed to be external to the formal grammar (though for a different perspective see, e.g., Flemming 1995; Kirchner 1998, 2001; Zhang 2001). However, these factors place restrictions on what the formal grammar can express. For example, Archangeli and Pulleyblank (1994) develop a formal phonological model in which all implicational statements of the form If A, then B must be grounded, which, in their terms, means that each implication must "reflect physical correlates of the [feature values] involved" (Archangeli and Pulleyblank 1994, p. 167). The idea that the formal objects and operations in the phonological system are directly constrained by functional factors has a long history; influential proposals along these lines include the discussion of markedness in Chomsky and Halle (1968, ch. 9) and the theories of Natural Generative Phonology (Vennemann 1974; Hooper 1976) and Natural Phonology (Stampe 1973; Donegan 1978; Donegan and Stampe 1979), as well as, for example, many of the theoretical debates concerning distinctive-feature theory or models of feature geometry. Grammar-internal functional grounding has also been implemented in Optimality Theory (OT; Prince and Smolensky 1993; McCarthy and Prince 1995), usually as a requirement that some, or all, of the constraints in the grammar are grounded (have 2 phonetic or psycholinguistic motivation). Examples of OT work that follows this strategy 2. It is important to keep in mind that, while the OT framework facilitates the assumption of grammar-internal functional grounding, it does not entail this assumption. OT provides a formal mechanism that allows functionally motivated tendencies to determine the grammars of natural languages, because the phonologies of individual 3

4 include J. Beckman (1995, 1998), Casali (1996), Flemming (1995), Hayes (1999), Jun (1995), Kirchner (1998, 2001), Padgett (1995, to appear), Pater (1999), Prince and Smolensky (1993, 5.1), Smith (2000, 2002), Steriade (1993, 1997, 2001), Walker (1998), Wilson (2001), Zhang (2000, 2001), and many of the articles in Hayes, Kirchner, and Steriade (2004). 2.2 Another view: Functional factors only influence diachronic misperception and phonologization There is an alternative strategy for handling the problem that formal grammars predict more phonological patterns than actually occur, which rejects the position that the formal phonological grammar is itself functionally grounded. According to this second view, functional factors operate separately from, and externally to, the formal grammar. The only thing that functional pressures do is constrain the way that a grammar is transmitted from one generation to the next, because they directly influence processes such as speech perception, language acquisition, and diachronic change. Specifically, functional factors such as coarticulation, articulatory undershoot, or perceptual confusion cause listeners (particularly learners) to misinterpret the phonological structures that speakers produce, assigning to perceived utterances a phonological structure that is different from that intended by the speaker (Ohala 1981, 1993). Because only certain kinds of phonetic contexts give rise to acoustic signals that are ambiguous enough to be misperceived and re-phonologized, not all of the formally possible phonological processes are actually found in natural language. An advantage of this approach is that, if all apparent functionally based restrictions on the phonological system can be traced back to diachronic changes caused by misperception and phonologization, then there is no need to duplicate functional factors or their effects inside the formal grammar, which would presumably complicate the grammatical model with extra machinery. Work that argues for some version of this position includes Ohala (1981, 1993), Anderson (1981), Blevins (to appear), Blevins and Garrett (1998, 2004), Hale and Reiss (2000), McMahon (2000), Hyman (2001), Kavitskaya (2001), Kochetov (2001), J.A. Barnes (2002), and Yu (2004). languages are the result of interactions among ranked and violable constraints, and constraints are a straightforward way of modeling functional pressures as part of the phonological grammar. However, the OT framework does not inherently require all of the constraints in the system to reflect functionally motivated tendencies. This fact has two consequences. First, if constraints are functionally motivated, we need separate theories of which constraints are grounded and how the grounding is enforced (Eisner 1997; Hayes 1999; Smith 2002). Second, a phonological system that does not assume grammar-internal functional grounding is equally compatible with the fundamental assumptions of OT. 4

5 2.3 Example: The vowel-reduction asymmetry Synchronic functional grounding To contrast these two models of phonetic influence on phonology, it is useful to consider an example of a functionally restricted phonological pattern and see how the two approaches would account for the functional restrictions on the pattern. As is well known, many languages reduce vowels to [ ] in unstressed syllables. However, there are no languages that reduce vowels to [ ] exclusively in stressed syllables. This asymmetry has a functional basis (see Crosswhite 1999 and Barnes 2002 for recent discussion). Namely, unstressed syllables are often shorter than stressed syllables; this is particularly true in languages with vowel reduction. The longer stressed syllables allow fuller realization of the articulatory target for each vowel. On the other hand, the shorter unstressed syllables are subject to articulatory undershoot and are therefore less well able to maintain contrasts among different vowel qualities. This basic pattern is fairly uncontroversial. What is currently under debate is just how the functional basis of the vowel-reduction asymmetry is related to the formal phonological grammar. An example of a theoretical approach consistent with the grammar-internal functional grounding model ( 2.1) is the positional markedness treatment of vowel neutralization (Steriade 1993; Crosswhite 1999). To account for the fact that vowel reduction targets unstressed syllables, this approach posits a constraint *VPLACE/F, a version of the markedness constraint *VPLACE that is relativized to the position 'unstressed syllable'. (2) *VPLACE/F 'No vowel place features, in unstressed syllables' Crucially, this proposal includes the following restriction: An analogous constraint for stressed syllables, *VPLACE/F, must be absent from the constraint set. Otherwise, the system predicts that some languages would rank the problematic *VPLACE/F high in the grammar; the result of such a ranking would be vowel reduction exclusively in stressed syllables, which is the unattested pattern. The difference between these two formally similar, but empirically quite different, constraints is that *VPLACE/F is functionally grounded, and therefore is a legitimate member of the constraint set, while *VPLACE/F is not functionally grounded, so it is not a legitimate constraint. In other words, the asymmetry between stressed and unstressed vowels is accounted for by means of grammar-internal functional grounding, which is implemented as a restriction on which of the formally expressible constraints are actually included in the constraint set Another possible implementation of the grammar-internal functional grounding model would be to propose that *VPLACE/F exists, but is in a universally fixed ranking 5

6 The role of functional factors in the vowel-reduction asymmetry, and the character of the formal phonological grammar, are viewed quite differently under the diachronic misperception and re-phonologization model. (The following discussion is based on the treatment of vowel reduction in Barnes 2002.) The diachronic explanation for the vowel-reduction asymmetry is as follows. Assume that there is a language with no phonological vowel reduction, but with unstressed syllables that are shorter than stressed syllables. In unstressed syllables, speakers may fail to achieve articulatory targets for vowels, as in (3). This leads to variation in the acoustic patterns produced, including tokens in which an intended [a] is produced as something that sounds like [ ]. (3) Failure to achieve articulatory targets in unstressed syllables UR Articulatory intent Acoustic pattern produced /pa ta/ [pa ta] [pa ta] ~ [pa ta ] ~ [pa t ] Given this kind of variability in the acoustic patterns for unstressed syllables, the next generation of learners may misperceive speakers' articulatory intent, and assume that the undershoot form [pa t ] is the intended articulation. If such a reanalysis occurs, vowel reduction will become phonologized. That is, there will have been a change in the phonological grammar (in OT terms, a new, higher ranking for the constraint *VPLACE/F ) such that speakers now intend to produce underlying /a/ as [ ] in unstressed syllables. (4) Phonological vowel reduction after misperception and re-phonologization UR Articulatory intent Acoustic pattern produced /pa ta/ [pa t ] [pa t ] If a scenario like this is responsible for every instance in which vowel reduction develops, then there is a clear extra-grammatical explanation for the lack of languages in which vowels reduce to [ ] in stressed syllables only. That is, if it is duration-based articulatory undershoot that leads to phonological vowel reduction, then reduction will never develop in stressed syllables only, because no speech community will ever have articulatory undershoot in stressed syllables only. below *VPLACE/F. The justification for the universally fixed ranking would likewise be the difference in functional grounding between these two constraints. See Steriade (2001) for a related proposal concerning functionally determined rankings among faithfulness constraints. 6

7 The crucial formal difference between this account and the positional-markedness account discussed above is that here, non-grounded constraints like *VPLACE/F are not formally excluded from the universal constraint set. The idea is that these undesirable constraints are harmless even if they do find their way into the universal constraint set, because the process of diachronic change through misperception will never lead speakers to rank them high enough to be active in the grammar of any language. 2.4 How general is the misperception/re-phonologization model of phonetic naturalness? The diachronic misperception and re-phonologization approach to the vowelreduction asymmetry outlined above is appealing; it relates the observed inactivity of a nongrounded phonological constraint such as *VPLACE/F directly to phonetic influences on language change, without explicitly needing to exclude the constraint from the universal constraint set. An important question remains, however: Can the misperception and rephonologization model account for all proposed cases of functional grounding in phonology? If so, then whenever a formally possible, but functionally unmotivated, constraint seems to be 'missing' from the constraint set, there should be a misperception and re-phonologization account of the constraint's phonological inactivity. However, positional augmentation constraints, which involve the addition of perceptually salient properties to phonologically strong positions, pose problems for this approach. The next two sections define ( 3) and exemplify ( 4) positional augmentation constraints and the functionally based restrictions on this constraint family; problems that these constraints raise for the diachronic misperception and re-phonologization model of phonetic influences in phonology are then discussed in Positional augmentation an overview The discussion in the previous section considers two formally similar constraints, *VPLACE/F and *VPLACE/F, of which the first is phonetically motivated and the second is not. The remainder of this paper focuses on a particular family of constraints that is likewise subject to a functionally based restriction; not all logically possible members of the family are attested. The family of constraints in question is the set of positional augmentation constraints. These are markedness constraints that are relativized to phonologically strong positions in other words, they are constraints that are responsible for enforcing phonological requirements that hold specifically of material in strong positions, rather than holding across-the-board of all relevant structures in a language (de Lacy 2001; Parker 2001; Smith 2000, 2002). Phonologically prominent, or 'strong,' positions are those that have particular phonetic or psycholinguistic salience (Steriade 1993; J. Beckman 1998). Examples of strong positions that have been discussed in the literature include the stressed syllable, the onset (or released consonant), the long vowel, the morphological root, and the initial syllable. These positions 7

8 are known for their characteristic ability to resist neutralization processes affecting other positions (Trubetzkoy 1939; Steriade 1993; Beckman 1998). However, there are also cases where strong positions are the exclusive targets of certain types of neutralization processes (de Lacy 2001; Parker 2001; Smith 2000, 2002). Examples of phonological requirements affecting strong positions are given in (5); the case studies discussed in the following section focus on the strong positions root (5d) and onset (5e). (5) Examples of requirements on strong positions (see Appendix for language examples) Strong position Phonological requirement a. Main-stress syllable Bimoraicity High tone Low tone High-sonority peak Onset Low-sonority onset b. Long vowel High-sonority peak c. Initial syllable Onset Low-sonority onset d. Root Stress e. Onset Supralaryngeal place Since these phenomena involve phonological requirements on strong positions, rather than the commonly observed resistance by strong positions to phonological requirements, they cannot be modeled with strong-position-specific faithfulness constraints (J. Beckman 1995, 1998; Casali 1996). Instead, these phenomena must be modeled with strong-positionspecific markedness constraints; in OT, phonological requirements on output forms 4 that 4. Another logically possible approach to these patterns is to propose weak-positionspecific faithfulness constraints, which if ranked high would protect contrasts in weak positions while allowing a general markedness constraint to affect strong positions only. However, this approach fails to capture the connection between phonological requirements that are specifically enforceable in strong positions and the addition of perceptual salience to those positions (see below). Furthermore, pursuing the weak-position faithfulness alternative would require us to abandon the insight that positional faithfulness constraints have a special relationship with strong positions, which are accessible to the grammar because of their increased salience (J. Beckman 1998). 8

9 override input featural specifications are enforced by the markedness constraints in the grammar. However, it is not just any markedness constraint that can be relativized to a strong position, because, empirically, it is not just any kind of phonological requirement that targets strong positions. As seen in the examples listed in (5), markedness constraints on strong positions always demand the presence of a perceptually salient property, which can be provisionally defined as a property that gives rise to a larger neural response than would an otherwise identical phonological form that lacks the property in question (see Smith 2002 for additional discussion). Other logically possible markedness constraints on strong positions that do not call for salient properties are not observed to be phonologically active. For example, HAVEPLACE/Onset is attested this is a markedness constraint, relativized to onsets, that requires a consonant to have a supralaryngeal Place feature, penalizing onset consonants with only a glottal Place specification (Parker 2001; see 4.2 below). However, constraints that simply ban typologically marked features from the strong position onset, such as *LABIAL/Onset, 'onset consonants are not [Labial],' are empirically unattested (see also Parker 2001, p. 362). The difference between the attested HAVEPLACE/Onset and the unattested *LABIAL/Onset is precisely one of perceptual salience. That is, a non-labial onset consonant would not consistently be more perceptually salient than a labial onset, but an onset consonant with some supralaryngeal Place feature would be more perceptually salient than a glottal onset consonant (Stevens 1971; Warner 1998). Because they demand the presence of perceptually salient properties in the strong positions that they target, markedness constraints on strong positions are also called positional augmentation constraints (Smith 2000, 2002; see also Zoll 1998 on positional effects involving 'augmentation of the input'). Crucially, degree of perceptual salience is a functional, extra-phonological characteristic. Therefore, the requirement that strong-position-specific markedness constraints must invoke perceptually salient properties is a functionally motivated one. This functionally based restriction on strong-position-specific markedness constraints complements other aspects of strong-position behavior given that strong positions are generally less susceptible to markedness requirements than weak positions, it is unsurprising that strong-position-specific markedness constraints should be highly restricted. Intuitively, what positional augmentation constraints do is to take a position with intrinsic salience on some dimension and give it additional perceptual salience, thereby 'making the strong stronger' (compare the harmonic alignment of prominence scales in Prince and Smolensky 1993, 5.1). Before turning to the question of whether this kind of functional restriction on possible constraints can be modeled with the diachronic misperception and rephonologization account described in 2.2-3, the following section presents case studies exemplifying two positional augmentation constraints and the phonological patterns that they are responsible for. 9

10 4. Positional augmentation case studies 4.1 A constraint calling for stress in roots Synchronic functional grounding One example of a phonological requirement specifically affecting a strong position is the requirement that roots bear stress. (On the status of the root as a strong position, see McCarthy and Prince 1995; J. Beckman 1995, 1998; Casali 1996; Alderete 1999, 2001). This requirement can be formally modeled with a positional augmentation constraint, HAVESTRESS/Root. (6) HAVESTRESS/Root Roots bear stress Because stress is correlated with perceptual salience, this constraint meets the functionally based criteria for a positional augmentation constraint. In some languages, roots are consistently stressed. In such cases, HAVESTRESS/Root is ranked above all other constraints that could determine stress placement, such as left- or right-edge Alignment constraints (McCarthy and Prince 1993), which would place stress on initial or final syllables respectively, or WEIGHT-TO-STRESS (Prince 1990), which would attract stress to heavy syllables. Examples of languages with consistent root stress include Diegueño (Langdon 1975, 1977), in (7), and Tahltan (Cook 1972; Alderete 1999; Alderete and Bob to appear), in (8). The Tahltan examples demonstrate that root stress is chosen even when this leaves a syllable containing a long vowel (8b,c), or a closed syllable (8f), unstressed. (7) Root stress in Diegueño Data from Langdon (1977, p ); roots are underlined a. ma t 'land' w b. t -x -m -k a n-p 'is tangled up' c. m-a ku xa p-c-m -ju 'Are you catching up with him?' 5 you-catch.up-same.subj-you-be 5. According to Langdon (1977, p. 239), the stem a+ku+xáp contains three prefixes. Since roots are monosyllabic (Langdon 1975, 1977), the root must be áp or xáp. 10

11 (8) Root stress in Tahltan Data from Alderete and Bob (to appear); roots are underlined a. u des- u t 'I whistled' b. e i -dli n 'We (dual) danced' < cv.cv -ccv c c. ude i -dle t 'We (dual) melted it' < cv.cv.cv -ccv c d. de -t o e 'soft' e. me e-k a he 'his/her fat' f. da #dah-se a 'Did you (pl) holler?' < cv #cvc-cv.cv g. me -det o j 'his/her pelts' Additional languages with consistent root stress include Chukchee (Krause 1979), Nancowry (Radhakrishnan 1981), and Mbabaram (Dixon 1991). Another pattern involving HAVESTRESS/Root is observed in Tuyuca (J. Barnes 1996). In this language, root stress is not surface-true; instead, it emerges as the default preference when all else is equal. Emergent root-stress patterns such as this confirm that the preference for root stress is integrated into the phonological grammar; it is a requirement that can interact with other processes in the phonology of a language. Tuyuca has lexical contrasts between stressed and unstressed roots and between stressed and unstressed suffixes, as shown in (9). (There are no prefixes in the language.) 11

12 (9) Stressed and unstressed roots and suffixes Roots (data from Barnes 1996) Suffixes Stressed ho a 'to write' -me na 'with' po a 'hair' -ma ke 'stuff' ho o 'to plant manioc' -di kv 'only' wai 'fish' -sotoa 'on top of' kape a 'eye' -ju 'beforehand' keero 'lightning bug' -wi (an evidential) - o (fem. sg. vb. sfx.) Unstressed hoo 'to submerge oneself' -a (an evidential) no a 'who' -i (an evidential) waka 'splinter' -je (change of focus) waso 'to change' -sa (thematic importance) There is exactly one stress per prosodic word in Tuyuca, so when roots and suffixes combine, not all morphemes necessarily surface faithful to their input stress specifications. As shown in (10), a stressed root (10a) always surfaces with stress (forms i, ii). However, not all roots are stressed in the language; combining an unstressed root (10b) with a stressed suffix (10c) leads to surface stress on the suffix (form iii). Crucially, however, if there is no lexically stressed morpheme, the root bears surface stress (form iv). (10) Root and suffix combinations; roots are underlined Roots stressed (Barnes 1996, p. 41) unstressed Suffixes a. /ho a/ 'to write' b. /waso/ 'to change' stressed c. /-ju / (ASPECT) i. ho aju iii. wasoju unstressed d. /-i/ (EVIDENTIAL) ii. ho ai iv. waso i The aspects of the Tuyuca pattern that attest to the ranking of HAVESTRESS/Root in the phonology of this language are first, the fact that lexically stressed affixes take 12

13 precedence over lexically unstressed roots (10, iii), and second, the fact that default stress is inserted into the root rather than into the affix (10, iv). Because word stress placement respects input stress location even at the cost of violating HAVESTRESS/Root appearing on a lexically stressed affix in preference to a lexically stressed root this shows that faithfulness constraints on the presence and affiliation of input stresses (11) must outrank HAVESTRESS/Root. (11) Stress faithfulness constraints dominate HAVESTRESS/Root (constraints from Alderete 1999, 2001) a. MAX-PROM A metrical prominence (=stress) in the input has an output correspondent; 'stress is not deleted' b. DEP-PROM A metrical prominence in the output has an input correspondent; 'stress is not inserted' c. NOSHIFT Corresponding prominences have corresponding sponsors and links; 'the location of a stress does not shift' (called NOFLOP-PROM in Alderete 1999, 2001) Specifically, either MAX-PROM or DEP-PROM (or both) must dominate HAVESTRESS/Root in order to rule out a candidate that deletes the input stress from the suffix and inserts a new one on the root (as in candidate (12a); the grave accent ( ) is used to indicate an output stress that is not a correspondent of the stress shown in the input). In addition, NOSHIFT must dominate HAVESTRESS/Root in order to rule out a candidate in which the input prominence surfaces in the output on a different sponsor, in order to appear within the root (as in candidate (12b)). With this ranking, the attested form (candidate (12c)), which maintains the input affix stress, is correctly chosen; HAVESTRESS/Root cannot force stress in a lexically unstressed root if a lexically stressed affix is present. 13

14 (12) Tuyuca ranking (I): Inserting/deleting or moving a stress is avoided { MAX-PROM or DEP-PROM }, NOSHIFT >> HAVESTRESS/Root /hoo + wi / submerge.oneself-ev 'he submerges himself' /hoo+wi / MAX-PROM or NOSHIFT HAVESTRESS/Root DEP-PROM a. hoo wi (newly inserted stress) *! b. hoo wi (shifted input stress) *! L c. hoowi (faithful input stress) * Although it is dominated, and frequently violated, HAVESTRESS/Root is nevertheless crucial for a complete account of the Tuyuca stress pattern. Only this constraint can force default stress insertion into roots when no morpheme in a word has lexical stress. The significance of a word with no lexically stressed morphemes is that MAX-PROM and DEP-PROM no longer penalize a candidate that has inserted a stress into the root. As noted above, all prosodic words in Tuyuca have exactly one stress. This pattern is enforced by the constraint CULMINATIVITY (which may in fact be an encapsulation of multiple interacting constraints), here given a characterization based on Alderete (1999). (13) CULMINATIVITY Every prosodic constituent has exactly one head (Alderete 1999) With CULMINATIVITY ranked above DEP-PROM, one stress must be inserted (14a); a DEP- PROM violation is unavoidable. On the other hand, there is no input stress to be deleted, so MAX-PROM is not violated by any candidate. In this situation, the effects of HAVESTRESS/Root emerge root stress (14b) is chosen over affix stress (14c) It is not an Alignment constraint (McCarthy and Prince 1993) forcing stress to the left edge of the word that favors root stress over affix stress, since default stress always appears at the right edge of the root (Barnes 1996). 14

15 (14) Tuyuca ranking (II): Default stress is inserted on roots CULMINATIVITY >> DEP-PROM; effects of HAVESTRESS/Root emerge /hoo + a/ submerge.oneself-ev 'I submerge myself' /hoo+a/ CULM DEP- MAX- HAVE DEP- PROM PROM STRESS/Rt PROM/Rt a. hooa *! * L b. hoo a * * c. hooa * *! It is important to note that, unlike the case of a stressed root plus a stressed affix, where the choice is which input stress to preserve and which to delete, default root-stress insertion cannot be driven by special faithfulness to roots. Stress insertion in the root actually violates root faithfulness, since an input property of the root is changed in the output. That is, the actual output (14b) performs worse on the root-specific faithfulness constraint DEP- PROM/Root than the candidate with affix stress (14c) does. Default root stress is not a logical necessity; languages like Thompson River Salish actively avoid default stress insertion into roots (Coelho 2002), showing the effects of DEP-PROM/Root. If the grammar contained no constraint favoring root stress that could be ranked above this root-specific faithfulness constraint, then DEP-PROM/Root, no matter how low-ranking, would always favor default stress on affixes. In summary, languages with a surface-true preference or an emergent preference for root stress, including Tahltan, Diegueño, and Tuyuca, provide evidence for a phonological requirement that roots bear stress. The constraint responsible for this requirement, HAVESTRESS/Root, is a positional augmentation constraint. This constraint targets the root, which is a strong position; bearing stress, which this constraint demands, makes the root more perceptually salient. 4.2 A constraint calling for supralaryngeal Place in onset consonants Another example of a phonological requirement that specifically affects a strong position is the requirement that onsets have a supralaryngeal Place specification. This pattern is found in Chamicuro, discussed in detail by Parker (1994, 2001), as well as in Tiriyó, Carib, and Macushi (Parker 2001, p. 362), and in Yatzachi Zapotec, where an onset [ ] is insufficiently salient to block vowel coalescence (Borroff 2003). The phonological requirement that onsets must have supralaryngeal Place is enforced by the following constraint, proposed by Parker (2001). 15

16 (15) HAVEPLACE/Onset Every onset segment in the output has a [supralaryngeal] Place specification (Parker 2001, p. 371) The crucial observation about Chamicuro is that glottal consonants are never onsets, although in coda position they contrast with each other, with other coda consonants, and with the lack of a coda. (16) Coda [h ] contrastive in Chamicuro (data from Parker 2001, p ) a. me sa 'sea lion' d. a tikana 'we' me sa 'party' ahtini 'path, trail' me sa 'table' uanasti 'I watch, look' b. it ehki 'it burns' e. sa pu 'lake' it e ki 'it is abundant' kahpu 'bone' c. me na 'woodpecker' netna 'how much?' jelna 'man, husband' me nu 'tongue' sjekput le 'pot-bellied' Parker (2001) provides supporting evidence from loanword adaptation showing that the absence of glottal onsets is a linguistically significant generalization. The Spanish fricative represented orthographically as <j> (which, according to Parker (2001, p. 373), is typically glottal [h] rather than dorsal [x] in the area of Peru where Chamicuro was spoken) is consistently realized as a coronal fricative in Chamicuro: [ ], or [ ] before round vowels. (17) Glottal onsets adapted in Chamicuro loanwords from Spanish (Parker 2001, p. 373) Spanish Chamicuro a. naranja na anha alan a 'orange' b. jabón ha on awona 'soap' c cojo koho ko o 'lame, crippled' The onset (or, perhaps, released consonant) is a strong position (Kingston 1985, 1990, to appear; Lombardi 1991; Padgett 1995; Steriade 1993, 1997). Crucially, as with the cases of mandatory root stress discussed in the preceding section, the prohibition against glottal onsets in Chamicuro cannot be modeled with special faithfulness for this strong 16

17 position. The only workable analysis involves the markedness constraint against onset glottals, HAVEPLACE/Onset, introduced in (15) above (see Parker 2001 for additional argumentation that such a constraint is needed). HAVEPLACE/Onset is a markedness constraint specific to a phonologically strong position, and it does meet the functional criteria for a positional augmentation constraint, because consonants with supralaryngeal Place features are more perceptually salient than glottal consonants (Stevens 1971; Warner 1998). With this constraint ranked appropriately, it is possible to model the Chamicuro pattern, allowing glottal codas but no glottal onsets. (The constraints and analysis presented here are based on those of Parker (2001), with minor changes.) First, the fact that glottal codas surface intact shows that any markedness constraint penalizing glottal consonants in general, such as *LARYNGEAL (18c), must be outranked by the faithfulness constraints against segment deletion (18a) and changes in Place features (18b). (18) Constraints potentially affecting glottal codas a. MAX-SEG Input segments have output correspondents; 'segments are not deleted' (McCarthy and Prince 1995) b. IDENT[Place] Corresponding segments agree in Place features; 'Place features are not changed' (McCarthy and Prince 1995) c. *LARYNGEAL Output segments have no laryngeal feature specifications (including [spread glottis], [constricted glottis]) (Lombardi 1999, 2001) The ranking among these constraints is illustrated in (19), where a glottal coda appears in the surface form (19a) instead of being deleted (19b) or altered (19c). (19) Chamicuro ranking (I): Glottal codas are permitted (after Parker 2001, (12)) MAX-SEG, IDENT[Place] >> *LARYNGEAL /nihpa/ 'louse' /nihpa/ MAX-SEG IDENT[Place] *LARYNGEAL L a. nih.pa * b. ni_.pa *! c. nis.pa *! 17

18 Unlike glottal codas, glottal onsets are subject to the positional constraint HAVEPLACE/Onset. Therefore, this constraint must dominate at least one faithfulness constraint that would otherwise protect glottal consonants in onset position. Following Parker (2001), the dominated constraint in (20) is taken to be IDENT[Place], based on the 7 evidence from loanword adaptation shown in (16) above. (20) Chamicuro ranking (II): Glottal onsets are avoided { HAVEPLACE/Onset, MAX-SEG } >> IDENT[Place] >> *LARYNGEAL /nihapa/ hypothetical form after Parker (2001, (15)) /nihapa/ HAVEPLACE/Onset MAX-SEG IDENT[Place] *LAR a. ni.ha.pa *! * b. ni._a.pa *! L c. ni. a.pa * The inclusion of HAVEPLACE/Onset in the ranking does not affect the results for glottal codas shown in (19), because no coda will ever violate an onset-specific constraint. In summary, Chamicuro and the other languages that prohibit glottal consonants specifically in onset position provide evidence for another strong-position-specific phonological requirement, one that forces onsets to have a supralaryngeal Place specification. The markedness constraint that enforces this requirement is Parker's (2001) HAVEPLACE/Onset. Like HAVESTRESS/Root in 4.1, HAVEPLACE/Onset invokes a perceptually salient property, thus respecting the functionally based limitation that restricts the inventory of markedness constraints on strong positions. The following section now demonstrates that this functional restriction on the constraints in the phonological grammar is not one that can be adequately modeled by the diachronic misperception and rephonologization account of phonetic influence on phonology. 7. The phonological repair chosen by a language during loanword adaptation is not necessarily the same as the default repair in the native phonological system (Yip 2002; Smith 2004). Here, however, it is the fact that a repair occurs at all that is important this confirms that glottals are actively avoided. Moreover, in the case of Chamicuro, there is no evidence that the feature-change repair in loanword adaptation does differ from the default repair strategy for glottals. Therefore, the tableau in (19) below follows Parker's (2001) assumption that MAX-SEG outranks IDENT[Place] in the core phonology of Chamicuro, making featural change the preferred repair for glottal onsets. 18

19 5. Accounting for the functional restriction on positional augmentation constraints In 2, two alternative accounts were presented of the tendency for phonological patterns to be phonetically motivated. The grammar-internal functional grounding model ( 2.1) holds that the contents of the formal phonological grammar are directly restricted by functional factors. The diachronic misperception and re-phonologization model ( 2.2) counters that phonological patterns tend to be phonetically plausible, not because the formal grammar is explicitly limited by functional factors, but because phonological patterns evolve when learners misperceive aspects of the acoustic signal and incorporate the misperceived structures into their grammatical systems. The question posed at the end of 2 was this: Is the misperception and rephonologization model able to account for all cases where a formally possible, but functionally unmotivated, constraint seems to be 'missing' from the constraint set? One example of a functionally based restriction on phonological constraints is the condition on positional augmentation constraints described in 3; namely, markedness constraints on strong positions that do not enhance perceptual salience are unattested. Can this gap in the inventory of phonologically active constraints be explained via patterns of diachronic misperception and re-phonologization? This section presents a number of difficulties raised by positional augmentation phenomena for the misperception and re-phonologization account, considering problems related both to specific characteristics of the 4 case studies and also to more general characteristics of positional augmentation phenomena ( 5.1-2). A grammar-internal functional grounding account of the functionally based restrictions on positional augmentation is then developed in 5.3. Finally, 5.4 shows that it is not possible to expand the diachronic model to handle positional augmentation without introducing a component into the model that is essentially indistinguishable from synchronic grammar-internal functional grounding itself. 5.1 Misperception-based accounts and the nature of the phonological patterns The first problem posed by positional augmentation constraints for the misperception and re-phonologization model is that the kinds of phonological requirements being enforced in strong positions cannot all have come about through misperception in the first place. As a consequence, it is difficult to relate the functionally based characteristics of this class of constraints directly to the nature of the diachronic misperception and re-phonologization process. For example, the languages discussed in 4.1 involve mandatory root stress. It is difficult to see how this situation could have arisen through misperception from a language that did not have root stress. The problem is perhaps most pressing for a case like Tuyuca, where root stress is an emergent effect of a relatively low-ranking constraint rather than a surfacetrue generalization. But even for the surface-true cases like Diegueño and Tahltan, it is unlikely that some acoustic factor could cause listeners to misperceive stress on a root that was not originally stressed. Plauché et al. (1997) and Chang et al. (2001), replicating a 19

20 known visual-perception effect in the domain of auditory perception, show that perceptual confusion is asymmetric with respect to perceptual salience. If X and Y differ primarily in that X has some salient cue that Y lacks, listeners may misperceive salient X as Y (i.e., they may fail to perceive the salient cue). However, listeners do not 'imagine' the presence of a nonexistent salient cue and misperceive Y as salient X. Since positional augmentation constraints always act to add perceptual salience to a strong position, their effects are therefore incompatible with a misperception origin. It is possible that particular instances of positional augmentation might come about through misperception and re-phonologization if a perceptually salient cue that is present in the signal for some other reason is later reinterpreted as being phonologically associated with a strong position. Positional augmentation effects calling for low-sonority onset consonants in initial syllables and stressed syllables (de Lacy 2001; Smith 2000, 2002) might be amenable to this kind of explanation, since phonetic articulatory strengthening effects for stressedsyllable onsets (M. Beckman et al. 1992) and domain-initial consonants (Keating et al. 2004) have been experimentally observed. However, this approach cannot provide a general explanation for the functional aspects of positional augmentation taken as a whole. The problem is that the prominent property enforced by a positional augmentation constraint need have no intrinsic connection to the phonetic characteristics of the position being augmented as with roots and stress, or onsets and supralaryngeal place. See 5.2 below for additional discussion of this 'abstract' character of the functional grounding in positional augmentation constraints. Chamicuro and the other languages listed in 4.2, with a requirement that onset consonants cannot be glottal, pose a different kind of problem for a diachronically based account. Specifically, what is known about the diachronic development of glottal consonants is not likely to lead to a Chamicuro-type pattern. Diachronic glottal 'fortition' or 'buccalization' change to a different Place feature is sporadically attested for [h] (Blevins, to appear), but essentially unattested for [ ] (Trask 1995), leading Trask to describe glottal Place as "a vast sink from which no segment ever returns." Crucially, the onsetspecific requirement for non-glottal Place in Chamicuro treats [h] and [ ] as a class, so no [h]-specific diachronic misperception account can adequately capture the general Chamicuro pattern. An alternative misperception scenario, involving diachronic loss of glottal onsets through failure to perceive them at all, would indeed give rise to a language with no glottal onsets. However, this scenario should entail the loss of glottal codas as well, since codas are less perceptible than onsets, a fact that is often emphasized in misperception and rephonologization accounts of coda neutralization. Instead, glottal codas persist in the Chamicuro-type languages. These facts about the nature of the positional augmentation case studies from 4 show that it is difficult to develop a convincing misperception and re-phonologization account of the phonological patterns involved because the observed phonological patterns 20

21 do not accord well with what is known about speech perception and paths of diachronic change. 5.2 Misperception-based accounts and 'abstract' functional grounding An additional problem that arises in trying to explain the functional restriction on positional augmentation constraints in a diachronic misperception and re-phonologization model has to do with the nature of the functionally based restriction itself. This restriction simply states that markedness constraints on prominent positions are legitimate, phonologically active constraints only if they demand the presence of some perceptually salient property in the strong position that they target. Crucially, it is not the case that the perceptually salient property must be related to some intrinsic characteristic of the strong position for the constraint to be acceptable. This can be seen in the positional augmentation effects discussed in 4: roots have no intrinsic connection to stress, and onsets have no intrinsic connection to supralaryngeal Place features. Additional cases listed in (5) in which there is no particular phonetic relationship between the strong position and the perceptually salient property include the requirement that long vowels have high-sonority peaks, and the requirement that stressed syllables be heavy (one of the best-known examples of a strongposition-specific requirement). In this sense, the functionally based restriction on positional augmentation constraints can be said to be 'abstract'; it is clearly functionally motivated, since it refers to perceptual salience, but it is not tied to the specific characteristics of particular constraints or particular strong positions. It is because the functional restriction on positional augmentation constraints has this degree of formal abstraction that it is difficult to reduce it to a mere byproduct of the way that particular phonological patterns might have arisen diachronically. An analogous argument against a misperception and re-phonologization account has been raised by Steriade (2001, p. 233) for the phonology of nasal place assimilation. Steriade argues against trying to relate the typological observation that nasals are the most likely consonants to undergo place assimilation directly to the results of perceptualconfusion experiments (Hura et al. 1992) showing that place features are more easily confused in nasals than in stops or fricatives. This is because in Hura et al.'s experimental results, nasals were most often misperceived, not as assimilated nasals, but as alveolar nasals. Nasals do tend to be misperceived, but not primarily in assimilatory ways. Therefore, bare misperception is unlikely to be the root of assimilation.... This may be an example of knowledge of perceptibility used as a phonological tool. Steriade (2001, p. 233) [emphasis added] Both the nasal-assimilation case discussed by Steriade (2001) and the positional augmentation effects discussed here show that functionally based considerations sometimes influence phonological patterns in ways that differ from the direct effects that the same functionally based considerations would have on a specific instance of speech perception. In 21