Locality in exceptions and derived environments in vowel harmony

The regular realm of vowel harmony in Assamese consists of right-to-left regressive [Atr] harmony. In contrast with this regular pattern of vowel harmony, the exceptional Assamese processes dealt with in this paper are symptomatic of the behavior of a pair of morphemes that trigger additional processes not seen elsewhere in the language. This pair of morphemes allows raising of the otherwise opaque vowel // and fronting/backing of // depending on the [Back] quality of a mid vowel adjacent to //. Raising is strictly local in the presence of preceding high and low vowels, but there is also another pattern which shows backness assimilation to a previous vowel if there are mid vowels preceding the // of the input. This exceptional raising occurs to allow [Atr] harmony to spread regressively by changing the [−Atr] low vowel into a [+Atr] mid vowel. I analyse these cases within Optimality Theory (Prince and Smolensky 1993/2004) and show that these exceptional occurrences are morpheme-specific. It is also shown that these exceptional occurrences lend themselves to an account based on indexation of markedness constraints (Flack 2007; Ota 2004; Pater 2000, 2006, 2009). Consequently, the Assamese examples show that indexed markedness constraints are able to deal with an exceptional alternation where a low vowel undergoes harmony locally. This article also shows that an emergence of the unmarked analysis is required to account for the low back vowel that alternates with a front vowel if there is a preceding front vowel. This article goes beyond the problems encountered in Assamese, and claims that there is no need to invoke locality in exceptional blocking in vowel harmony, as both exceptional and non-exceptional blocking in vowel harmony are always local and bounded. The goal of this paper is to shed light on exceptional and emergent processes, arguing that they are always local and governed by strong universal properties of grammar.


Introduction
Assamese has eight surface vowels, and it exhibits regressive [ATR] harmony: the high vowels /i/ and /u/ systematically trigger [+ATR] in all the extant [−ATR] high and mid vowels in the language (i.e. / /, / / and / /) except for in the [−ATR] low vowel / /. In this language, exceptional patterns emerge when a root morpheme containing the vowel / / is harmonized to either [e] or [o] by the presence of the suffixes /-ij / and /-uw /. In this paper I present the intricacies of exceptional triggering of [ATR] vowel harmony by the two morphemes /-ij / and /-uw / in Assamese, and I also present an analysis of this exceptional pattern within the theory of lexical indexation in Optimality Theory.
Section 1 presents a general background to Assamese vowel harmony, and delineates the nature of the sequential markedness constraints that will be employed to analyze regressive vowel harmony. Section 2 deals with data and problems relating to exceptional triggering of harmony by the affixes /-ij / and /-uw / in Assamese. Its subsections illustrate the theoretical precept of locality in exceptionality. Section 2 also discusses the typological implications of the patterns of exceptionality thus far encountered in languages. Section 3 discusses the untenability of alternative approaches to exceptions in vowel harmony. Section 4 discusses the / /-raising data, and offers an emergence of the unmarked solution to the problem encountered in these data. The subsections in 4 discuss alternative theoretical approaches, and the paper ends with a conclusion in Sect. 5.
The main thrust of this paper is to discuss exceptions in vowel harmony in order to broaden our understanding of exceptional processes, with reference to data from a hitherto under-described language. In order to do so, we must first understand the regular vowel harmony pattern of Assamese. To this extent, Sect. 1.1 presents a descriptive backdrop that acquaints the reader with the vowel system and the processes of vowel harmony in the language.

Introduction to Assamese vowel harmony
Assamese has the eight surface vowels [i, e, , , , o, , u], as shown in Table 1.
The two high vowels [i] and [u] are pronounced with an advanced tongue root (indicated in phonological representations by the feature [+ATR]) as are the mid vowels The featural distribution system above shows that there is a contrast in terms of the feature [ATR], but the contrast is less than perfect: There is a gap reflected in the absence of a [−ATR] high front vowel. A very common gap in seven vowel systems like Yoruba is the absence of any [−ATR] high vowels. Assamese is therefore atypical for seven or nine vowel systems in ATR harmony languages (cf. Stewart 1967Stewart , 1971 where gaps are encountered with respect to the low vowel / / and [−ATR] high vowels.
The existence of vowel harmony as a regular synchronic process is evident from the distribution of vowels in the language. A thorough investigation has revealed that even in the underived part of the vocabulary, the underlying [−ATR] vowels / /, / /, and / / are not attested in conjunction with the [+ATR +High] vowels /i/ and /u/ to their right. Assamese imposes severe restrictions on the occurrence of the features [−ATR] and [+ATR] consecutively, in derived as well as underived domains. Although the absence of */ . . . i/, */ . . . u/, */ . . . i/, */ . . . u/, */ . . . i/, and */ . . . u/ sequences demonstrates that / /, / /, and / / do not surface in such sequences, it is the harmonic alternations in morphologically complex words that demonstrate that these sounds change.

3) [−ATR] vowels in the absence of [+ATR] vowels
Word Gloss a. b s n 'powdered pulse' b. g r m 'hot' c. b l g 'different' d. p x k 'week' e. z k r 'shake' These distributional characteristics lead us to surmise that the underlying vowel inventory of Assamese consists of six vowels, /i/, /u/, / /, / /, / / and / /, in its stem phonology.  (4) and (5) show that whenever /i/ and /u/ are in a suffix, they consistently impose changes on the [ATR] specifications of the vowels / /, / / and / / in the stem/root. 1 Note the overabundance of suffixes bearing the vowel /i/. It is a fact of the language that /i/ suffixes far outnumber suffixes with the trigger /u/, but whenever /u/ appears it unfailingly triggers vowel harmony. 2 Some combinations, like / / + / / were found not to exist in the native vocabulary. The examples below show how the prefixal vowels / -/ and / -/ change their feature value to [+ATR] in an environment where there is a /i/ or a /u/ on the right side of the morphological word: The examples above show how the prefixal vowels / -/ and / -/ change their feature values to [+ATR] in an environment where there is /i/ or /u/ on the right side of the morphological word. The data provided thus far provide sufficient evidence that Assamese harmony is regressive, and is always triggered by an immediately following /i/ or /u/, regardless of the morphological affiliation of the trigger or the target. This  [u] is not. Assamese vowel harmony is typically word-based, excluding compounded words and larger morphosyntactic domains. Vowel harmony in Assamese is a leftward process, and there are no morphologically significant positions that either trigger or target it. This paper, however, concerns itself to a large extent with certain exceptionalities that stray from the regular harmony pattern of Assamese. Of utmost importance in understanding these exceptional processes is the fact that the generally opaque vowel / / in words like /k p hi/ 'made of cotton' and /g z li/ 'sprout' (more examples in Sect. 1.3) is raised as a result of the presence of the vowels /i/ and /u/ contributed by the morphemes /-ij / and /-uw /. / /-raising does not occur when / / is not adjacent to the triggering vowel. It is argued in Sect. 2.1 that /-ij / and /-uw / are exceptional triggers, and need to be analyzed with a lexical indexation approach, in which a convention of locality has to be followed. In Sects. 2.2 and 2.3 we will discuss the typological implications of this pattern to show that exceptional harmony is the only environment that requires a locality convention. The entirety of Sect. 4 discusses the pattern where a front harmony pattern emerges after / / raising in words like / l h/ + /uw / → [elehuw ] 'lazy'. I argue that this phonologically derived environment has to be analyzed with the help of a highly ranked positional markedness constraint and a LICENSE constraint leading to the emergence of the unmarked pattern. vacuously satisfy the constraint because they do not provide the right context for its application. These sequential markedness constraints can also be seen as sub-components of the Agree family of constraints (Bakovic 2000;Beckman 1998;Butska 1998;Lombardi 1996Lombardi , 1999 The tableau in (10) shows that, in the absence of any reliable morphological or prosodic motivation, AGREE by itself does not lead to regressive or progressive directionality. In the persistently regressive harmony of Assamese there is no substantive positional relevance of the trigger, therefore other theoretical devices like positional faithfulness and local conjunction also prove to be redundant. As such, in the proposed analysis, I proceed to disentangle the behavior of AGREE from its inherently asymmetric nature, and modify it into a more specific constraint which identifies a specific marked sequence of features. In the tableau above, AGREE[F] assigns a violation mark to the desired output candidate (10b) . This is only a partial analysis of harmony in Assamese, as the full harmony facts have not been laid out as of yet.
Once all the data are in place, we shall see that a ranking between *[−ATR] [+ATR] and the undominated constraints is required because of blocking by / / (see (23)).
In the absence of a harmony trigger to the right, only [−ATR] mid vowels can occur. This can be accounted for by the articulatorily grounded constraint *[−High +ATR]. Following Archangeli and Pulleyblank's (1994) grounding conditions 5 , I propose the feature co-occurrence constraint defined below, which becomes relevant when mid vowels are present in the input.
The tableau in (23) below shows how IDENT[Low] is responsible for blocking harmony: (23) Input /k p hi/:/ / remains unaltered in the presence of a following trigger The opacity of / / to the harmony process is accounted for by high-ranked IDENT[Low] and *[+ATR +Low]. These constraints are ranked higher than the harmony driving constraint *[−ATR][+ATR]. Therefore, the candidate in (23a), which does not undergo any / / alteration, is the winning candidate.

Regular harmony triggered by /-ij / and /-uw /
Thus far we have seen that the spectrum of regularity in the process of vowel harmony in Assamese encompasses harmonic effects of the following nature: The basic harmony pattern is that of regressive The examples above show that there is ample evidence that the adjectival suffixes /-ij / and /-uw / participate in regular [ATR] harmony when the preceding [−ATR] vowels are / /, / /, and / /. With this background we can now engage in the description and analysis of some of the vagaries (i.e. exceptional patterns) of vowel harmony in Assamese. In Sect. 2, we begin with monosyllabic roots containing / /, followed by bisyllabic roots with two consecutive / /'s. In Sect. 2.1 we present an analysis based on lexical exceptions and subsequently in Sect. 2.3 we present a broad overview of the typology of exceptions in vowel harmony.

/ /-raising: local exceptional triggering
The morphemes /-ij / and /-uw / trigger [ATR] harmony in accordance with the harmony characteristics outlined in Sect. 1.4. Section 1.4 however, abstains from demonstrating any phonological opacity of / / in the presence of /-ij / and /-uw /, precisely because this is the environment in which / / loses its static quality as far as [ATR] harmony is concerned. Under normal circumstances, the vowel / /, whether in the prefix, suffix or stem position, invariably behaves phonologically opaquely. If in certain circumstances / / appears to contravene this normal behavior and is able to undergo vowel harmony, then that behavior has to be characterized as exceptional and the characteristics of such exceptional participation in harmony need to be examined in detail. The [−ATR +Low] vowel / / in Assamese undergoes raising to the [+ATR] vowels [e] and [o] in order to participate in [ATR] harmony, but only when it is in the environment of the previously mentioned triggering morphemes /-ij / and /-uw /. In this section we will only deal with those environments where / / alternates with [o]. This pattern will be elaborated in Sect. 4, where / / sometimes undergoes height har-mony and fronting to alternate with [e]. The data below show that / /-raising alone appears in words of the following configuration: In words containing the vowel sequences above, the locality of the process always results in words where only the / / adjacent to the triggering vowel undergoes raising to [o], resulting in words of the type [(C)VCoCij ]. / /-raising occurs when the two affixes /-ij / and /-uw / trigger harmony in morphemes containing / /. 8 As a result of this raising, / / raises to [o] under exceptional morphological triggering in these environments, but remains inalterable in the regular phonology. 9 In monosyllabic stems, / / always adapts itself to [o] when followed by /-ij / or /-uw /.
The data below show that / /-raising is restricted to the vowel adjacent to the triggering morpheme and there is no / /-raising when / / is not adjacent to the triggering vowel: (28) / / does not change when it is not adjacent to the triggering vowel Root The examples (28a-c) have the segmental composition /C C . . . / and harmony triggered by /-ij / only affects the immediately preceding [−ATR] vowel / /, but the non-adjacent / / does not undergo harmony. However, this is not different from the behavior of similar sequences when harmony is triggered by suffixes other than /-ij / and /-uw /, as they would all produce the same result. The local triggering behavior of /-ij / and /-uw / is exemplified very clearly by the examples in (28d-e). In these cases, there are two instances of / /, but only the vowel adjacent to the triggering vowel undergoes harmony. See the illustrations below for a more explicit illustration: / /-raising triggered by /-ij / and /-uw / 11 violates IDENT[Low], but IDENT[Low] violations are as minimal as possible, because / /-raising is restricted to the smallest possible domain. The participation of only two morphemes /-ij / and /-uw / in triggering exceptional realization of harmony can be characterized as morphologically induced harmony, which is obtained at the cost of flouting the highly ranked phonological constraint IDENT[Low] that prohibits any alteration of the [+Low] vowel / /. This violation leads to the harmonizing behavior of the normally opaque vowel / / in such a way that it changes to a vowel which is already present in the surface phonetic inventory. Exceptional triggering of the type discussed in this paper cannot be deemed to be the same as dominance (Aoki 1966;Chomsky and Halle 1968;Hall et al. 1974;Anderson 1980;Hall and Hall 1980;Dimmendaal 1983) in vowel harmony or other kinds of exceptionalities recorded in the literature. In Assamese, there are no instances of exceptional root or suffix morphemes which undergo harmony under special circumstances, nor are there cases where morphemes do not undergo harmony because they are opaque to the spreading process. The Assamese data are unique cases of exceptional triggers. However, they are only unique as far as exceptionality in vowel harmony is concerned. Such cases of local exceptionality are found in other morpheme-specific phonology as well. Pater (2006Pater ( , 2009 shows that most of the problems tackled in morpheme-specific constraint ranking or cophonology, whose proponents include Anttila (1997Anttila ( , 2002, Inkelas (1998), Inkelas and Zoll (2005), Orgun (1996Orgun ( , 1998Orgun ( , 1999, and Orgun and Inkelas (2002), as well as faithfulnessonly constraint indexation (Fukuzawa 1999;Mester 1999, 2001;Kraska-Szlenk 1997 theories can be analyzed in terms of constraint indexation of both markedness and faithfulness constraints. (See Sect. 3 for a critique of these approaches.) At the same time, however, the fact that exceptional triggering or blocking by morphemes is never an unbounded phenomenon is only predicted by lexically indexed constraints. Constraint indexation is of special relevance in this paper because the predicted 'local' behavior of morphemically indexed constraints is borne out in the exceptional data of Assamese. In the constraint indexation approach, morphemes that trigger a process are indexed for a lexically specific faithfulness or markedness constraint. It is assumed that these indexed constraints are cloned from already existing constraints, which are ranked lower in the hierarchy. In recent work, indexation of markedness constraints has also been shown to be operative in Dinka (Flack 2007) where an exceptional prosodic template imposes a fixed number of morae on vowels. Gouskova (2007) employs lexical indexation of markedness constraints to analyze reduplicative affixes in Tonkawa which vary in their syllabic shape following the requirements of non-templatic constraints. Ota (2004) has also shown that indexed markedness constraints could be required to analyze native and loan words in the Japanese lexicon, if certain aspects of learnability are taken into consideration.
We can consider the analysis for Finnish (Pater 2007(Pater , 2009), a language in which the stem-final low vowel /-a/ either deletes or is raised to /o/ under the influence of a following /-i-/. Examples like /tavara+i+ssa/ → [tavaroissa] 'thing' (PLURAL INESSIVE) and /itara+i+ssa/ → [itaroissa] 'stingy' (PLURAL INESSIVE) in Finnish show the locality of alternations in certain exceptional forms, given the environment in which /-a/ occurs. Pater addresses the problem of exceptionality in Finnish by introducing a locality convention which captures the locality encountered in Finnish.
(30) *[ai] L Assign a violation mark to any instance of *[ai] that contains a phonological exponent of a morpheme specified as L.
The definition of the constraint above shows that the constraint is violated only when the specified string *[ai] occurs in the output. Without any further elaboration, I will simply repeat the (partial) tableau from Pater in order to provide an illustration: Although discussion of indexation of faithfulness constraints is not within the scope of this paper, it is tacitly assumed that faithfulness indexation is involved in other exceptional phenomena which do not involve triggering. In exceptional trigger-ing, locality effects will not be captured in a faithfulness-only constraint indexation approach.

Constraint indexation and morphologically conditioned exceptions in phonology
Given the description of the pattern of alternation and / /-raising in Sect. 2, I will now discuss the locality requirements which can be adequately addressed by the lexical indexation of the relevant constraints. 'Locality', or the application of a phonological process to a certain smallest possible domain, is of special relevance in this paper. Locality falls out from the locus of violation function, which has been implemented in recent OT grammars (Lubowicz 2005;McCarthy 2003aMcCarthy , 2003bMcCarthy , 2005McCarthy and Wolf 2005;Riggle and Wilson 2004) by employing the notion of position of violation of a constraint. In the context of markedness constraints, this approach locates the specific position where a constraint will be violated. To apply a markedness constraint to a string, the locus of violation function of a markedness constraint will find the set of points where the constraint is violated. (33) Loc*[ai] L ≡ Return each instance of /i/ that is immediately preceded by / /, where /i/ is a phonological exponent of a morpheme specified as L.
The domain of relevance for the 'locus of violation' of indexed constraints is restricted to co-indexed input-output pairs, where each constraint violation is assessed with regard to a specified location in the input. With regard to indexed markedness constraints this means that not only is the markedness constraint constrained to apply in the indexed part of the candidate, but also that the number of violations that the constraint can incur is also limited. The set of points that the locus of violation function for Loc *[ai] L identifies for the candidate /tavara-i L -ssa/ is limited to the indexed morpheme,and this limitation in the locus of violation defines the locality convention of exceptional occurrences. In Assamese, the two morphemes /-ij / and /-uw / exceptionally trigger harmony in the otherwise opaque vowel / /. This kind of triggering behavior is exceptional, as it is confined only to these two morphemes. But it is also systematic: / / systematically changes only when it is adjacent to the harmony triggering morpheme; i.e. if / / does not occur in immediate proximity of the triggering vowel, it does not harmonize (see diagram (29)). The local effect schematically represented in (29)  . This analysis has been presented in Mahanta (2008) and also mentioned in Pater (2009), but I now move on to discuss the exceptional typology of vowel harmony and the place of exceptional triggering in the typology.

Revisiting locality: towards a typology of exceptions in vowel harmony
Locality is a fundamental insight of generative grammar. Phonological processes are known to typically operate on adjacent elements, which in turn aids the learnability of such processes. I propose that strict locality in vowel harmony in the manner argued for in the phonological literature is only to be considered an attribute of exceptional triggering. Exceptional blocking in vowel harmony does not need any additional theoretical tool of locality. In effect, both exceptional and non-exceptional blockers behave in precisely the same manner (Bakovic 2000). In both types, the blocking vowel does not allow the harmonic feature value to affect any more vowels in the direction of spreading. Exceptional triggering, on the other hand is to be understood as noniterative harmony and it is mainly a matter of triggering an exceptional pattern on an adjacent syllable in a string. Note that the locality principle (McCarthy and Prince 1986) espoused for prosodic morphology is as follows: (39) A rule may fix on one specified element and examine a structurally adjacent element and no other.
McCarthy and Prince (1986) consider locality of adjacent structural elements to be a 'two elements only' configuration. Locality in these terms is seen as a process which can consider only two grammatical elements; in other words, 'grammars can only count up to two'. McCarthy and Prince (1986) argue that rules count moras (µ), syllables (σ), or feet (F) but never segments. McCarthy and Prince also reject the idea that languages count only the number of segments in a form. Templates can only count prosodic elements, and any approach which only counts segments is bound to fail. Thus McCarthy and Prince (1986) refute the templatic theory of reduplication and argue for a locality which is prosodically defined. This sort of categorical locality may also be necessary to account for exceptions in vowel harmony. As seen in (34), the locality of exceptional triggering considers only the vowel in an adjacent syllable in a string. It is important to see this distinction because exceptional triggering in vowel harmony is essentially a non-iterative process (unlike regular processes of vowel harmony) and the units of the process are confined to two adjacent syllables only.

Locality and exceptional blockers
Exceptional blockers in vowel harmony are mostly vocalic elements 12 which exceptionally stem the tide of unbounded vowel harmony. Such blockers have traditionally been called opaque vowels in vowel harmony. An opaque vowel is a disharmonic vowel, and the reason for the disharmonicity of the opaque vowel can be either phonological or morphological. Morphologically governed exceptional blockers or opaque vocalic segments do not behave any differently from opaque segments which are phonologically governed. To analyze the striking similarity of opaque vowels which are both phonologically driven as well as morphologically governed, we consult Bakovic (2000) where this is explicitly discussed in the context of Turkish vowel harmony. Bakovic also reliably demonstrates that in the presence of both phonologically predictable opaque vowels and morphologically induced exceptionally opaque vowels, an affix vowel uniformly agrees with the opaque vowel if it is adjacent to it. This shows that in vowel harmony, blocking by vowels needs no extra requirement of locality. Locality in phonological blocking is ubiquitous in vowel harmony systems (as also in Assamese; see Sect. 1.3 and the examples and tableaus therein) and therefore blocking is essentially the same in both exceptional and regular vowel harmony.

Non-existence of transparent vowels in exceptional patterns
The purpose of the previous section was to demonstrate, with the aid of opaque vowels which are either phonologically determined or morphologically constrained, that exceptional blocking shows the same requirements of locality under both conditions (contra Finley 2010). Given this behavior and its locality in exceptional triggering situations, a locality convention in the analysis of exceptions must restrict the absence of morphologically idiosyncratic transparent vowels in vowel harmony languages. Socalled transparent vowels cannot bear one of the two values of the harmonic feature, but these are different from opaque vowels in that they seem to permit the incompatible value to pass through them, allowing agreement between the vowels on either side, while the transparent vowel is itself incompatible with the feature values of the vowels on both sides.
The obvious explanation for the absence of morphologically idiosyncratic transparent vowels is that vowel transparency establishes a non-local relation, and therefore transparent vowels are not a choice which can be exploited by exceptional morphemes. Therefore, the typology of exceptional occurrences in vowel harmony excludes vowel transparency by incorporating a strict locality criterion between trigger and target in exceptions in vowel harmony. A constraint which typically operates to execute locality in exceptional vowel harmony is the following: Locality in exceptional vowel harmony: Only adjacent elements in a string (defined as syllable, mora, foot) can bear the phonological exponence of morphological elements.
This locality criterion excludes the presence of transparent vowels in exceptional patterns of vowel harmony, but it includes exceptional triggering.

A typological sketch of exceptions in vowel harmony
The typology of exceptions in vowel harmony needs to be stringently based on the locality criterion. While exceptional triggers will always trigger non-iterative harmony, exceptional blockers will not allow the harmonic feature value to percolate. In this context the typology of exceptions in vowel harmony presented by Finley (2010) becomes relevant. Finley's typology considers all exceptional patterns in vowel harmony languages to be morpheme-bound. I propose that there needs to be a more fine-grained distinction between the different types of exceptionalities regarding their respective locality conditions. Finley's typology divides the exceptional patterns in vowel harmony into exceptional undergoers, exceptional non-undergoers, and exceptional triggers. Finley seeks to isolate exceptional triggers from exceptional nontriggers and exceptional undergoers from exceptional non-undergoers. The exceptionally undergoing vowel harmony morpheme in Korean (which is otherwise not a vowel harmony language) allows for non-iterative vowel harmony. While Finley's typology can separate the behavior of exceptional non-undergoers from exceptional undergoers, this typology fails to predict and account for the non-existence of exceptionally transparent vowels which are also exceptional nonundergoers. Exceptionally transparent vowels are not only absent crosslinguistically, but they are also predicted to be non-existent by the very nature of locality of exceptions. Exceptionally transparent vowels which would establish a non-local relation between the trigger and target are categorically prevented from being available as one of the choices of exceptionality in vowel harmony languages. On the other hand, exceptional non-triggers, although predicted by Finley to be non-existent, are predicted by the theory of exceptionality in vowel harmony, and a high-ranked indexed markedness constraint would be able to account for it, if such a pattern is attested.
This excursus on the typology of exceptions in vowel harmony shows us that there may be only two types of exceptionality in vowel harmony, and while one type is subject to a criterion of locality, the other does not need any additional requirements.
(i) Exceptional harmony. This is demonstrated by the non-iterative harmony in exceptions in Assamese vowel harmony and Korean harmony. Assamese exceptional triggering and Korean exceptional undergoers are both instances of exceptional harmony. (ii) Exceptional blocking. This is demonstrated by exceptional morphological blocking in Turkish. Insofar as locality is concerned, this type of morphological blocking is not distinct from phonological opacity.
The two unattested patterns are exceptional non-triggers of harmony and exceptionally transparent vowels. On one hand, if exceptional non-triggers are found to exist, then exceptional non-triggers can be predicted to fall under the locality pattern exhibited by exceptional blocking (i.e. bound to the exceptional morpheme). On the other hand, if exceptionally transparent vowels are found to be prevalent in vowel harmony languages, then this will contravene the locality principle which has been espoused for exceptional occurrences.

Alternative approaches to exceptional phenomena within optimality theory
In recent theoretical discussion in the OT framework, there has been considerable interest in the way exceptional morphological interferences in phonology can be modeled (Anttila 2002;Inkelas and Zoll 2003;Pater 2000Pater , 2009. It is of special interest in an OT framework where all constraints are universal and individual grammars are a result of permutations of these constraints. In the cophonology approach of Anttila (2002), morphemes select their own ranking from a set of partially ordered constraints. Accordingly, only constraints that are unranked in the grammar can have lexically specified rankings. I will not go into the details of the cophonology approach. (See Anttila 1997Anttila , 2002Inkelas 1998;Inkelas and Zoll 2005;Orgun 1996Orgun , 1998Orgun , 1999Orgun and Inkelas 2002 for an elaboration of the framework; and Pater 2000Pater , 2006Pater , 2009, for arguments against indexed constraint rankings and in favor of constraint indexation.) Among the diacritic approaches, the ones favoring faithfulness-only constraint indexation are many and varied (e.g. Fukuzawa 1999;Mester 1999, 2001;Kraska-Szlenk 1997; see also Alderete 2001;Benua 2000). It has been argued that morpho-phonological processes are the result of the grammar at large. The indexation of faithfulness constraints only was supposed to lend force to the argument that such grammar dependent processes have manifest limits on their range of occurrences (e.g. Alderete 2001;Benua 1997;Itô and Mester 1999). It was argued that lexical indexation can pose limitations on the scope and extent of exceptional occurrences simply because of the prohibition against the indexation of markedness constraints. Therefore, 'faithfulness only' indexation predicts languages in which there would be a limitation on the extent of variation possible in the markedness patterns of exceptional processes. The local conjunction approach adopted by Lubowicz showed that derived environment blocking effects can only be captured by the conjunction of markedness and faithfulness constraints. In the following subsections, I show how each of these approaches would lead to wrong results while analyzing exceptional vowel harmony in Assamese.

Morpheme-specific exceptions in cophonology theory
A theoretical approach which would founder with respect to the Assamese patterns of morpho-phonological interaction is that of cophonology, more specifically in the manner argued by Anttila (2002; one of the main proponents of the cophonology approach). In general, cophonology requires exceptionalities in the lexicon and morphology to be analyzed with the aid of lexically specified rankings of constraints which are already present in the grammar. In the approach propagated by Anttila (2002) only pairs of constraints whose ranking is unspecified in the general case can have lexically specified rankings. When a lexical item is unspecified for the respective ranking, then it should show variation.
I have shown in Sect. 2.1 that lexical indexation accounts for non-iterative raising in the context of a pair of morphemes in Assamese. The indexed markedness constraint bans a configuration of two adjacent vowels, of which the first one is [−ATR] and the second is [+ATR]. Since the indexed constraint is only sensitive to [ATR] alteration, it checks whether there is an adjacent [−ATR] vowel, which is repaired even if the [+ATR] feature of the vowel is contributed by another morpheme. While this is feasible in a lexical indexation approach, a cophonology analysis would predict iterative application of raising in stems with more low vowels than just the final one in the presence of the affixes in question. The remainder of this section shows how the approach advocated by cophonology will not work for the Assamese data. According to the lexically indexed constraint ranking approach, only unranked constraints can be lexically specified. In Assamese, the relevant unranked constraints would be IDENT[Low] and *[−ATR] [+ATR]. The specific lexical items in the lexicon then choose their ranking from the unranked pair. Accordingly, indexation of constraint ranking will produce [k p hi] 'made of cotton' and [morij ] 'beat' respectively in each of the two lexical items listed in (44a)  It is quite easy to show that the indexed constraint ranking approach would then generate *[oloxuw ] for the input / l x-uw / as a result of indexed constraint ranking. The tableau below shows how cophonology theory would predict incorrect results for the input / l x-uw /.
The candidate in (45c) would be the predicted outcome in ranking indexation. The actual output in (45b) incurs a fatal violation of the constraint *[−ATR][+ATR]. Anttila (2002) also presents this model to establish a connection between exceptionality and variation, so that whenever the constraints are unranked, the resultant output has the potential to vary either on the markedness or on the faithfulness dimension. This kind of variation is not attested in Assamese, and therefore this attribute of cophonology theory cannot be applicable to all instances of exceptionality in the grammar of languages. Further, as argued in Pater (2006Pater ( , 2009

/ /-raising and [Back] alternation: data and problems
In this section I lay emphasis on some of the intricacies of vowel harmony phenomena in the presence of mid vowels, which have not been elaborated upon until now. Section 2.1 dealt with the low vowel / / which exceptionally undergoes harmony when followed by /-ij / and /-uw /. Further complications arise when / / occupies the second position from the left in the root but there is a [−ATR] mid vowel in the initial position.
In sequences like /C C C/ + /ij /, where the output is [CeCeCij ], the derivation presents itself as a case of what is called emergent [Back] harmony in Assamese in this paper. Assamese vowel harmony is iterative and regressive, hence if / / undergoes harmony and produces a [+ATR] output, it is plausible that the resultant [+ATR] vowel will also trigger harmony in the preceding / / and / /. Hence in instances of /C C C/ + /ij /, it could be predicted that the outcome of / /-raising would result in [CeCoCij ] because [C CoCij ] will violate the constraint driving iterative vowel harmony, i.e. *[−ATR] [+ATR]. A quick look at the data below shows that this prediction is true, but the attested pattern of / /-raising in this environment shows another alternation along with / /-raising. In disyllabic stems, where / / is preceded by a mid vowel, apart from [+ATR] harmony, the feature that / / assumes for harmony is determined by the root-initial vowel. In other words, the presence of an initial mid front vowel leads to the emergence of front harmony in all these cases, where /C C C/ + /ij / → The process of / /-raising is sensitive to morphological factors, but [Back] harmony is sensitive to / /-raising. In other words, only an underlying / / which undergoes raising may alter its [+Back] value. This emergent [Back] harmony shows the characteristics of a derived environment process. The process addressed here requires the presence of / /, which is a candidate for raising because of the presence of the specific morphemes. There is a morphological domain restriction on this phonological process, as it is blocked if the preceding / -/ or / -/ is a prefix vowel. The allegiance to the morphological boundary of a prefix so that the prefixal vowel remains unchanged in its [Back]  To summarize, there are two phenomena involving preceding vowels: Prefix and root mid and high [−ATR] vowels harmonize when they precede [+ATR] vowels, but preceding root mid [ATR] vowels trigger backness harmony in harmonizing / / but prefix vowels do not. In order to make this clearer, I schematize this below: As shown above, prefixal vowels fail to initiate progressive front harmony. If exceptional / /-raising is demonstrated in the presence of preceding mid root vowels, the root vowel determines the [Back] feature that / / might assume, so that it becomes [+ATR] ([e] or [o]). It is clear from this behavior that the root-initial vowel is responsible for initiating a type of progressive [±Back] harmony, where the vowel which raises to satisfy [ATR] harmony can also change its [Back] specification in order to agree with the preceding vowel's specification for [Back]. These changes fail to apply if the preceding mid vowel belongs to the prefix.
Before proceeding to an analysis, it may be useful to summarize the facts involved in harmony in Assamese. / / undergoes regressive [+ATR] harmony and occasionally progressive [Back] harmony, and both of these occurrences happen in the presence of /-ij / and /-uw / (although progressive [Back] harmony is not directly dependent on /-ij / and /-uw /, but rather on a raised / /). The locality conditions are such that / / becomes [+ATR] only if the following vowel is a part of [-ij ] or [-uw ]. / / agrees in backness with a preceding mid root (but not prefix or high/low) vowel, if one exists; otherwise / / maps to [o]. Iterativity in this exceptional domain appears when an underlying non-/ / [−ATR] root or prefix vowel preceding a harmonized / / also becomes [+ATR], but if / / is the preceding vowel, it never harmonizes. Although IDENT[Back] is indispensable, the use of faithfulness constraints alone does not lead us to a satisfactory analysis of the back harmony phenomenon. The back harmony problem needs to be addressed as a reflex of a prohibition already present in the phonology of Assamese-that is, the avoidance of sequences of front and back vowels in the root. Neither the addition of IDENT[Back] nor any other faithfulness constraint 13 alone will predict the attested outputs with [Back] harmony, where the initial front vowel leads to the alternation of the / / in the following syllable as in / l h/+/uw / → [elehuw ] 'lazy'. In evaluating inputs like / l h/+/uw /, the constraint hierarchy above will predict * The reason for the resistance of / / to an account based on general markedness and faithfulness constraints is that it would be a far cry from presenting an analysis of the resultant harmony that the derived / / undergoes under the influence of the preceding mid front vowel / /. In the section below we address the problem of a non-underlying / / undergoing front harmony and try to contextualize it to the phonological lexicon of the language.

Derived environment effect in harmonic patterns: the analysis
Cases where phonological rules are only applicable to derived structures but are systematically blocked in nonderived environments are known to be instances of nonderived environment blocking (NDEB; Kiparsky 1993). A derived structure is one where (a) the segments participating in the phonological process traverse a morpheme boundary, or (b) the relevant segment, i.e. the segment undergoing a phonological process, arises as a result of another phonological process (applying in the same cycle, as per Kiparsky 1982 andMohanan 1986). In much of the literature which has dealt with non-derived environment blocking, these two processes have been identified as morphological NDEB and phonological NDEB, respectively.
I argue below that the derived environment effect (DEE) in Assamese is not necessarily blocked in non-derived domains. Regular phonological processes are often blocked from applying in languages if the result would lead to some phonologically ill-formed structure. The Assamese derived environment effect shows that there is no active blocking in non-derived environments and this is evident from factors related to the occurrence of the trigger and target in phonologically simplex domains. As will be shown shortly, there is a paucity of data showing the relevant trigger and target occurring in that environment. Apart from the absence of non-derived [e. . . o] sequences, even in derived domains the phonological DEE is the result of a derived environment created by the presence of an exceptional morpheme. This, however, does not make it a morphological DEE as this phonological DEE is also not productive in morphologically complex domains. Wherever the boundary of the trigger and target span a morpheme boundary, the DEE is not seen.
A scanning of the most comprehensive Assamese dictionary 14 showed that apart from restrictions on sequences of [−ATR] and [+ATR] demonstrating [ATR] vowel harmony (see Sect. 1), constraints on sequential occurrences are extended to mid vowel sequences also, but only in a specific domain. Dictionary entries showed that in roots, the sequence / . . . / is definitely marked, even though it is not non-existent. Mahanta (2008) found only four lexical items with underlying / . . . / sequences, three of which were loanwords. They are given below: Sequences of / / followed by / / -initial Gloss a. t r 'thirteen' b. b s n 'powdered pulse' c. r s n 'provisions' d. b t n ' s a l a r y ' While there was no discussion of the productivity of these sequences, a more thorough investigation has revealed that these are among the very few words containing underlying lexical / . . . / sequences. Interestingly, we may be tempted to surmise that these sequences of mid [−ATR] vowels do not undergo raising, either because of the opacity of the final /n/ or because in a word like /t r / suffixation is not possible. There is no way to test if these words would have resulted in harmony because /b s n/, /r s n/, and /b t n/ are also loan words and non-native loans normally do not combine with native suffixes. 15  We can now conclude that the accidental gap of the absence of [e. . . o] in roots in the phonology of Assamese is not so accidental, and the reason for this gap may be assigned to a constraint prohibiting unbridled occurrences of front and back mid vowels. Alternatively, one can analyze the absence of these instances with a constraint allowing front harmony, such as AGREE [BACK]. While that may be plausible, it must be borne in mind that there are no instances of front harmony holding in all domains, not across a morphological boundary. More importantly, front harmony is not found among the [−ATR] counterparts of [e] and [o]. High vowels never demonstrate an inclination for front harmony, and the process must be limited only to mid vowels. In such cases, cross-linguistic evidence for the occurrence of such a pattern of front/back harmony among mid vowels (restricted to root vowels only) will be quite marked. The other reason for the absence of [e. . . o] sequences may be the universal markedness of mid back rounded vowels. Studies presenting the markedness of [o] over [e] are available from work independent of the data attested in Assamese. Crothers (1978), in his hierarchy of implicational universals, has shown that languages with six, seven, or eight vowel inventories generally prefer to have [e] in their vowel inventories. [o] figures lower in this implicational hierarchy, which means that only those languages which have more than eight vowels will normally prefer to have [o].
[e] is also more preferred than [o] whenever there is vocalic epenthesis.
(Some examples are Spanish (Harris 1986a(Harris 1986b, Farsi (Shademan 2003), and Mohawk (Postal 1968).) The relative paucity of languages which use [o] as an epenthetic vowel is also a pointer in the direction of the markedness of [o] vis-a-vis [e]. 16 The Assamese data do not provide evidence for iterative front harmony across morpheme boundaries. We find many examples in which such highly marked seg-ments are permitted only in prominent positions, where prominent includes stressed, word-initial, and root segments (see Beckman 1995Beckman , 1997Beckman , 1998Steriade 1997;Zoll 1997). Therefore, the constraint which is required to analyze this part of the Assamese data is a markedness constraint which prohibits occurrences of the vowel [o]. In order to account for this we posit a licensing constraint prohibiting instances of the mid back vowel in non-licensed positions. LICENSE[−High−Low+Back] has the effect that a mid back vowel must be in the root, and it must be associated with a following [+High +ATR] vowel, as in /bohi/, 'book' /b tori/, 'news' /porohi/ 'day before yesterday' or /xogun/ 'vulture'. Prominent positions such as roots and initial syllables are also often dependent on higher-ranking faithfulness constraints, as has been well documented; see for example McCarthy and Prince (1995) and Beckman (1995Beckman ( , 1997Beckman ( , 1998 In the analysis that follows I account for the phonological blocking effects described above. The first constraint required is the specific faithfulness constraint in (58) which militates against a change of back values. The second component is the markedness constraint in (57) which prohibits a sequence of two adjacent front and back mid vowels. The combined effect of these constraints is shown below:

The Emergence of the unmarked analysis vis-à-vis other contending analyses
In the first Emergence of the Unmarked (TETU) analysis proposed by Prince (1994, 1995), it was shown that a markedness constraint M which may be inactive in all other conditions in a language may surface in that language in an environment conditioned by the presence of a specific morpheme. This analysis was posited for reduplication phenomena showing sensitivity to markedness preferences. The ranking proposed by McCarthy and Prince for such phenomena was one in which the markedness constraint M is ranked lower than the faithfulness constraint IO-FAITH. The markedness requirement M emerges in the language because it outranks other faithfulness constraints on Base-Reduplicant Identity. Consequently, reduplicants obey the markedness constraint but violate the constraint requiring faithfulness to the base (Alderete et al. 1999;McCarthy and Prince 1994). The surfacing of [Back] harmony in Assamese also shows that emergent unmarkedness may sometimes be underscored by phonological processes. I have offered an account that relies on positional faithfulness along with featural licensing, so that the analysis captures the desired effect that some instances of root controlled [Back] harmony are the result of emergent markedness restrictions.
In current OT approaches, problems presented by derived environment processes have been analyzed by revising the basic architecture of Optimality Theory. Approaches like that of local conjunction and comparative markedness try to address the difficulties presented by a process which is sensitive to a phonological derivation or a morphological boundary. Comparative markedness refers to differences between old and new markedness. Markedness constraints can refer to 'old' violations (those already underlyingly present) vs. 'new' violations (those created by Gen). In local conjunction approaches (as per Lubowicz 2002Lubowicz , 2005, DEE are seen to be a result of the conjunction of faithfulness and markedness constraints. The critiques in Sects. 4.4 and 4.5 discuss the challenges that would be presented by Assamese if these approaches were applied to the data.

The unsuitability of a local conjunction solution to Assamese derived environment effects
Local constraint conjunction was first proposed by Smolensky (1995) as a proposal about the internal structure of constraints. Some proponents of local conjunction (LC) (Itô and Mester 2003;Kirchner 1996;Moreton and Smolensky 2002) advocate for the local conjunction of faithfulness constraints only. We can compare the derived environment effects presented in this paper with the derived environment processes presented in Kiparsky (1993), Lubowicz (2003), and elsewhere. To analyze the problem of derived environment blocking effects (DEE), Lubowicz conjoins two constraints in order to explain phonological processes that are blocked in derived domains. The pattern we need to analyze here is such that unfaithfulness to [Back] specifications is required in the presence of mid vowels only, to the exclusion of all others, and only in the position where it is adjacent to the triggering vowel. The LC approach, if successful, would have predicted alternations which are not noticed in any language, such as alteration of [+Back] values of all vowels which are underlyingly [+Back] and [−High] if there is any change in the value of [Low]. In reality, the attested alternations in Assamese are not so effusive and random, and they follow a pattern which eludes the LC analysis.

Comparative markedness analysis
Comparative markedness argues for a model of grammar where underlying segments violate 'markedness old' whereas derived segments are violated by 'markedness new'. Markedness constraints can refer to 'old' violations, i.e. those already underlyingly present. The same markedness constraints can incur 'new' violations which are created by Gen. This approach will not provide us with a correct analysis of the facts in Assamese. In Assamese, derived /o/ is permitted in certain contexts and there is no absolute ban on the occurrence of /o/ derived from / /. The derived environment effect appears only in the presence of a preceding / /. This result seems intractable with the present theoretical tools offered by comparative markedness. and is therefore erroneously selected as the optimal candidate. This shows that comparative markedness is not a viable option for analyzing the derived environment effect in Assamese.

Conclusion
The exceptionally triggered vowel alternations discussed in this paper can be characterized as morphologically induced harmony, which is obtained at the cost of flouting the highly ranked phonological constraint IDENT [Low] (which prevents any alteration of the low vowel / /). In the first case, the suffixal morphemes /-ij / and /-uw / trigger exceptional alteration of the otherwise opaque vowel / /. The resultant raising of / / to a [+ATR] vowel in this context violates the constraint IDENT[Low], which is otherwise highly ranked in the phonology of Assamese. Secondly, this leads to the emergence of [Back] harmony whenever there is a preceding front mid vowel. In the rest of the paper, I describe these morpho-phonological interactions and discuss several aspects of locality and emergent unmarkedness which have consequences for theories of exceptions and DEEs. The derived environment data show a DEE effect which is masked in the regular phonology. It only emerges as a consequence of raising in exceptional vowel harmony. The data show that this effect, although appearing in a derived domain, is a reflex of the grammar at large and is actually an emergent unmarkedness pattern. The analysis proposed in this section is superior to other competing analyses because it can effectively contain long distance effects in exceptions and it also shows how such restraining of harmony in the local domain may facilitate the emergence of the unmarked segment [e] and result in the process of [Back] harmony in Assamese.
This work also addresses the typology of exceptions in vowel harmony and shows that locality constrains exceptional harmony processes. In a strain different from Finley (2010), I have argued that exceptional triggers and exceptional undergoers are the only exceptional types which need to establish a harmonic relation with an adjacent element. Other exceptional patterns like blocking do not need any additional convention of locality as they behave like regular phonological blocking, which is also locally bound. Thus, this paper shows that exceptional triggering is the only truly local phenomena specific to exceptional vowel harmony. In all other types of exceptionalities, the facts require only mechanisms which are observable in other vowel harmony phenomena. The only type of exception that a theory of exceptional vowel harmony has to constrain is the total absence of exceptionally transparent vowels. This may boil down to the fact that transparent vowels make way for the interaction of two non-adjacent units, i.e. the trigger and the target.
This article is a contribution to the understanding of harmony processes in general. Vowel harmony is normally iterative, but iterativity caused by feature spreading in vowel harmony is typically myopic (Wilson 2003(Wilson , 2006. In a typical harmonic environment, when a feature spreads to a neighboring element it is only sensitive to whether the neighboring segment can undergo harmony; harmony is not dependent on any element beyond the immediate segment in a string. If harmony is not dependent on factors beyond string-adjacency, the operation is dependent on local factors, not global ones. This paper demonstrates how exceptional and emergent processes in an iterative harmony language show harmony patterns which are determined strictly by local factors. However, this operation has to be different from locality in iterative harmony where a lack of morphological indexation allows harmony to proceed iteratively. In the case of DEEs and exceptional harmony, a pre-conceived idea of locality in iterative harmony would give rise to an analysis with pathological harmony. Although exceptions and DEE are different as phonological processes, the interaction of the two phenomena with vowel harmony shows that both are local.