Research Interests


My research philosophy may be summed up as follows:

·Descriptive adequacy. “Chaque langue est un système où tout se tient” (Meillet 1913); or, in Sapir’s words, each language has its own “structural genius” that underlies the “sundry facts” (Sapir 1921). Most linguistic work done today, however, looks only at bits and pieces of systems, rather than whole systems. Sometimes this atomistic approach is even argued for explicitly (McCarthy 2008: 31-32).

A descriptively adequate grammar is a grammar that generates all of the grammatical forms of a language, and excludes all of the ungrammatical ones (Chomsky 1964: 62-63). But how is descriptive adequacy to be evaluated? If we restrict our analysis only to a small part of a system, how can we distinguish between genuine counterexamples, i.e. those forms which the analysis has failed to explain, from those forms which are merely beyond the scope of the analysis? In fact, I believe this is impossible: analyses based on small data sets are always left with an element of uncertainty, since we don’t know what else is out there in the language. There is no descriptive adequacy without descriptive completeness.

·Formal precision. While there has been much discussion in recent years on the need for empirical evidence to support theoretical claims (Wasow & Arnold 2005), an equally serious problem, in my view, is theoretical claims in the literature which are not clearly defined. How can we determine whether a claim is true or false, if it is stated in a vague or imprecise manner?

Specifically, since the introduction of Optimality Theory (Prince & Smolensky 1993), it has become common practice to omit any reference to autosegmental tier structure and the feature-geometric representation of segments (e.g. Clements 1985) when constructing phonological analyses. That is, in OT, it is commonly assumed that either (1) the unstructured feature matrices of SPE are adequate for phonological analysis; or (2) some sort of autosegmental or feature geometric structure is necessary, though it need not be defined precisely nor referred to explicitly (presumably, since almost any set of representational assumptions would work equally well).

Insofar as the introduction of ranked and violable constraints can lead to simpler representations, this is of course a desirable result. It is also true that, for some particular sets of data, there may be several autosegmental or feature-geometric models that work equally well. Failure to specify any representational model, however, means that the resulting phonological analysis will be poorly defined, with unclear empirical predictions, since it is to representations that all OT constraints ultimately refer, and it is on the basis of these representations that constraint violations are ultimately computed.

For this reason I am rather conservative as a phonologist, in making use of a number of 1980’s style representational devices in my work: lexical phonology, autosegmental phonology, and feature geometry—even when using OT (see Phonology).

·No theory without description, and no description without theory. There is no such thing as pure “bottom-up” linguistics (Gil 2001). We always need some sort of hypothesis to guide empirical work, otherwise we would not know what data to collect, or how to interpret it. I thus take issue with popular catch-phrases such as “empirical foundations of linguistic theory”—one could just as well say that formal theory is the foundation of empirical research since, without it, even the simplest observations would be impossible (see Language Documentation and Description).

Back to Research Interests contents ↑


I have three main areas of interest in phonetics, all of which broadly relate to my work in phonology).

1) Consonant length and gemination. Weledeh has a categorical long-short distinction for consonants as well as vowels, although only the latter is written in standard orthography. The Prosodic Reversal Hypothesis (i.e. my dissertation—see Publications) makes specific predictions about where these should and should not occur. In addition, I am interested in more subtle effects of consonant length—for example, singleton consonants tend to be shorter in the middle of a foot.

2) Phonetic correlates of stress. Stress, in both Weledeh and Chipewyan, is realized by increased F0, increased duration, and increased amplitude of the stressed vowel, even though tone and vowel length are contrastive in both languages. That is, the stressed vowel (if short and Low-toned) is pronounced higher and longer, but not so high or long that it would be confused with a phonemically long or High-toned vowel. There are a number of potential confounds in measuring stress in Athabaskan languages—one is that, given the very large phoneme inventory, it is often difficult to come up with a set of true minimal pairs.

3) Nasality and vowel dispersion. Nasality has a deleterious effect on vowel distinctness, because coupling of the oral and nasal resonators results in an overal spectral “flattening”—this is especially true of F1 (House & Stevens 1956, Fant 1960, Krakow & Huffman 1993, Stevens 1998, Ladefoged 2003). This means that, all other things being equal, nasal vowels will be perceived as more like one another, even if the actual articulatory effort expended to produce them is the same. In terms of Dispersion Theory (Flemming 1996, 2001), this could lead to two different types of results: either the inventory of nasal vowels will be reduced, as happens in many languages, or additional articulatory effort will be expended to keep the vowels distinct. Interestingly, in Weledeh, both outcomes are attested in the same language. In Weledeh verbs, there is a process of nasal raising at the Inner Stem Level, where ę → į and ą → ǫ, such that there are only 3 nasal vowels in the inventory at this level, į, ǫ, and ų. At the Word Level, however, nasal vowels do not raise, and there is a full inventory of five nasal vowels: ą, ę, į, ǫ, ų. I have not investigated nasals in Chipewyan though the situation there appears to be similar. I hypothesize that this nasal raising process is the result of formant integration (Chistovich & Lublinskaya 1979, Beddor 1984)—that is, nasal raising occurs in order to align the frequencies of oral and nasal resonances. I am planning a series of experiments to investigate these issues in more detail.

Back to Research Interests contents ↑


Jump to Phonology subsection:

As a fieldworker and Athabaskanist (or Dene-ist, as it were), my interests center on those areas of phonological theory which are relevant for the study of Dene languages. Dene (Athabaskan) languages are famous for their highly complex verbs and verb morphology. Traditionally, descriptions of the phonology of Dene languages have relied heavily on morphological conditioning: phonological processes are described as applying only within some particular morphological context. An example is given in (1).
(1) “Further checking is required on the wh perfectives, especially the third plural and first personal plural/indefinite forms. the low tone which results from deletion of the perfective prefix at a conjunct boundary does not always appear. Also, where more than one prefix precedes the perfective marker, low tone or stress sometimes appears on the second syllable to the left of the mode in non-third person forms” (Ackroyd 1982: 116).
If taken literally, statements such as the above suggest that any phonological process may be triggered, or blocked, in almost any morphological environment. We should not expect, under this view, that phonological processes should apply in any general way throughout the language. Furthermore, the phonological processes proposed often seem phonetically unnatural—why, for example, should deletion of the segment wh result in the reation of Low tone, which then floats two syllables leftwards? And why does this process apply only in 1st and 2nd person forms?

The goal of my phonological research then, is to eliminate morphological conditioning from phonology—that is, the view that morphosyntactic features can directly manipulate phonological material, either by word formation rules (Anderson 1992, Stump 2001, Brown & Hippisley 2012) or morpheme-specific co-phonologies (Inkelas 1997, Inkelas & Zoll 1997). In my dissertation (Jaker 2012) I showed that the phonology of Weledeh is almost completely regular, once a few simple prosodic rules are taken into account. If Dene languages—seemingly the poster-child for morphologically conditioned phonology—can be shown to be phonologically regular, this casts serious doubt on whether morphological conditioning is needed in phonology in general (or, if the reported cases are the result of mis-analysis). This in turn casts doubt on the need for “inferential-realizational” theories of morphology, since a regular phonological system can be handled by simple concatenation of morphemes (see Morphology).

The following sections outline my approach, and the areas of phonology I am interested in.

Back to Phonology contents ↑Back to Research Interests contents ↑

·Lexical Phonology and the Dene (Athabaskan) Verb

Broadly speaking, three models of Dene verb structure have been proposed: the “interrupted synthesis” model (Kari 1992), the scope model (Rice 2000), and what Halpern (1992) has termed the “stem-core” model (Hargus 1988), based on Lexical Phonology. This last model—the one which I assume—maintains that the Dene verb is built inside-out, from right to left, so that the structure of the whole word has the appearance of a right-branching tree, as in (2). Template position numbers (Rice 1989) are given under each group of prefixes.

Under this model, first the root is combined with the classifier and suffix, and phonology applies, at the Root Level. Then the output of this level, the “stem,” is combined with the inner conjunct prefixes at the Inner Stem Level, and phonology applies again, etc. This model is compatible with both rule-based and OT analyses—indeed, I explored both in my dissertation. Under this model, it is possible for different groups of prefixes to exhibit different phonological behavior, without resorting directly to morphological conditioning. Rather, different groups of prefixes are added at different levels, and undergo whatever set of phonological processes are characteristic of that level .

Back to Phonology contents ↑Back to Research Interests contents ↑

·Metrical Stress Theory and Prosodic Typology

“…an understanding of this structure, the prosodic structure, will provide fresh inisght into the basic operation of Athapaskan languages and will ultimately yield an understanding of Athapaskan languages that makes them look less bizarre and more like better understood languages” (Rice 1990: 243).

Metrical phonology is the branch of phonology that studies stress and prosodic constituency (Hayes 1995). It is generally agreed that groups of stressed and unstressed syllables are grouped together into units called feet. The two main types of feet are the iamb and the trochee. The definition of iamb and trochee in modern metrical phonology is somewhat different from that found in classical metrics (where they are defined in terms of quantity). In metrical phonology, feet are defined in terms of stress. Thus, any group of syllables with initial stress is a trochee, and any group with final stress ia an iamb, by definition. The most well-formed, i.e. ‘canonical’ feet, are illustrated in (3). Stress is indicated here by underlining, except in monosyllabic feet, where the underlining is omitted.

The feet in (3) represent the so-called ‘canonical’ feet in iambic systems (3a) and moraic trochee systems (3b), that is, the most well-formed feet that are allowed in most or all such systems. Which other types of feet if any should be allowed, both cross-linguistically and within particular languages, is a point of debate within metrical phonology (Hyde 1999, Mellander 2003). However, note that feet are defined as iambs or trochees strictly in terms of stress. Thus, a foot such as (ta.taa), is still a trochee, by definition, because it has initial stress, whereas (taa.ta) is an iamb, by definition, by virtue of its final stress, even though their weight pattern does not match the stress pattern. This differs, again, from classical metrics, where feet are defined quantitatively, and the same feet would be defined as an iamb and a trochee, respectively.

There are certain types of phonological processes that occur within feet. These processes are so common cross-linguistically, in fact, that they can be used as diagnostics: if a certain phonological process has applied, it is evidence that a certain type of foot was present.

All of the processes in (4) are attested in the NE Dene languages. In my work, I use metrical phonology, and constraints on foot structure and foot type, to motivate many of the complex morphophonemic alternations found in Dene languages (see here). Of course, feet cannot be observed directly—all that one directly observes are stressed and unstressed syllables, long and short vowels, High and Low tones, and a large number of unexplained morphophonemic alternations. So to put things another way, I use morphophonemic alternations in Dene languages as diagnostics for the presence of feet. In my view, the strongest argument for feet is that they virtually eliminate the need for morphologically conditioned phonology, and allow us to formulate an analysis of Dene phonology that looks far more regular and typologically natural.

Back to Phonology contents ↑Back to Research Interests contents ↑

·Feature Geometry

The primary motivation for feature geometry is constituency. That is, certain groups of phonological features seem to behave as a unit for the purposes of phonological processes such as deletion or spreading. In Dogrib, for instance, all vowel features behave together as a unit, as do all place features. For example, in vowel coalescence, when /e-a/ → aa, all vowel features behave as a unit: all of the vowel features of e are deleted (i.e. [front]), and replaced with all of the vowel features of a (i.e. [low]); it is not the case that these two features are both preserved in the output, as this would generate a new segment, [ææ] ([low], [front]) which is not part of the inventory of vowels in the language. In feature geometry, this is explained by the existence of a node called VPlace, which carries all vowel features (Padgett & Ní Chiosáin 1993). Similarly, when voiceless fricatives such as ł are debuccalized syllable-finally, i.e. ł → h, all of the place features of ł, namely [lateral] and [coronal], are deleted; it is not the case, for example, that only [lateral] deletes, leaving behind a plain coronal fricative such as s.

Prior to feature geometry, in SPE phonology, the fact that certain groups of features behaved together as units was expressed using variables: α, β, γ..etc. The use of this notation was argued to be too powerful, however:

“…feature geometry allows for the elimination of powerful variable-notation in phonological rules. Such variable-notation was formerly used in the statement of common assimilatory and dissimilatory processes: under place-assimilation a segment becomes [α coronal, β anterior, …] when adjacent to [α coronal, β anterior …]. Such assimilation is now expressed as simple spreading of the place-node. Variable notation can be eliminated, and as a consequence, the theory is more appropriately constrained” (Blevins 1994: 306).

In other words, the earlier, SPE-style variable-notation allowed, in principle, for spreading and dissimilation rules consisting of arbitrary combinations of features, whereas under feature geometry such rules are restricted to natural classes of features, such as place or manner features. It should be noted that Blevins’s critique of variable-notation in SPE also applies to those OT analyses which, just like SPE, assume that segments consist of unordered feature bundles. Thus, while it is empirically well-motivated to posit a constraint which specifically requires adjacent coronal consonants to agree in place features—as is the case in Tamil for example (i.e. aɳ.pa is allowed, but not *aɳ.ta—Christdas 1988: 154-157, 438-446)—yet if segments have no internal structure, one might equally well require adjacent labial consonants to agree in manner (i.e. anta is allowed, but not *ampa), something which is unattested. In other words, if it is possible to write a constraint AGREE(CORONALPLACE), why not AGREE(LABIALMANNER)? Thus, feature geometry serves essentially the same function in OT as it does in rule-based phonology: to simplify the characterization of assimilation and dissimilation processes and to rule out processes involving unnatural classes of features. The difference is that, in OT, this restriction holds at the level of constraints rather than rules. In other words, using feature geometry in OT helps us simplify the universal constraint set (CON) and clean up our factorial typology.

The model I use is based largely on Clements (1985), as shown in (5).

The diagram in (5) presents only those features and class nodes which are necessary to account for the inventory of segments and the range of phonological processes attested in NE Dene languages; I do not claim that additional features or class nodes do not exist in the phonology of other languages. Near the top of the feature hierarchy is the root node, located on the root tier. The root node, indicated by a dot (•) is what defines a segment as such. That is, a segment is a bundle of features dominated by one and only one root node. Immediately dominating the root tier is the CV tier. This is the tier on which a segment is labeled either as a consonant (C) or a vowel (V) for the purpose of constraint evaluation (i.e. for constraints such as MAX(C) or IDENT(V)). Located above the CV tier are the various suprasegmental tiers, such as the moraic tier, the syllable tier, the stress tier, and the tone tier, not shown in (5).

Below the root node, following Parker (2001), I assume three major class nodes: Laryngeal, Manner, and Place. The properties of voicing, aspiration, and glottalization are determined by the Laryngeal node. Thus, voiced segments such as n and m bear the feature [voice], aspirates such as t and k bear the feature [spread glottis], or [spr gl], and ejectives such as t’ and k’ bear the feature [constricted glottis], or [cnstr gl]. The Manner node is responsible for the range of properties commonly referred to as sonority. Thus, fricatives bear the feature [continuant], or [cont], stops bear the feature [stop], and affricates bear both features [stop] and [continuant] (Sagey 1986, Lombardi 1990)¹. Like fricatives, sonorants, such as r, are also [continuant], and in addition bear the feature [sonorant], or [son]. Nasals, such as n and m, are a special type of sonorant which also bear the feature [nasal], or [nas]. Finally, we come to the Place node. Following Padgett & Ní Chiosáin (1993), I assume that there are two types of place features: consonantal place, located directly under the Place node, and vocalic place, located under a subordinate class node called VPlace.

An example of how feature geometry works is given in (6), where vowel assimilation is represented as spreading and de-linking of VPlace nodes.

For more on my views of representations, see Chapter 6 of my dissertation.

Back to Phonology contents ↑Back to Research Interests contents ↑

·Prosodic Morphology

So far, we have examined Lexical Phonology, metrical phonology, and feature geometry. These three theoretical devices, taken together, go a long ways towards explaining away the apparent irregularity and need for morphological conditioning in Dene phonology (see Publications). Yet, there is a residue of cases that seem resistant to a straightforward phonological analysis. What should we do with them?

The first thing to consider is that some apparently irregular phonology is really not phonology at all—it is really just morphology, which may require some special lexically listed allomorph. More interesting, however, are cases that seem to involve prosodic morphemes. That is, morphemes which consist, either wholly or in part, of suprasegmental phonological material. There are really two types of such morphemes. The first type is the prosodic morpheme proper, as shown in (7), which consists only of suprasegmental material, and has no segmental content. In Weledeh, Outer Stem Level iambic lengthening is motivated by a special prosodic affix of the shape (Light-Heavy), which occurs in the perfective paradigms of d- and l-classifier verbs. This prosodic affix is given in (7), in LFG notation (Bresnan 2001, Dalrymple 2001).

The lexical entry in (36) contains three constraining equations (Bresnan 2001: 60-62, Dalrymple 2001: 115-117), marked with the symbol ‘=c’. That is, the presence of this prosodic morpheme is constrained, in that it can only appear in verb forms that are middle voice, perfective viewpoint aspect, and activity situation aspect, though the (Light-Heavy) morpheme itself does not contribute this semantic content. The presence of this affix triggers iambic lengthening in Weledeh: (ge.ghe)dǫ → (ge.ghaa)dǫ → (ge.aa)dǫ ‘they drank’ (perf). It is necessary to achieve this lengthening using a prosodic morpheme, because iambic lengthening is otherwise unmotivated given the ranking of prosodic constraints in the segmental phonology of Weledeh (Jaker 2012: 554-562).

However, there are other cases which represent sort of a hybrid between a regular segmental morpheme, and a pure prosodic morpheme as described in (7). In these cases, a segmental morpheme comes lexically pre-associated to some suprasegmental content, as is the case with my reconstruction of the prefix *the in Proto Northeast Dene, shown in (8) (see here).

In (20), the conjugation marker *the is lexically pre-associated to an iambic foot, which is stressed on the second syllable, and bears a High tone on the first syllable, i.e. the syllable preceding the itself. The morphophonemics of this prefix are the most notoriously complex of any prefix in Dene languages. However, the reason the phonology of the is so complex, I argue, is that (a) this prefix has its own pre-specified stress and tone properties, which disrupt the normal iambic prosody of the language, and (2), these lexically pre-specified properties are themselves mutually conflicting, in that they involve a mis-match between stress and tone. Thus the manner in which this conflict is resolved, as well as the manner in which the normal iambic prosody of the language is altered, will be different in different prosodic environments, depending on both the stress pattern and tone pattern of neighboring syllables. See here for a full account of the phonology of this prefix in NE Dene languages.

A possible criticism of this approach is that it is really just morphological conditioning by another name, i.e. “morphological conditioning encoded as prosody”. However, I believe that lexically pre-specified suprasegmental structure is a particularly good way to encode morphological irregularity, because it is far more constrained in its predictions than, e.g. Paradigm Function Morphology (Stump 2001), or A-Morphous Morphology (Anderson 1992), where morphosyntactic features can directly manipulate phonological material in almost any conceivable way (e.g. “replace all uvular stops with High toned vowels in the first person dual”). There is only a small set of foot types (iamb, trochee) and a small set of phonological processes that can be conditioned by feet (e.g. syncope, lengthening, consonant lenition). Thus a Lexical Phonology approach, combined with metrical phonology and prosodic morphology, provides for a far more constrained and explanatory theory of grammar (see Morphology).

Back to Phonology contents ↑Back to Research Interests contents ↑


Jump to Morphology subsection:

The main goal of my work in morphology is to develop a model which is descriptively adequate for Dene languages, makes sound typological predictions, and is compatible with my phonology work on these languages. The basic model I assume in morphology is the same as in phonology, i.e. the Lexical Phonology model (Kiparsky 1982, 1985, 2000; Kaisse & Shaw 1985), which, when applied to Dene languages, is also called the “stem-core” model (Rice 1982, 1989; Hargus 1988; Jaker 2012). This is illustrated in (1).

Note that this is a morpheme-based model. That is, it assumes that “there is only one operation in morphology: combination” (Stonham 1994). The Dene verb is organized into a series of levels, and built up incrementally, from right to left, so that it has the appearance of a right-branching tree. At each level, the phonological side of morphemes undergoes various phonological processes, while the semantic content is built up compositionally. Of course, this model is in some sense more challenging to use than A-Morphous Morphology (Anderson 1992) or Paradigm Function Morphology (Stump 2001) which allow the use of much more powerful formal devices (such as replacement rules and subtraction rules), especially in a Dene language, with so many irregular, morphologically-conditioned phonological processes, non-compositional semantics, and discontinuous dependencies—or so it would appear.

However, I believe the restrictiveness and lesser theoretical power of the Lexical Phonology model are its greatest virtues. As I have argued elsewhere (see Phonology) there is in fact no such thing as morphologically conditioned phonology in Dene languages, and hence no need for subtraction or replacement rules. This is especially true if metrical phonology is taken into account. Apart from this, the suppletion, ablaut, verb classes, selection and blocking restrictions, and discontinuous dependencies of Dene languages are typologically marked patterns in morphology, and should be theoretically problematic: from a morphological perspective, Weledeh and Taltsą́ot'ıné really are more complex than Tamil or Turkish. A truly explanatory theory of morphology should explain both how the unique properties of the Dene verb are possible, and also why they are so unusual cross-linguistically.

Read on to learn more about the topics I’m working on in morphology.

Back to Morphology contents ↑Back to Research Interests contents ↑

·Morphological Typology.

One perspective which I believe to be important in morphology, and which was neglected for most of the 20th century, is morphological typology—that is, the classical division of languages into morphological types: isolating, fusional, agglutinating, and polysynthetic. In the 19th and early 20th centuries, morphological typology was a central concern of linguistic theory, as well as being a culturally important concept. For example:

“One celebrated American writer on culture and language delivered himself of the dictum that, estimable as the speakers of agglutinative languages might be, it was nevertheless a crime for an inflecting woman to marry an agglutinating man. Tremendous spiritual values were evidently at stake” (Sapir 1921: 124).

However, morphological typology is largely neglected today:

“Sapir’s elaborate typology (1921) was never sucessfully applied or developed by other scholars, and the resurgence of interest in language universals since 1957 has not provoked any serious re-examination of the old classification. Yet the old labels ‘isolating’, ‘agglutinating’, ‘inflecting’ (or ‘fusional’) and ‘polysynthetic’ continue to be widely used. They therefore have a peculiar status in concemporary linguistics. They are not part of the technical vocabulary of any currently influential theory of grammar except Natural Morphology…. On the other hand, these labels do provide a convenient shorthand for clusters of morphological characteristics which are implicitly assumed to go together” (Carstairs-McCarthy 1992: 166-167).

Arguably, morphological typology should be at the top of the agenda for generative grammar:

“All syntactic differences reduce either to differences in the properties of lexical items or to visibility conditions.” These visibility conditions “ensure that meaning is (to a tolerable degree) recoverable from form” (Baker 1996: 505).

Morphological typology is thus the study of what is visible, where, and how.

There is no single criterion adequate to characterize all morphological types. There are three dimensions along which languages may vary:

--what kind of information is expressed within the word
--how much information is expressed
--what technique is used.

What kind of information refers to Sapir’s “pure relational” vs. “mixed relational” languages, which corresponds loosely to the modern head-marking vs. dependent marking distinction (Nichols 1986). The terms isolating, agglutinating, and fusional are somewhat of a conflation of morphological technique and degree of synthesis; all other things being equal, agglutinating and fusional languages differ only in technique.

(3) Definitions of Isolating, Agglutinating, and Fusional.
Isolating: Lexemes are invariant. No affixation, ablaut, or stem suppletion.
Agglutinating: Prefixes or suffixes are attached to stems. Words can be neatly segmented into morphemes (stem, prefixes, suffixes), each affix has a uniquely identifiable meaning, and each affix expresses only one morphosyntactic feature.
Fusional: A single affix expresses multiple morphosyntactic features (portmanteau morphs). Stems may also exhibit ablaut and/or stem suppletion. Words may exhibit nonconcatenative morphology (e.g. triconsonantal roots, tonal melodies). The inflectional system as a whole may comprise multiple noun and/or verb classes.

However, the diachronic development of these types is not entirely clear. Isolating languages are known to become agglutinating through grammaticalization or morphologization (Lehmann 1995), which has been well-researched in the literature. Fusional languages may also become isolating through loss of inflection, which has been observed in Indo-European and many other language families. This process may be attributed largely to sound change, which erodes word-final affixes. However, the process by which agglutinating languages become fusional (or “inflecting”) is still poorly understood. Lehmann calls this process demorphemicization (1995: 14) but there are very few attested examples, and no existing formalization. The Northeast Dene Languages can help fill this gap in our knowledge, since they represent a rare transitional state typologically: they are underlyingly agglutinating, but surface-fusional, to varying degrees.

When an agglutinating language becomes fusional, basically what was previously a concatenation of several different affixes becomes re-interpreted as a portmanteau morpheme (see How do new morphemes arise?), and this new portmanteau morpheme can then spread to new paradigm cells, or new verb classes, by analogy and semantic bleaching (see Syncretism and Analogy). In a morpheme-based theory, these diachronic changes are described largely as changes to the lexical entry of individual morphemes, the addition of new morphemes to the lexicon, and the loss of others.

However, the lexicon is not a “black box” that cannot be inquired into. We can and should ask the question, “what kinds of morphemes will a language have?” That is, just as, in phonology, the inventory of phonemes in a language is determined by constraint ranking (e.g. high-ranked *Uvular means there will be no uvular consonants), so too in morphology a language’s morphological type is determined by what kinds of morphemes it has in its lexicon, which is in turn the result of constraint ranking. Specifically, I have proposed the following factorial typology of morphological constraints, to derive the classical morphological types.

I explored an application of these constraints in this paper. Of course, isolating, fusional, and agglutinating are very general, global properties of whole morphological systems, while most work in morphology deals with particulars (selectional and blocking restrictions, suppletion, verb classes, paradigmatic gaps, etc.). How to integrate the general and the particular in morphological analysis is a long term research question for me.

Back to Morphology contents ↑Back to Research Interests contents ↑

·Selection and Blocking Effects.

While there has been much discussion of “blocking” in the morphological literature, in the sense of how, e.g. better and best block *gooder and *goodest, or mare blocks *horsess (Kiparsky 2004), the kinds of selection and blocking effects found in Dene languages have not been formalized in any morphological framework, to my knowledge. Sometimes, two prefixes must obligatorily occur together, under certain circumstances. For example, certain stems require certain situation aspect markers (i.e. conjugation markers) in position 10 in the perfective—although this can be overridden by other factos (Rice 2000: 262). More commonly, two prefixes are forbidden from occurring together in the same word. For example, the accomplishment situation aspect morpheme the in position 10 is forbidden from co-occurring with the perfective viewpoint aspect morpheme ne in position 11, and similarly ne is also forbidden from co-occurring with the d- or l-classifiers in position 13. I refer to this as blocking, although it might also be referred to as “negative selection”. A summary of morphological dependencies in the conjunct domain in Dogrib is given in (5) (taken from page 119 of my dissertation).

It is not immediately obvious at what level of representation these selectional and blocking effects should be formalized. That is, it could be that certain morphosyntactic features incompatible with other features (for example, middle voice perfective is not allowed). Or these effects could operate at the level of phonological forms (i.e. “morphs”). Or it could be certain morphemes (sound-meaning pairings) that select or block other morphemes (Spencer 2004: 101-106), or perhaps some combination of all above, on a case-by-case basis. The dotted lines in (5) represent selectional restrictions that have no discernible semantic motivation—for example, in conservative Dogrib, special allomorphs ı and aa are used in place of h (1sgS) and ah (2plS) in the perfective of Ø/h-classifier verbs, even though there is no special semantic connection between, e.g., perfectivity and the 2nd person plural.

The approach I have taken thus far is to adopt the formal device of constraining equations from LFG (Bresnan 2001, Dalrymple 2001). That is, by using the equals sign with subscript “c”, it is possible to assert that some morpheme is only allowed to occur in a specific morphosyntactic environment—or, just as easily, a morpheme can be forbidden from occurring in some environment. For example, the prosodic morpheme in (6) contributes no new meaning of its own, but is constrained to occur in the perfective forms of activity (i.e. durative atelic) middle voice verbs.

As a slightly different example, the morpheme the in NE Dene languages (also realized as fe or whe in Slave and Dogrib) does contribute its on meaning, accomplishment situation aspect (Rice 2000), but is only allowed to appear in the perfective (disregarding so-called “neuter verbs”), even though the prefix the itself (arguably) does not carry the meaning of perfectivity.

Thus, under this approach, it is morphemes (i.e. lexically listed pairings of sound and meaning) which are being selected by morphosyntactic features. I chose this approach largely because a formal notation for it already existed, but it seems to work in every case I am aware of, in Dene languages. A potential downside of this approach is that it requres that every morpheme “mean” something, or at least be selected by some set of meaningful morphosyntactic features—it is not possible to model selectional effects that are truly semantically empty. This might be a problem, for example, in a language with completely arbitrary verb classes, e.g. the conjugation vowel a appears in class 1 and e in class 2—unless of course “class 1” and “class 2” are taken as a kind of “meaning” (cf. Carstairs-McCarthy 1994, Spencer 2004: 81). However most selectional effects in Dene languages do seem to be governed by some set of meaningful morphosyntactic or semantic features.

Back to Morphology contents ↑Back to Research Interests contents ↑

·Syncretism and Analogy.

A common criticism of morpheme-based approaches to morphology is that they cannot account for properties of whole paradigms or whole morphological systems—for example, the fact that, in Latin, the dative and ablative plural of nouns are always identical (Spencer 2004). And it is true that, if the only theoretical device we were allowed to use in morphology were information in the lexical entries of morphemes, then any properties of whole systems would be epiphenomena of these lexical entries. For example, we could say that Tamil is agglutinating because the lexical entry for each morpheme generally contains information about just one morphosyntactic feature, e.g. gal ‘plural’, gir ‘present’, v ‘future’, d ‘past’, etc.—there are very few empty, suppletive, or portmanteau morphemes in the language. But why should all of the morphemes in the language be so simple and transparent in this way? And why should Latin and Sanskrit have so many complex and irregular lexical entries by comparison? This is no different from asking, why do English and German allow so many complex coda consonant clusters (asks, fifths, clothes, masks) while Italian and Tamil allow only allow geminates and sonorants (alto, arte, anzi, atto)? In phonology, no one would suggest that the notion of the segment should be abandoned, in order to capture generalizations about the inventory of segments in a language. In morphology, it makes just as little sense to argue that the notion of the morpheme should be abandoned just because there are generalizations to be made about the inventories of morphemes in particular languages. Rather, in both cases, what is needed is a theory of markedness, to define which segments or morphemes are possible in which languages, and in what environments. What morphological theory has been lacking—since the days of Sapir—is such a theory of morphological markedness. Once we have such a theory of markedness, it will be possible to state morphological generalizations over whole paradigms and whole languages, without abandoning the notion of the morpheme.

Below we will consider one case, that of the 1st person dual forms in Weledeh. For conservative speakers, 1st person dual forms can be realized in a number of ways depending on aspect (perfective, imperfective, optative) and verb class, of which there are 6 in the perfective, under my analysis. Some full perfective paradigms are given below, for all 6 verb classes. The underlying forms are given in (8)-(13), and the surface forms are given in (14)-(19).

We now return to the case of the 1st person dual morpheme. For conservative speakers, the 1st person dual subject prefix is always /wìd/, whose surface form varies according to aspect and verb class, depending on what other prefixes it occurs next to. But in innovative speech, this prefix has been replaced by an invariant prefix dıì ‘1duS’ on the surface. This prefix dıì is invariant in all verb classes and aspects. Thus, the 1st person dual forms in (14)-(19) would be shèdıìtı̨ ‘we (2) ate’ (PERF), nàdıìdè ‘we (2) lived’ (PERF), enèdıìjǫ ‘we (2) were embarrassed’ (PERF), nìdıìt’a ‘we (2) arrived’ (PERF), nàdıìzè ‘we (2) went hunting’ (PERF), and xàdıìt’e ‘we (2) cooked’ (PERF). Historically (and still today for conservative speakers), dıì was a 1st person dual optative-inceptive /de-we-wìd/ or /de-u-wìd/, which spread to all 1st person dual forms. What motivates this analogical spread?

As I have argued elsewhere (see here), the NE Dene languages present a sort of transitional state between agglutination and fusion: they are underlyingly agglutinating, but surface-fusional (with lots of complex morphophonology in between). In Weledeh, this transition towards fusion is most advanced, such that the language is beginning to exhibit effects of syncretism and analogical levelling, typical of fusional systems. dıì is a surface-fused form (from /de-we-wìd/) that has replaced other surface-fused forms, e.g. whì (from /whe-wìd/) and ì (from /e-wìd/). This type of process would seem to suggest some kind of surface-oriented paradigm uniformity effect—but how could we account for this type of process in an abstract, derivational, morpheme-based theory? After all, strictly speaking, dıì shouldn’t even exist as a morpheme under a traditional morpheme-based account (it is derived from 3 other morphemes)—how can it spread if it doesn’t exist in the first place?

I propose that this happens through a 2-step process of bracket erasure and semantic bleaching. According to the Bracket Erasure Convention (Pesetsky 1979, Kiparsky 1982), separate morphemes in the input are fused into larger units at the end of each cycle. Morphosyntactic features are combined via unification (Sells 1995, Nordlinger 1997). In this sense, all agglutinating languages are thus fusional in the output of the Lexical Phonology. dıì is derived from /de-we-wìd/, an optative-inceptive, in this manner.

Thus, bracket erasure generates new morphemes in the course of the derivation. /dıì/ in (20b) combines all of the phonological and morphosyntactic features of /de-we-wìd/ in (20a). Analogy happens when certain derived morphemes as in (20b) take on a life of their own: they are bleached of some of their original semantic content, and spread to paradigms / verb classes in which they did not occur previously. Innovative speakers use dıì in all 1st person dual forms, not just optatives or inceptives. For these speakers, then, the lexical entry for /dıì/ is simpler, as in (21).

Once a morpheme has been semantically bleached, as in (21), it is not difficult to see how it would spread to other paradigms: /dıì/ is now just a 1st person dual prefix, so there is no reason why it shouldn’t occur wherever wìd formerly occurred.

Finally, a puzzle: when semantically bleached dıì ‘1st person dual’ replaces other level 3 morphemes such as ù ‘1st person dual optative’, whì ‘1st person dual accomplishment perfective’ and ì ‘1st person dual activity perfective’, in effect, a less-specific form is replacing a number of more-specific forms, contrary to Panini’s Principle, i.e. the Elsewhere Condition. I suggest this is a question of markedness versus faithfulness: dıì is the unmarked form, which replaces the other more faithful, but also more marked forms. I leave it to future work to formalize the exact markedness and faithfulness constraints involved.

Back to Morphology contents ↑Back to Research Interests contents ↑

·Prosodic Morphology: How do new morphemes arise?

All Dene languages have, historically, shown some tendency towards suprasegmentalization (Leer 1999, 2001)—that is, segments or segmental features such as the glottal stop and nasal consonants become floating features tone or nasality, respectively. However, the NE Dene languages, and Weledeh in particular, have taken this process a step further, so that now many prosodic features have been, or are in the process or being, morphologized. That is, tonal, moraic, and/or stress patterns have themselves become affixes, associated directly with morphosyntactic features.

When we think about it, tonal Dene languages are typologically unusual, in that tone seems to be always lexically pre-associated to the segmental portion of some affix: every morpheme comes lexically pre-specified as either bearing High tone or Low tone (or perhaps neither). As Dene languages are usually described, tone seems to behave more like a segmental property (akin to vowel height or nasality) than a true suprasegmental feature. While I believe this view is oversimplified, i.e. tone does play an active role in NE Dene morphophonemics (see here), it does have some basis: tone in Dene languages evolved fairly recently from glottal constriction, which in turn came from glottal stops (Leer 1999, 2001). Since glottal stops are segments, it should not surprise us if tone still behaves more like a segment, for some purposes.

Contrast this situation with the way tone is described for many African languages, where one speaks of “tone mapping” (Zoll 2003). That is, tone is usually some sort of word-level melody, such as HL, LH, or HLH, often associated with some morphosyntactic category (such as past tense or fuure tense), which maps onto the segmental portion of words, either from left-to-right or right-to-left, depending on how many syllables it has. So while on the one hand it is unusual that, in Dene languages, there is generally no “mapping” to speak of, on the other hand it should not surprise us that certain prosodic patterns or tone melodies are now being morphologized in Weledeh—this development actually makes the languages more typologically normal.

Consider, for example, the (Light-Heavy) prosodic morpheme that occurs in the perfective forms of d/l-classifier (i.e. middle voice) verbs in Weledeh (Jaker 2012: 555), as shown in (22).

I have proposed that this prosodic morpheme maps onto segmental material using the formal device of melodic overwriting (Ussishkin 2000, 2005a, 2005b), as shown in (23).

While this model is good for describing mapping relationships where affixal melodies already exist (and faithfulness to them is high-ranked), it does not address the question of how such affixal melodies arise in the first place. For the (Light-Heavy) prosodic morpheme described above, the answer is intuitively obvious, from a historical perspective: in what I call “Proto-Mackenzian” (the most recent common ancestor of Dogrib and Slave), there was an iambic lengthening process that turned (Light-Light) iambic feet into (Light-Heavy) feet, under certain conditions (see here). Subsequently, this process became opaque and unlearnable, and was morphologized, i.e. re-interpreted as a (Light-Heavy) prosodic affix.

But what are the formal mechanisms by which this re-analysis occurs? Let us work backwards. In all cases I know of, the affixal melody creates a prosodic configuration which either did or still does arise transparently, by regular phonological processes, in some environments. That is, the prosodic affix must have been created redundantly by the grammar, at a time when its effects could still be derived by regular phonological rules. It is only later when this affix spreads beyond its original domain (or the prosodic configurations with which it is associated can no longer be explained by the synchronic phonology) that we become aware of its presence.

To explain how such prosodic morphemes can arise redundantly, I have argued that we need a morpheme-based faithfulness (see here). That is, in cases where the affixal melody is identical to material already contained in the base, then the output material corresponds to both the affixal and base material simultaneously. Under these circumstances, if a certain prosodic pattern (say, a (Light-Heavy) iambic foot) arises by a regular phonological process in some environment, it incurs no cost to the grammar (in terms of faithfulness) to re-interpret it as a morpheme.

But why would the prosodic pattern be re-interpreted in this way, i.e. what advantage is there? Very briefly, I propose that this has to do with the interaction between Bracket Erasure (Pesetsky 1979) and NoAllomorphy. As the derivation proceeds (in successive strata) and more and more brackets are erased, larger and larger morphemes are created, which carry larger sets of morphosyntactic features (see Syncretism and Analogy). Now here’s the twist: I do not equate bracket erasure with tier conflation (McCarthy 1986)—actually quite the opposite. I propose that bracket erasure merges morphemes within a given linear (i.e. horizontal) domain, but does not affect vertical association lines. Thus, a configuration such as in the output of (23) will be unaffected by bracket erasure. This means that we can avoid extra NoAllomorphy violations, by creating new morphemes on different autosegmental tiers. This predicts that all agglutinating languages have a natural tendency towards suprasegmentalization: as segmental morphemes fuse horizontally, new morphemes will be elaborated vertically, under the right conditions.

Thus to conclude, the morphologization of suprasegmental features is predicted by the Lexical Phonology model, Bracket Erasure is taken seriously. I hope to formalize and explain these intuitions in greater detail in future papers.

Back to Morphology contents ↑Back to Research Interests contents ↑


1. The traditional assumption in the Athabaskan literature is that affricates pattern as stops, which are distinguished from other stops by having additional place features (e.g. Shaw 1991). Here, on the other hand, I follow the more mainstream assumption that affricates are both [stop] and [continuant]. Return to originating paragraph ↑

Back to Research Interests contents ↑


Ackroyd, Lynda. 1982. Dogrib Grammar. Manuscript, University of Toronto.
Anderson, Stephen. 1992. A-Morphous Morphology. Cambridge University Press.
Baker, Mark. 1996. The Polysynthesis Parameter. Oxford University Press.
Beddor, Patrice. 1984. Formant Integration and the Perception of Vowel Height. Haskins Laboratories: Status Report on Speech Research 77/78: 107-120.
Blevins, Juliette. 1994. A Place for Lateral in the Feature Geometry. Journal of Linguistics 30 (2): 301-348.
Bresnan, Joan. 2001. Lexical-Functional Syntax. Blackwell Publishers.
Brown, Dunstan & Andrew Hippisley. 2012. Network Morphology: a defaults-based theory of word structure. Cambridge University Press.
Carstairs-McCarthy, Andrew. 1992. Current Morphology. Routledge.
Carstairs-McCarthy, Andrew. 1994. Inflection classes, gender and the Principle of Contrast. Language 70: 737-788.
Chistovich, L. & V. Lublinskaya. 1979. ‘Centers of gravity’ and spectral peaks as the determinants of vowel quality. In B. Lindblom & S. Ohman (eds.) Frontiers of Speech Communication Research. New York: Academic Press.
Chomsky, Noam. 1964. Current Issues in Linguistic Theory. In Fodor, Jerry & Jerrold Katz (eds.) The Structure of Language: Readings in the Philosophy of Language. Englewood Cliffs, New Jersey: Prentice-Hall, Inc.
Christdas, Prathima. 1988. The Phonology and Morphology of Tamil. Doctoral Dissertation, Cornell University.
Clements, George. 1985. The Geometry of Phonological Features. Phonology Yearbook 2: 225-252.
Dalrymple, Mary. 2001. Lexical Functional Grammar. Syntax and Semantics vol. 34. Academic Press
Fant, Gunnar. 1960. Acoustic Theory of Speech Production. The Hague: Mouton.
Flemming, Edward. 1996. Auditory Representations in Phonology. Doctoral Dissertation, UCLA.
Flemming, Edward. 2001. Scalar and categorical phenomena in a unified model of phonetics and phonology. Phonology 18: 7-44.
Gil, David. 2001. Escaping Eurocentrism: fieldwork as a process of unlearning. In Paul Newman & Martha Ratliff (eds.) Linguistic Fieldwork. Cambridge University Press.
Halpern, Aaron. 1992. Topics in the placement and morphology of clitics. Doctoral dissertation, Stanford University.
Hargus, Sharon. 1988. The Lexical Phonology of Sekani. Doctoral dissertation, University of California, Los Angeles.
Hayes, Bruce. 1995. Metrical Stress Theory: Principles and Case Studies. University of Chicago Press.
Hyde, Brett. 1999. Overlapping Feet in Polish. In Dziwirek, Katarzyna, Herbert Coats & Cynthia M. Vakareliyska (eds.) Annual Workshop on Formal Approaches to Slavic Linguistics vol. 7: The Seattle Meeting 1998: 185-203. Ann Arbor, Michigan Slavic Publications.
House, A. & Kenneth Stevens. 1956. Analog studies of the nasalization of vowels. Journal of Speech and Hearing Disorders 21: 218-232.
Inkelas, Sharon. 1997. The theoretical status of morphologically conditioned phonology: a case study of dominance effects. In Geert Booij & Jaap van Marle (eds.) Yearbook of Morphology: 121-156.
Inkelas, Sharon, C. Orhan Orgun & Cheryl Zoll. 1997. The Implications of Lexical Exceptions for the Nature of Grammar. In Iggy Roca (ed.) Derivations and Constraints in Phonology. Oxford University Press.
Jaker, Alessandro. 2012. Prosodic Reversal in Dogrib (Weledeh Dialect). Doctoral dissertation, Stanford University.
Kaisse, Ellen & Patricia Shaw. 1985. On the theory of Lexical Phonology. Phonology Yearbook 2: 1-30.
Kari, James. 1992. Some concepts in Ahtna Athabaskan word formation. In Mark Aronoff (ed.) Morphology Now. Albany: State University of New York Press.
Kiparsky, Paul. 1982. Lexical morphology and phonology. In I.S. Yang (ed.) Linguistics in the morning calm. Hanshin Publishers, Seoul.
Kiparsky, Paul. 1985. Some consequences of Lexical Phonology. Phonology Yearbook 2: 85-138.
Kiparsky, Paul. 2000. Opacity and cyclicity. The Linguistic Review 17: 351-365.
Kiparsky, Paul. 2004. Blocking and periphrasis in inflectional paradigms. In Geert Booij and Jaap van Marle (eds.) Yearbook of Morphology 2004: 113-135.
Krakow, Rena & Marie Huffman. 1993. Instruments and Techniques for Investigating Nasalization and Velopharyngeal Function in the Laboratory: An Introduction. In Krakow & Huffman (eds.) Phonetics and Phonology 5: Nasals, Nasalization, and the Velum. Academic Press.
Ladefoged, Peter. 2003. Phonetic Data Analysis: An Introduction to Fieldwork and Instrumental Techniques. Blackwell Publishing.
Lehmann, Christian. 1995. Thoughts on Grammaticalization. Lingcom Europa.
Leer, Jeff. 1999. Tonogenesis in Athabaskan. Cross-linguistic Studies of Tonal Phenomena: Tonogenesis, Typology, and Related Topics ed. by S. Kaji, vol. 1: 37-66. Tokyo: Institute of the Study of the Languages and Cultures of Asia and Africa, Tokyo University of Foreign Studies.
Leer, Jeff. 2001. Shift of tone markedness in Northern Tlingit and Southern Athabaskan. Cross-linguistic Studies of Tonal Phenomena: Tonogenesis, Japanese Accentology, and Other Topics ed. by S. Kaji, vol. 2: 61-86. Tokyo: Institute of the Study of the Languages and Cultures of Asia and Africa, Tokyo University of Foreign Studies.
Lombardi, Linda. 1990. The Nonlinear Organization of the Affricate. Natural Language and Linguistic Theory 8: 375-425.
Lubowicz, Anna. 2003. Contrast Preservation in Phonological Mappings. Doctoral Dissertation, UMass Amherst.
McCarthy, John. 1986. OCP Effects: Gemination and Antigemination. Linguistic Inquiry 17 (2): 207-263.
McCarthy, John. 2008. Doing Optimality Theory. Blackwell Publishers.
Meillet, Antoine. 1915. Introduction à l’étude comparative des langues indo-européennes. Paris: Hachette.
Mellander, Evan. 2003. (H L)-creating processes in a theory of foot structure. The Linguistic Review 20: 243-280.
Nichols, Johanna. 1986. Head-Marking and Dependent-Marking Grammar. Language 62 (1): 56-119.
Nordlinger, Rachel. 1997. Constructive Case: Evidence from Australian Languages. CSLI Publications, Stanford, California.
Padgett, Jaye & Máire Ní Chiosáin. 1993. Inherent VPlace. Manuscript, Linguistics Research Center, UC Santa Cruz.
Parker, Steve. 2001. Non-optimal onsets in Chamicuro: an inventory maximised in coda position. Phonology 18: 361-386.
Pesetsky, David. 1979. Russian morphology and lexical theory. Manuscript, MIT.
Prince, Alan & Paul Smolensky. 1993. Optimality Theory: Constraint Interaction in Generative Grammar. Manuscript, Rutgers University.
Rice, Keren. 1982. On stems and Lexical Phonology: an Athapaskan example. Handout presented at the Canadian Linguistics Association.
Rice, Keren. 1989. A Grammar of Slave. Mouton de Gruyter.
Rice, Keren. 1990. Prosodic Constituency in Hare (Athapaskan): Evidence for the Foot. Lingua 82: 201-244.
Rice, Keren. 2000. Morpheme Order and Semantic Scope. Cambridge University Press.
Sagey, Elizabeth. 1986. The representation of features and relations in non-linear phonology. Doctoral Dissertation, MIT.
Sapir, Edward. Language. Harvest Books, 1921.
Sells, Peter. 1995. Korean and Japanese Morphology from a Lexical Perspective. Linguistic Inquiry 26 (2): 227-325.
Shaw, Patricia. 1991. Consonant harmony systems: the special status of coronal harmony. In Paradis, C. & Prunet, J.-F. (eds.) The special status of coronals: internal and external evidence (Phonetics & Phonology 2): 125-157. Academic Press.
Spencer, Andrew. 2004. Morphology – an overview of central concepts. In Sadler & Spencer (eds). Projecting Morphology. CSLI Publications.
Stevens, Kenneth N. 1998. Acoustic Phonetics. MIT Press.
Stonham, John. 1994. Combinatorial Morphology. Amsterdam: John Benjamins.
Stump, Gregory. 2001. Inflectional Morphology: a theory of paradigm structure. Cambridge University Press.
Ussishkin, Adam. 2000. The emergence of fixed prosody. Doctoral Dissertation, UC Santa Cruz.
Ussishkin, Adam. 2005a. A fixed prosodic theory of nonconcatenative templatic morphology. Natural Language and Linguistic Theory 23: 169-218.
Ussishkin, Adam. 2005b. Morpheme position. In Paul de Lacy (ed.) The Cambridge Handbook of Phonology. Cambridge University Press.
Wasow, Thomas & Jennifer Arnold. 2005. Intuitions in linguistic argumentation. Lingua 115: 1481-1496.
Zoll, Cheryl. 2003. Optimal Tone Mapping. Linguistic Inquiry 34 (2): 225-268.

Back to Research Interests contents ↑

Home | Languages | Language Documentation and Description | Language Revitalization |
Research Interests | Teaching | Papers | Presentations | Personal | CV | Contact