OSU08 Morphosemantics Notes 2

Emmon Bach, SOAS, UMass(Amherst)
OSU Linguistics
e m a i l: firstinitialsurname (at) linguist (dot) umass (dot) edu
Copyright Emmon Bach 2008. All rights reserved.
Office hrs: M W 1-3 and by appointment, Oxley 201a
Notes for classes will be posted at / linked to

"http://www.people.umass.edu/ebach/courses/osu08-pl.htm"

What's morphology? Various answers. Languages: Pacific Northwest, Wakashan.

Part A Theoretical

1. Four Models of Grammatical Description

The title of this section is adapted from the famous article by Charles F. Hockett (1954), which was first published in Word, but is probably read most often -- if at all these days -- in the reprint in Joos (1957). Hockett's paper was called "Two Models..." but he mentions a third in the early paragraphs of the paper. We will add one more which has gotten a lot of press recently.

History: Hockett's paper was widely read at the time and later and the title no doubt inspired Chomsky's 1956 paper "Three models for the description of language." (There is a reference to Hockett's paper in Chomsky's 1957 Syntactic Structures.) Hockett was not well-treated in the early days of generative grammar, despite his early openness. [story: more in my Retrospect 2008 talk in a couple of weeks.]

Model 1: Item and Arrangement (IA). The item and arrangement model of Hockett flows directly out of the procedural orientation of (American) structural linguistics, in its heyday around the middle of the last century. By procedural I mean here the emphasis on procedures of analysis: segmentation and classification. History: Zellig Harris's 1952 book Methods in Structural Linguistics (later called just Structural Linguistics) epitomizes the dominant view of the process of linguistic analysis: Harris describes this procedure as a process of segmentation and classification. A central notion was the morpheme understood as the minimal meaningful unit of language. It is probably what you first go with when you teach beginning linguistics, using word-structures as your take-off point. [Story: to be told at my "Great Moments" lecture on April 18.] Under this view, a descriptive grammar is basically a set of lists of "items" -- morphemes and their allomorphs, for example -- together with statements about their distribution. In Hockett's words: "The essence of IA is to talk simply of things and the arrangements in which those things occur." When Hockett wrote the cited paper, this way of treating linguistic structures was dominant in American linguistics. Because of the "building block" nature of the IA model, it was particularly well-suited to the analysis of agglutinative languages, and some writers (Spencer 2004) have called the general program of such approaches "Radical Agglutination." Here's Andrew Spencer on the word cats: In the American Structuralist tradition associated with such authors as Bloomfield, we would say that the forms /cat/ and /s/ are morphemes , the minimal meaningful components of a word. The full semantic characterization of the word results from combining the meaning of the parts.... This way of thinking is crucial to syntactically-based approaches deriving from the proposals of Pollock (1989), and I have called this set of assumptions Radical Agglutination (Spencer 1992). (Spencer 2004: 68.) Curiously, although one might think there was more affinity between classical generative, particularly transformational generative approaches and the Item and Process model to be discussed immediately, the IA assumptions have been dominant early and late in the generative tradition. In the generative tradition this kind of description is realized most consequentially in theories where there is no distinction between basic members of lexical classes and the morphemes that are attached to them both in derivational and in inflectional morphology (Lieber 1992).

Model 2: Item and Process (IP). The prototype for this sort of description was the kind of grammar that Sapir and his students wrote, going back ultimately to the Sanskrit grammarians. In this model basic lexical items are assigned a special form (root or base). Inflectional and derivational forms are derived by applying various processes to these roots and the further derived forms. Quite regularly the root form may be an abstract form that never exists as a surface form, but is set up to facilitate the statement of the various processes that might apply to get surface forms. Very often in a descriptive grammar and dictionary the citation form is chosen for convenience, say as the nominative singular form of a noun, or the first or third person singular or infinitive form of a verb.

Interlude: a comparison of IA and IP treatments of reduplication. The basically IA nature of the generative tradition of morphology can be easily seen in the treatment of reduplication. Reduplicative patterns are treated as items, that is morphemes, like any others except in their segmental content: CV-, CVC- and the like are examples of prefixes, just like re-, en- and so on. An IP treatment would spell out operations based on root or base shapes. One pattern of reduplication in Haisla goes like this: Given a root of the form C1VC2X, make an extension of the form C1iC1VC2H, for example: begʷanem `person' ==> bibegʷanem `people' The item and arrangement treatment (as in Marantz, 1982) would simply treat the preposed bit as a prefix Ci- (with prespecified i and then appeal to spreading of some sort to link the C to the first C of the host. An IP account would specify some formal operation corresponding to this: "copy the first consonant (or possibly onset) of the root followed by i and prefix to the root."

Model 3: Word and Paradigm (WP).

Hockett mentions the model called then and since Word and Paradigm (WP) and opines that it deserved as full a treatment as the other two models that were focussed on in his paper. In recent years, the model has come to the fore quite a bit. (Matthews 1974 and 1991 are primary references; Spencer 2004 is a good fairly recent discussion, as is Blevins 2006.) Basic to the approach is the rejection of any attempt to build up the form and meaning of words accretively, whether by agglutinative means or by applying processes to bases or roots. We will return to the WP way of looking at morphology as we go along. As to semantics, one firm assumption of WP theories is that words -- not morphemes -- have meaning, and hence would be the basis for any recursive specification of interpretations.

Model 4: Construction Grammar (ConstG) This model has been increasingly influential in the last decade. The basic idea is that a language can be best characterized as a collection of constructions. Emphasis is on idiosyncratic form and meaning combinations. Goldberg's books (1995, 2006) are basic references. The approach has its modern roots in work by Charles Fillmore and Paul Kay, and is related to Cognitive Grammar (G. Lakoff, Ronald Langacker). Formal assumptions about the model have relied (recently at least) on HPSG implementations.

Caveat emptor! Invocations of the history of our discipline often serve a justificational goal, rather than a sober historiographical purpose. In particular, when a linguist cites the evils or virtues of some previous era, follow Jim McCawley's advice about mentions of ¡¡EXPLANATION!!: put your hand on your wallet! And my advice: read the originals!

2. The Legacy of Transformational Generative Grammar

A constant of the tradition of Transformational Grammar is the idea that grammars generate or specify systems of representation: phrase-markers, logical forms, deep structures, surface structures, conceptual structures, semantic markers, or whatnot. A general theory and a grammar spell out mappings or relations among these various objects. In so far as a real semantic interpretation is included it is given on the basis of one or another of these representions, using whatever properties accrue to that type of representation, for example, C-command etc. in the case of phrase-structures. This approach may be called "configurational" and can be constrasted with another view in which the objects of the grammar are built up simultaneously. This approach has been called the "rule-to-rule" or "rule-by-rule" relation among the ojects of the grammar and interpretation (Bach 1976).

3. The Linguistic Sign

What are the objects of grammar and language? It is easy to fall into the trap of identifying an item with its phonological (or orthographic) form. But is easy to see that this won't do. In the spirit of Saussure, it is best to think of a linguistic object as a cluster of elements or properties. We will follow this route here: a lexical item is a k-tuple of elements. What k is and what the elements are is matter of theory. I will assume that a linguistic sign is minimally a quadruple: < α, φ, γ , σ > standing for an identifier, a phonological representation, a category (syntactic or possibly morphological), and a semantic specification. The identifier is strictly not necessary but convenient: think of it as an address or key in a data-base. For the identifier, I will generally use conventional orthography supplemented with numbers, in the way that is often used in linguistic discussion: read₁ vs read₂ etc. This characterization is good for lexical items. I leave open for now what if any is the status of the identifier in tuples that are built up in a grammar.

Excursus: My implementation of categorial grammar in Scheme (Lisp) uses indexed labels to carry the feature values of the grammar as an item is built up. The indexed labels then correspond to occurrences of the various categories. Relative to a derivation then the indexed labels together with the identifiers of the lexical items can be thought of as the identifiers of the k-tuples.

4. A Fifth Model?

The reason for the question mark, is that we will be looking first at a general or "universal" model, roughly that of Montague's Universal Grammar, which can be looked at as a meta-theory. [Remarks] What I take as the essential requirements on grammars instantiating this general model are these:

1. explicitness
2. simultaneous recursion on k-tuples, where k is at least = 4
(cf. above)
3. among the elements of the tuples: identifier, interpreted phonological/phonetic item, category (syntactic, morphological) including appropriate feature-values, semantic element (say, a formula in an interpreted language like Montague's Intensional Logic)
4. separation of rules and operations
Each rule specifies the input tuple(s), the output tuple, the operations to be applied to the input tuples to yield the output tuple.

To get into just what is involved in spelling out a syntactic rule in this mode or model, let's look at PTQ's rule for putting together a subject term phrase and intransitive verb phrase: incorporating the interpretation rule, roughly this:

If α ∊ P_T and β ∊ P_IV then γ₁₆(α, &beta) ∊ P_t, where γ₁₆(x,y) is the result of concatenating x with the third-person singular form of y. Let τ(z) stand for the translation of z, then &tau(γ₁₆(α,β)) = τ(α)(^τ(β)) [if I got all that right!]
In English: the denotation of the result of the rule is the application of the denotation of the subject to the intension of the denotation of the predicate.

The proper treatment of this rule in ordinary English might look like this:

Put the subject termphrase (in the nominative case) in front of the tensed (intransitive) verb phrase and let the tensed verbphrase agree with the subject in person and number.

Note that this is more general, in that Montague simplified by just treating singular terms, as number was not pertinent to his interests in developing this fragment.

Montague (if I recall correctly) did not spell out adequately what it was to make "the third person singular" of an IVP in two respects: the structural location of the verb in the IVP, the problems of conjoined IVP's (I think he gave a condition by picking out "the first verb" or the like).

5. Metatheory

We can take something like Montague's theory as laid out in his Universal Grammar (Montague 1970) as a general framework, picking and choosing for more specific purposes. (A good exposition may be found in Dowty 1979.)

Keenan and Stabler's Bare Grammar (2003) might be an alternative base to use for comparing grammars and general theories.

6. Rules and Operations

7. Morphological operations

More detail on this question as we go along. Part of the general formal questions about morphology (inflection, compounding and other derivational processes) has to do with similarities and differences between the operations of phrase grammar and word grammar. For example, reduplication is frequent in word grammar, but not unheard of in phrase grammar (Carlson 1983).

8. Feature Theory

More on this as we go along. For now, think of a feature as a function from an expression to a (set of) values. The space of morpho-syntactic features is structured by the syntactic categories. So for example the features defined for TermPhrases (= DP's or NP's in pre-Abney syntax) include (for English at least): Kase, Number, Person. The Kase values for German, for instance, = {Nom, Acc, Dat, Gen}. German has a Gender feature for nominals, etc. Feature theory must include principles of Government, Agreement, and Percolation. Compare Bach 1983 (a pre-Unification treatment) where two implementations are considered, a strictly lexicalist version and a "not-so-lexicalist" version.
9. What's Morphology?
Some theories seem to say: Nothing.
10. Where's Morphology
Some theories seem to say: Nowhere or Everywhere.

11. Categories and Types

Here are my initial assumptions about categories and types:

Categories

:

1. BCAT = {n, s, w}
basic categories;
2. Γ = {LCON, RCON, UP, LWRAP, RWRAP,...}
UP is used for category changing rules which have no affect on the phonology or orthography; i-place operations for i = 1, 2 (,...)
3. BCAT ⊆ CAT
This needs discussion (later) on the relation between lexical categories and syntactic categories;
4. if α, &beta ∊ CAT and γ is a one or two-place operation in Γ, then < α, β; γ > ∊ CAT
if there are i-place operations for i > 2 then add them here
5. that's it!

The more or less usual way to do this in categorial grammars is like this: Abbreviations: write (a/b) for < b, a; RCON >
write (b\a) for < b, a; LCON >
suppress outermost parentheses
NB: using fractional rather than argument last notation!
That is:

FA schemata:

a/b b ==> a
b b\a ==> a


ARG-last version:

a/b b ==> a
b a\b ==> a


Mapping from syntax to semantics; for a start we will take the usual one, with one possible special wrinkle:

Take the model structure as including:

1. A type: e
set of individuals or entities;
2. BOOL = {0, 1, zilch} type: t
zilch is some ugly object to extend possible partial functions to total functions;
3. W type: w
set of possible worlds (or -- here's the wrinkle -- set of situations: as in Kratzer 1989, 1995, 2007 for a general account of situations in semantics and philosophy*;
4. F: set of all functions < β, α > ... built on the above:
... intended to leave open the possibility of more than unary functions
*we need also various relations such as inclusion, temporal relations, including the possibility of following Montague in splitting times from worlds, etc.

By saying "including" I wish to leave open whether there are further primitive types of denotations, for example, Properties, as in Chierchia's work. I also want to assume that there can be structures and Sorts in A such as those introduced by Greg Carlson, Godehard Link, and so on. Cf. references.

We then need a mapping T from categories to types:

1. T(n) = e
2. T(s) = t
3. T < b,a; γ> = <T(b), T(a) >
or T < b,a; γ> = <T(b), ^T(a) > for intensional environments/discussions;

We'll use IL to indicate the denotations of items and Montague's priming convention for constants (or English glosses in caps when convenient).

Exercise: proof-read/copy-edit the above and send me bloopers or typos!

Part B: Languages: Introduction to the Pacific Northwest Coast Linguistic Area

These languages are really different.
Bob Levine (P.C., some time ago)

The Pacific Northwest of Northamerica is one of the richest linguistic areas of the world. Here are some maps to give an idea of the languages to be found there.

• Eskimo-Aleut: Yup'ik (Central Alaskan Yup'ik*), Aleut, Inupiaq (Eskimo) ...
• NaDene
Recently confirmed: related to the Siberian langugea Ket (= Yenisei-Ostyak), the first confirmed genetic relationship between an American language and an Old World language -- not counting Eskimo-Aleut via Siberian Yuppik).
Tlingit*
Eyak / Athabaskan: Carrier, Chilcotin, Wets'uwet'en, Tahltan,...

• Haida (isolate! recently vindicated not Na-Dene)
• Tsimshianic: Coast Tsimshian*, Nisga'a, Gitxsan
• Wakashan:
  Northern: Haisla* / Henaaksiala*, Ooweky'ala*, Heiltsuk; Kwakw'ala*;
  Southern: Nuuchahnulth (Nootka)(Ahousat*), Ditidaht (Nitinat), Makah
• Salishan: Nuxalk (Bella Coola), Lillooet, Lushootseed, Shuswap, Squamish, Halkomelem, Musqueam,...
• Kutenai (isolate)
• Chemakuan: Quileute, Chemakum (extinct)
...
*EB has some direct acquaintance with these languages, hence perhaps more to be trusted in claims about them.

In the Pacific Northwest the resemblances among languages exist at all linguistic levels: phonetics and phonology, morphology, syntax, semantics. Nevertheless, within these commonalities there are striking differences of detail.

• Sounds and sound systems:
Rich inventory of consonants:

Some favourite sounds and features:

• glottalized consonants:

  stops: pʼ tʼ kʼ qʼ ƛʼ vs p t k q ƛ (common) and b d g ḡ λ (Haisla, uncommon)
  sonorants: mʼ nʼ lʼ wʼ yʼ (scattered: Wakashan, Salishan)
  fricatives: xʼ xʼ łʼ (rare: Tlingit)

Examples: Tlingit glottalized fricatives (voice of Nora Dauenhauer, used with permission):

l - l' (ł łʼ)
x̱ʼ - xʼ x̱ʼéishxʼw `blue jay'
s - s'

• front - back velar (uvular) stops and fricatives

g k x vs. ḡ (g) q (k) x̄ (x)

• laterals:

ł (hl, lh) λ ƛ ƛʼ

• laryngeal / pharyngeal consonants:

ʔ ʕ h̩ (Nuuchahnulth)

Examples (Ahousat, Katie Fraser):

glottal stop: ʔaʔamas `cheek'
pharyngeal fricative: ʔiih̩ `big'
pharyngeal stop: ʕiiniƛ `dog'


• Interesting gaps:

no labials (m b p): Tlingit
no nasals: Nitinat (Ditidaht) - Wakashan, Lushootseed - Salishan, Quileute - Chemakuan
no voiced l only voiceless ł Tlingit no plain (vl) ƛ only glottalized ƛʼ Salishan (Lushootseed)

• long clusters of obstruents with no vowels: Nuxalk (Bella Coola), Ooweky'ala

Example: Ooweky'ala (voice of the late Hilda Smith, courtesy of Darin Howe, University of Calgary):

clusters: th'pkv, th'pkvki, th'pxvkski `closed' (several inflected forms):
(ƛʼpkʷ - ƛʼpkʷki - ƛʼpxʷkski )

• Word grammar:
• polysynthesis (in classic senses)
  
(1) single word sentences, (2) words composed of relatively many pieces and processes

Examples:
Central Alaskan Yup'ik (Jacobson, 1984, see Notes 1):

angyaliciqsugnarquqallu `he/she will probably make a boat also'
(for analysis see Notes-1)
nanvarpagtengnaqngaicugnarquq he/she probably won't try to go the big lake

Nuuchanulth: (Nuu-chah-nulth) (Nootka)(from Sapir's Language (1921: 141-2, see Notes 1):
ʔinikʷiłm̓inih̗ʔisita `several small fires were burning in the house.'

Excursus: on hyphens (cf SSILA bulletin a few years ago).

Haisla (from a text by Jeffrey L Legaic):
Kʼaˈqex̄damʼuasuxʼilapin ʼEbuˈkʷs Tlalemqʷʼaˈx̄s his tlʼaˈsiagʷemikax̄i.
The wolf put its head on the knees of ʼEbuˈkʷs Tlalemqʷʼaˈx̄s
kʼaˈqex̄damʼuasuxʼilapin
√kʼaq- touch with head
-x̄damʼua -on knees
-su -passive
-[x]ʼilap -momentaneous
-in -then/now/perfective
ʼE. Tl. Personal name (= `mother of Tl')
his (= -s) oblique marker: here agentive/instrumental
tlʼaˈsiagʷemik `wolf' [Story: etymological knowledge.]
-a Primary deictic suffix
-x̄i Secondary deictic suffix (-ax̄i `remote visible')

So the sentence means more literally:
`ʼEbukʷs Tlalemqʷʼax̄s was head-touched-on-knees by the wolf.'


• high degree of fusional allomorphy

Examples:
Wakashan: `hardening' (glottalizing) and `softening' suffixes
Haisla lexical suffix -[k]!al(a) `make sound of X, language of X': k appears only if host ends on vowel or sonorant, ! signifies glottalization of (k or) final plain stop of host.

Yup'ik large number of combinatory effects: deleting final consonant, amalgamating final consonant of host with first consonant of suffix,...

Here's a Scheme (Lisp) function for doing part of the allomorphy in Yup'ik morphology: Yup'ik suffixation

• lexical suffixes

Wakashan lexical suffixes: `in house,' `outside on beach,' `in forest,' `(on) head,' `inside hole,' `on(to) rocks,' ... (more anon)

• rich reduplicative patterns in some languages of the area: Wakashan, Salishan, Tsimshianic
.

• Syntax (phrase grammar):
• free word order: Yup'ik
  
• verb final: NaDene, Haida
  
• verb initial: Tsimshianic, Wakashan, Salishan, Chemakuan
• lexical categories: Nouns vs Verbs (vs Adjectives)? Wakashan, Salishan, Chemakuan
• Semantics: cultural similarities reflected in vocabulary:
• some favorite derivational patterns: eating, hunting and gathering/preparing
Haisla:

saax `grizzly bear' sasen̓á `going after grizzly bear' [story: the late Gordon Robertson on how grizzlies know their names but not Haisla derivational morphology]

Coast Tsimshian:

hoon `fish, salmon' xhoon `going after fish, salmon'

[Local English: `going after' = hunting and gathering]

• shape classifications

Haisla: hená `long thing -- canoe, boat, log -- is located (somewhere)'

• number classification

Coast Tsimshian:

baa `run (sg)' kʼoɬ `run (pl)' [emmon: check!]

Remark: language particulars and language universals (Bach 2004a).

If time: focus on Haisla .

References:

For the first half of today's session, some general discussions and comparisons: Anderson 1992; Blevins 2006; Dowty 1979; Hockett 1954; Matthews 1974, 1991; Spencer 2004. For the second half: Thompson and Kinkade 1990 (excluding Eskimo-Aleut), on Yup'ik Steven A. Jacobson 1984, 1995. Generally on North American languages: Mithun 1999.