Emmon Bach
UMass(Amherst) / SOAS
Copenhagen Linguistic Circle
21 October 2008
© Emmon Bach 2009. All rights reserved.

Opening

1. The Central Question

The term "morphosemantics" in the title of this talk is intended to raise a fundamental question about linguistic expressions and their meanings. When we talk about the meanings of morphemes and their combination into words should we expect to find the same kinds of meanings and combinations of meanings that we associate with the processes of putting together words into phrases? The answers to this question vary widely or even wildly across different linguists and their schools or theories. For example, some linguists would say: "Decidedly not, since morphemes don't have any meanings at all. Meaning only begins with words and their combinations." Others would say: "Yes, of course! Complex words and their meanings are built up in syntax, so we predict that their complex meanings are made up uniformly across words and syntactic phrases."

So the broad general question I want to address here is this:

Are word-internal and word-external meanings the same?

This question requires that we take up a number of preliminary questions about languages and grammars, about meaning, and about words. Various answers to the question also lead to basic issues about language diversity and typology. I will be leaning heavily on evidence from North American languages that enjoy rich systems of word-formation, especially languages of the Pacific Northwest. And I will take time to fill in some background on this linguistic area.

2. Synthesis: poly- and other

There are spectacular differences among language in how much complexity is accorded to words. This dimension of diversity is the source for some of the oldest typologies of language:

• analytic - synthetic - polysynthetic
• isolating - agglutinative - fusional

and the like. Popular books on language as well as elementary textbooks of linguistics like to cite £100 words from languages like Nuuchahnulth (Nootka) and contrast them with languages like what we might call "mythical Chinese." English will do as well as Chinese for contrasts like these, citing a Northern Wakashan language Haisla (that I will return to):

1. Lanis tlakemliselasuʼina.
[lànis ƛàkemliselàsuʼina]
2. Someone is paddling along in front of our (village) beach.

(2) is a fair translation of (1), I'll unpack (1) later. (1) expresses in two (typographical, phonological, morphological -- see below) words what the English rendering takes nine to do. Sometimes the dimension we are talking about here is described as a difference in the number of "ideas" or "concepts" that can be expressed in a single word (Sapir 1921). Talking about concepts or ideas is not quite right, as I am sure Sapir would have agreed. Here's a simple root again from one of Sapir's favourite languages, Nuuchahnulth (Nootka):

3. √yah̩ʷ- `to shout in a prescribed manner in the woods as a daily morning practice for a mother of twins for a year after their birth'

Another nice meaning for a simple word from Kwakw'ala (Boas ms.: 231):

4. nəxá `to be pleased at having attained one's end by plaguing someone.'

Or again another simple root from Haisla:
5. √k̕elt- `reluctant to go out (of harbour etc.) because of the weather.'

We don't actually have to go far afield to find examples of this sort of conceptual complexity. Think of the English verb to bean; it requires a contextual reference point in the game of baseball, and means to hit a batter (what's a batter?) on the head with a pitched ball (what's that?) intentionally (?). Quite a few concepts there!

So the synthesis dimension really has to do with the complexity of the formal manifestations or reflexes of concepts in "words." The scare quotes here anticipate the hard question: What counts as a word? We'll get to that a bit later.

3. Basic Assumptions about Languages and Grammars

In general, I follow the programme of formal grammar. A formal grammar is an explicit linking of the expressions and meanings of a language. The particular form of grammar that I assume is roughly that laid out in Richard Montague's paper "Universal Grammar" (1970), with some particular choices as to implementation. A language is described by a recursive specification of the signs of a language, starting with a lexicon as a listing of the basic signs of the language and a system of rules that define how the expressions are to be combined (recursively) into complex expressions. Each sign is a k-tuple of items specifying the phonology, phonetics, morphology, syntax and semantics of the expressions. The individual elements in the sign correspond to levels of representation and interpretation. In so far as possible the principles of combination conform to a requirement of compositionality:

Compositionality: the form and meaning of a complex expression is a function of the form and meaning of its parts and the manner of their combination.

An important point that I wish to stress is that the division drawn here between grammar and lexicon is not the same as the division between syntax and morphology:

Grammar v Lexicon ≠ Syntax v Morphology

The reason for this inequality is that the lexicon can contain items that are both smaller than and greater than the word. And the grammar can affect and combine words, phrases, and discontinuous sequences of items (compare Sadock's (1991) autolexical model).

We'll come back to this point in a moment, but first we need to take up the problem of different meanings for "word."

4. What is a word?

There are several different meanings for the word "word" (di Sciullo and Williams 1987 and many other sources)

These need to be distinguished:

1. phonological word;

Phonological words are part of a phonological hierarchy that reflects the organization of sounds in a language. Other terms in this hierarchy include: segment, onset, coda, syllable, foot, phonological phrase. Languages differ a good deal in how such hierarchies are defined and how they are interpreted.

Example: Underlined and bracketed off pieces are:

[Sally's a]   [friend of mine]
[sǽlizə fréndəmayn]
compare:

Helena anodyne
[hɛ́lənə ǽnədayn]


2. morphological word;

Morphological analyses of words use labels like these: Root, stem, affix, infix, etc. These are parallel to the bar-levels of X-bar syntax, perhaps. In addition categorial assignments are made. It is an open question whether these are the same as those of syntax proper -- N, V, etc -- or different, and if different, how they are related to the syntactic categories.

3. syntactic word;

A syntactic word is a terminal element in the syntax. This notion is heavily theory dependent so that its significance depends on complementary ideas about morphology, considered as theories about the word. One line (for example, di Sciullo and Williams 1987) has it that syntactic words are atoms and hence their internal structure and characteristics are invisible to the syntax, The principal cheats here arise by feature percolations: the syntax may not be able to look at a case morpheme, for example, in a syntactic word, but the relevant information can be conveyed by feature values. But in any case, information about morphological allomorphy cannot be available to the syntax.

4. lexical word (lexeme, "listeme").

As indicated already, I take a lexeme to be the basis for the recursive specification of form and meaning in a formal grammar. The lexeme is the point at which the famous Saussurean arbitrariness of the linguistic sign is most evident. This point is emphasized in the term "listeme."

Now, various writers and theories or frameworks differ in what kinds of items are supposed to be listed in the lexicon, for example in whether items that are smaller than words should be listed in the lexicon. On the face of it, it seems that some items that are larger than words need to be listed. Some examples will make this clear. In each case I underline what I take to be a lexical item or lexeme:

English:
6. I will no longer put up with your behaviour!
7. They looked up the word in the dictionary.
8. They looked the word up in the dictionary.
9. Please don't crack wise in this context!
10. take ... to task
11. it beats the shit out of ... `it baffles ...'
12. keep tabs on ...
13. get ... 's goat
14. x hold x's tongue
15. x hold x's breath

Two kinds of word-internal elements

It follows from our general picture of a language that there will be two kinds of word-internal elements, operations, relations, and so on. On the one hand we have those items (to choose a neutral term) that are part of the recursive specification of how to make complex expressions starting from a base of lexemes, and on the other those items that are used for making or analyzing lexemes themselves. This is the wellknown but perhaps not always clearly understood difference between inflectional or grammatical items and derivational items and processes (including compounding). The first type are part of a synchronic grammar, the second are diachronic in nature. This picture is implicit in Montague's general theory, it has been most forcefully enunciated early on by David Dowty (1988, 1989).

Today, we will be looking almost exclusively at derivational processes.

Now we can state a central claim about languages:

The semantics of grammar is compositional, the semantics of derivation is not.

We need to understand how this is meant and how it is not meant.

5. Meanings of Meaning

Generally, I follow the programme of model-theoretic semantics, which assumes that a central task of linguistic semantics is to assign (sets of) semantic values or denotations to the expressions of a language as laid out by formal or generative grammar. In line with the outline just given it also assigns possible phonetic interpretations to these expression as well as other elements that might be considered components of meaning in the pretheoretical sense: implicatures, presuppositions, indication of sociolinguistic level and so on. (Why "sets of semantic values"? Because natural languages are in general ambiguous.)

The very idea of an explicit or formal or generative grammar has been under attack in recent years (as it was from the very beginning). I consider most of these critiques to be best answered by providing a formal theory to accomodate these criticisms to the extent that they are valid, thus providing better still formal theories. I believe that a great deal has been learned in the decades since this programme was initiated by Montague, Davidson, and others.

Note that the programme of model-theoretic semantics, or more broadly named formal semantics presupposes a base of formal grammar. (Chomsky's thesis: a natural language can be characterized as a formal system. Montague's thesis: a natural language can be characterized as an interpreted formal system.)

Here are some examples to make this a bit more concrete, stripped to essentials:

To interpret simple natural languages we need at least this much machinery: a set of individals (entities), a set of truth values, a set of situations or worlds, functions built up out these elements, a set of assignments of values to variables.

16. Guinevere is unhappy.
17. Guinevere loves Lancelot.
18. Guinevere loves Arthur.
19. Guinevere might not love Lancelot.
20. Every knight loves Guinevere.

Expressions are evaluated relative to a world. (16) is true relative to a world if and only if Guinevere is in the set of unhappy entities in that world. (I ignore the complications introduced by tenses.) Similarly (17) and (18) are true iff Guinevere is in the set of Lancelot-lovers and the set of Arthur-lovers. Finally (19) is true in this world iff there is another world -- way of being -- in which Guinevere does not love Lancelot.

Assignments of values to variables come in if we adopt a standard story about quantification: (20) is going to be true iff the open sentence x loves Guinevere is true on every assignment of values to x, where x is a knight.

6. Some Languages
1. Haisla

Pacific Northwest Language Area

2. Basics on Haisla: notes from a short dictionary

   So let's take a quick look at the formal elements involved in our initial example: the Haisla sentence, provided to me by the late Sampson Ross in Kitamaat Village, British Columbia. [Remark]

   21. Lanis tlakemliselasu'ina. (=1)
   [lànis ƛàkemliselàsuʼina]
   `somebody's paddling along in front of our village'
   more literally: `we're (incl.) being paddled down along the beach in front of'
      la- aux(with_ʼin(a))
       -nis -1pl-incl
      √tlak- paddle-
      -em -front
      -l -go
      -=is -on_beach
      -el(a) -continuative
      -su -passive
      -'in -a (with la- perfective/progressive)
   22. 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.'
   (From a Haisla text by Jeffrey L Legaic^†))

3. Some derived words in Haisla:

Haisla open class words are all based on a root or extended root followed by one or more derivational (`lexical,' `semantic') suffixes. Here are some typical examples (√marks roots, =, !, - for voicing, glottalizing, plain suffixes):

23. √ʼik- `good'
    ʼíxsdu `bright, white colored' -sdu `colour'
    ʼíxsduqʷia `bald-eagle' -qi(a) `head' cf. híxtʼi `head'
    ʼíxp̕a `tasty' -p̕a `smell'
    ʼíxp̕ala `smelling good'-p̓ala `sensing, tasting'
24. √lhaxʷ- `hurt, love, strain'
Lincoln and Rath (1986) gloss the root like this: `strong, valuable, painful, love, difficult, tight'    lhaxʷela tight, fit tightly, stormy, rough water, firmly fastened, wound up tight
   lhaxʷmala ache, be in pain, hurt, suffer
   lhaxʷmecʼuaqiala (have) headache (many items for hurting or aching in a particular part of the body)
   lhaˈwʼinaiˈ hard, difficult, close friend, like, care, love
   lhawʼinaiˈnukʷ love (someone), like
   lhaˈwʼenem, lhilaˈwʼenem husband
   lhaˈwad have husband
25. -!ex̄d `arse, foundation, basis'
    gélt̓ex̄d `tall' √gelt- `long'
    démk̓ʷex̄ `kick in the arse'
    k̓ʷenc̓ex̄d `lazy'
    q̓emc̓ex̄d `have a wrinkled bum' [story]

4. Nuuchahnulth (aka Nootka Southern Wakashan )
(A famous example repeated in many publications, from Sapir's Language (1921: 141):
26. ʔinikʷiłm̓inih̩ʔisita
`several small fires were burning in the house'
analysis: (forms from Sapir and Swadesh, 1939)

    ʔink(ʷ)- `fire; burning'
   `-ił `in the house, on the floor'
  2 -m̓inh̩ plural
   -ʔis, -ʔic- diminutive
   -(m)it past
   -ma / -a (a: with past -(m)it)
[some details omitted EB]

5. Yup'ik (Eskimo-Aleut = Yuppik, in the practical orthography C' marks a geminate consonant: CC)
27. Angyarpaliciqsugnarquq.
[ángyárpalíciqsúgnarquq]
S/He probably will make a big boat.
analysis
    angyaq- boat-
   --rpak -big
   --li -make
   +ciqe -will
   ~+yugnarqe -probably
   --uq -3sgInd

(The markings on the beginnings of the affixes signal various special morphonological effects and allomorphy), for example "--" marks affixes which delete the last segment of their hosts, "~" : delete final e of host, and so on.)
6. English as a Wakashan language!

Part of the value of studying a wide variety of languages is that they often throw light on each other. It is rare that a certain kind of formal or semantic feature does not find counterparts in other languages that seem at first sight to be very different. In this section I will draw attention to some global ways in which English is very different from languages like Haisla, but then take a look at English through Wakashan glasses, so to speak, and mention some derivational items that would seem very familiar to a Wakashan linguist.

It should first be noted that Haisla and other Wakashan languages are strikingly different from English, in that they show no examples of "compounding" in the strict sense: putting together roots or stems in derivational ways. In this way, Wakashan lines up with the Eskimo-Aleut family, and is opposed to other groups in the area, such as Salishan, Tsimshianic. This fact has led some linguists to look to compounding as a source for the extensive lexical affixes of Wakashan. Whether that historical story is right or not, it is a fact that in Wakashan, the lexical affixes in general show no formal resemblance to independent words that might overlap semantically with the affix: compare the `head' affix -qia and the independent word hixt̓i `head.' It is in fact misleading to identify the affix semantically with the item meaning head, as the affix generally has an adverbial force meaning something like `on the head' or the like.

That said, it is evident that many of the derived words of Wakashan languages can be matched most straightforwardly by English compounds. So we can easily reproduce the semantic structure of the Haisla word for baldheaded eagle by means of a compound like whitehead or (more exactly still) brightcolorhead (where we have to understand brightcolor as already having the conventionalized meaning `white'), and this would even be a common pattern for forming nouns for species of animals, birds, kinds of people, etc.: bluefin, redbreast, graybeard, paleface. What is different is that the Haisla word is built not by compounding (putting together two or more `free' forms) but by affixing a bound morpheme. So we might ask: does Haisla have something that English doesn't? There is a tradition that seems to say Yes to this question, namely Haisla (Wakashan, other families like Salishan) has things called variously `semantic' or `lexical' suffixes. More on this below. But first, some English things that are somewhat but not exactly like these suffixes.

English (like many languages) shows special forms of morphemes that occur only in close combination inside words. Most of these are learned Latino-Graeco [!] morphemes, but there are some perfectly ordinary native ones as well.

28. -mən, -lənd, -səl as in postman, Finland, mainsail (only for sealubbers)
But the majority of these special combining forms occur in learned or `foreign' items (like Graeco-): we have series like:
29. syntax, syntact-; syntactic, syntactico- (semantic) etc.
Somewhat closer are items that occur only as combining forms. We can coin a word like leucocephalic (`white-headed'), but there are no free forms like leuk(os) or cephal(os). Japanese also has lots of Sino-Japenese morphemes that occur as such only in combination:
30. Japanese: den `electricity'; niti `day' (denwa, densha; nitiyoobi etc.) [gratia M. Takahashi] (den is like English e- in email, etc.)
Two things are different about these bound forms of English (and Japanese) as against the Haisla items: (1) in the former case the items seem more like roots or stems, whereas in Haisla there is no doubt about the assignment of the items to the class of suffixes (note that the English items can often occur as first members or later members of words); (2) the learned English (and Japanese) items seem to be marked off as participants in a special subsystem of the language (often signalled in linguistic discussions and analyses by features like Foreign, -Yamato, etc.) with special things to say about their combinatory potential and phonology. To approach the Haisla eagle word we would have to have some monstrosity like white-cephal or imagine that pate say could only occur as a suffix and make up a word like white-pate. STILL and all: English does have some genuine affixes with meanings that are quite comparable to the meanings of particular Haisla lexical suffixes. A few examples:
31. -ade: orangeade, lemonade, limeade, grapefruitade ?
Compare items like Haisla -bes meaning `liquid from or small particles of' (snow, milk, etc.), Kw -asdi `dried meat of' etc. (Names for foods seem to excite the energies of Amerenglish wordsmiths: -burger, -furter, -wich etc.).
32. -(a)teria: cafeteria, washateria `place or establishment for'
This is quite comparable to Ha and Kw suffixes like =as, =ilas `place (etc.) for.'
33. -gate: Watergate, Whitewatergate, Irangate, skategate (= Tonyagate), Cherie-gate
34. -er: not just agentive -er but also -er for tool names, person from, birdnames as in baker, thresher, thrasher, warbler, New Yorker
35. -oid: humanoid, android, factoid etc.
36. -ee: payee, employee, promisee, etc.
Finally, English has at least one "phrasal" derivational suffix
37. -ish X-ish: somewhat X, somewhat like an X greenish, boyish, Monday morning-ish
English seems to be poor or entirely lacking in `adverbial' affixes like Haisla =is `on beach' =ilh `in house' but particles and prepositions in English seem to function in somewhat the same way, with idiomatic combinations acting as lexical derivatives.

7. Conclusions: typology, parameters, and Sapir

Look back at the traditional typology that I cited early on in this presentation:

1. typology

• analytic - synthetic - polysynthetic
• isolating - agglutinative - fusional

The traditional typology that we cited early in this essay is based exclusively on properties of words as phonological or morphological entities. Another dimension comes in from the perspective of this essay.

2. parameters

There was a time when it was hoped or claimed that differences among languages were to be explained on the basis of "parameters." For example, there was supposed to be a "null-subject" parameter. If a language tolerated sentences with no overt subject then a whole cluster of other properties was supposed to follow. This approach was characteristic of the (aptly-named) "Principles and Parameters" model of Chomsky and his followers, of roughly the eighties of the last century. (A few leading references: Chomsky 1982, Baker 1996, Chomsky 1995, for critique: Culicover and Jackendoff 2005.)

Variation within languages approximates variation across languages (Bach 1974).

An easy example in this context is afforded by languages that allow null subjects but only in certain circumstances (person, tense etc.) such as Modern Irish or Modern Hebrew. In the realm of word-order a clear example is afforded by Modern German which shows three different patterns of word-order: SVO, VSO, SOV, each of which is characteristic of a larger number of different languages (see Bach 1974 for discussion of this point). This is, however, not the place to enter into an extended discussion. Instead I give the

3. last word to Edward Sapir

An analytic language is one that either does not combine concepts into single words at all (Chinese) or does so economically (English, French). In an analytic language the sentence is always of prime importance, the word is of minor interest. In a synthetic language (Latin, Arabic, Finnish) the concepts cluster more thickly, the words are more richly chambered but there is a tendency, on the whole, to keep the range of concrete significance in a single word down to a moderate compass. A polysynthetic language, as its name impliies, is more than ordinarily synthetic. The elaboration of the word is extreme. Concepts which we should never dream of treating in a subordinate fashion are symbolized by derivational affixes or "symbolic" changes in the radical element, while the more abstract notions, including the syntactic relations, may also be conveyed by the word. A polysynthetic language illustrates no principles that are not already exemplified in the more familiar synthetic languages. It is related to them very much as a synthetic language is related to our own analytic English. The three terms are purely quantitative -- and relative, that is, a language may be "analytic" from one standpoint, "synthetic" from another. I believe the terms are more useful in defining certain drifts than as absolute counters. Edward Sapir (1921: p. 128 [paper bound edition])

References