Professional Documents
Culture Documents
Semantics and Pragmatics
Semantics and Pragmatics
Introduction
Semantics is the study and representation of the meaning of every kind of constituent and
expression (from morph to discourse) in human languages, and also of the meaning
relationships among them. Twentieth century semantics, especially in the period 1960-2000,
has roots that stretch back to the Pre-Socratics of Greece in the sixth to fifth centuries BCE.
To chronicle the annual achievements in semantics and pragmatics from 1960 to 2000
would be almost unreadable and certainly unhelpful. Instead more or less chronological
threads of development will be presented within thematic areas. These are divided into four
major sections: Lexical Semantics; The Semantics~Syntax Interface; Logic and Linguistic
Meaning; and Aspects of Pragmatics. Developments in any one topic area were often
influenced by developments elsewhere, and the subdivisions within them are not discrete, as
we shall see. The chapter concludes with a short essay on the importance of scripts
(predictable personae and sequences of events) and conversational implicatures (probable
inferences that arise from conventional expectations) which allow for the underspecification
of meaning in language that renders natural languages much less explicit than most computer
programming languages.
One problem that has been raised repeatedly over the course of the period in question and
which underlies much of the discussion here is the scope of semantics (and thus the definition
of meaning). For some scholars semantics studies entities such as words in isolation, and thus
semantics concerns only the type of decontextualized, abstract information that would be
2/38
given in a dictionary. However, for others, semantics covers much more. This can be the
type of information found in an encyclopedia – a structured data-base containing exhaustive
information on many (perhaps all) branches of knowledge.1 Or, it can be the relation of the
word to other words in the co-text (the surrounding text), or the relation to the context of the
speech event in which the word is used, experiences, beliefs and prejudices about the context
in which it is used.
Lexical Semantics
Here I discuss componential analysis, semantic fields, semantic primes, prototype semantics,
stereotype semantics and lexicography.
Componential analysis is typically based on the sense of a lexeme such as can be found in
a dictionary, that is, decontextualized meaning, abstracted from innumerable occurrences (in
texts) of the lexeme or combination of lexemes. In componential analysis, this sense is
identified in terms of one or more semantic components. The principal means of
accomplishing this has been through the structuralist method of contrastive distributional
1. The place of proper names and the problematic relationship between the dictionary and encyclopedia are
examined by e.g. Allan (2001; 2006a); Hanks (1979).
2. The term ‘lexeme’ is used here because it is the word most used in the literature. Others have used
‘listeme’ for this meaning (Di Sciullo and Williams 1987; Allan 2001).
3/38
analysis. Lexemes that share semantic components are semantically related. There is no
consistent one-to-one correlation between semantic components and either the morphs or the
morphemes of any language. Being components of sense, semantic components reflect the
characteristics of typical denotata. The denotatum, ‘denotation’, of a language expression is
what it is normally used to refer to in some possible world. So, the denotata (‘denotations’)
of cat are any animals that the English word cat (and cats) can be used to refer to. There is a
hierarchy of semantic components which corresponds to perceived hierarchies among
denotata. For instance, FELINE is a semantic component of cat and entails the semantic
component ANIMAL which is also, therefore, a component of cat. This suggests a thesaurus-
like structure for semantic components. It follows that the set of semantic components for a
language can be discovered by identifying all the relationships that can be conceived of
among the denotata of lexemes. In practice, this could be everything in all worlds, actual and
non-actual, a procedure that has never been successfully achieved.
In American linguistics up to 1960 (see chapter 16), Leonard Bloomfield (see especially
1933) – the major figure of this period – was sympathetic to the cultural context of language,
but he came to exclude semantics from the Bloomfieldian tradition in American linguistics on
the grounds that semantics is not directly observable in the way that phonemes, morphemes,
and sentences are manifest in phones. So, from the 1940s until the 1960s, semantics was
regarded by many American linguists as metaphysical and unfit for the kind of scientific
enquiry into observable language structures that they believed linguistics should undertake.
The advance towards semantic analysis was therefore made within morphosyntax and by
European linguists, using as a model Roman Jakobson’s work on the general theory of case
(1936). Jakobson identified the conceptual features of each case in Russian using the
methodology of Prague School phonology (see chapter 14, Part two, and chapter 22). Thus,
according to Jakobson (1936) each Russian case is distinguished from other cases in terms of
the presence or absence of just the four features [±directedness], [±status], [±scope] (the
scope of involvement in the context of the utterance), and [±shaping] (identifying a container
or bounded area). The idea of characterizing cases in terms of distinguishing components can
be applied to the traditional analysis of the nominal suffixes in Latin so as to identify features
from the categories of case (NOMINATIVE ∨ GENITIVE ∨ DATIVE ∨ ACCUSATIVE ∨
3
ABLATIVE) , gender (MASCULINE ∨ FEMININE ∨ NEUTER), number, and declension type. For
instance ACCUSATIVE ∧ FEMININE ∧ SINGULAR ∧ FIRST_DECLENSION generates the suffix -
am as in fēminam “woman”; ABLATIVE ∧ MASCULINE ∧ SINGULAR ∧ SECOND_DECLENSION
generates the suffix –ō as in puerō “by the boy”. In ‘Componential analysis of a Hebrew
paradigm’, Zellig Harris (1948) analyzed the verb paradigm using the categories of tense,
person, and gender on a similar distributional basis to this; the result corresponds to
Jakobson’s analysis in terms of conceptual features.
It is a small step from the componential analysis of closed morphosyntactic systems like
case systems to the componential analysis of closed semantic fields like kinship systems.
Anthropologists had for many years been comparing widely differing kinship systems in
culturally distinct societies by interpreting them in terms of universal constituents (such as
FEMALE_PARENT_OF, MALE_SIBLING_OF) that equate to semantic components. Two of the
earliest articles in the componential analysis of meaning (Lounsbury 1956 and Goodenough
1956), appeared consecutively in the same issue of the journal Language and were both
analyses of kin terms. Without stepping far outside the Bloomfieldian tradition, these early
writers on componential analysis were responsible for changing contemporary linguistic
opinion by showing that semantic analysis could be carried out using approved methods of
structural analysis, similar to those used, for example, to filter out the phonetic components of
the Sanskrit stop phonemes. For instance, Floyd Lounsbury’s paper begins with a comparison
of Spanish and English kin terms: ti-o, hij-o, abuel-o, herman-o (‘uncle’, ‘son’, ‘grandfather’,
‘brother’) vs ti-a, hij-a, abuel-a, herman-a (‘aunt’, ‘daughter’, ‘grandmother’, ‘sister’). He
notes that English has no gender morphs corresponding to the Spanish suffixes -o and -a, but
gender is nonetheless a significant component in the meaning of the English kin terms. Their
covert gender must be compatible with the sex of the person denoted; consequently, it is
anomalous to call one’s uncle aunt, or one’s sister brother. There are grammatical
consequences, which provide evidence for the covert gender: the personal pronoun anaphoric
to uncle is he/him; the one for aunt is she/her; *My brother is pregnant is anomalous. Father,
uncle, and aunt have in common that they are FIRST_ASCENDING_GENERATION. Father and
uncle additionally have in common that both are MALE, whereas aunt is FEMALE. Aunt and
uncle are both COLLATERAL, whereas father is LINEAL. Thus are meaning relationships
systematically identified.
occur; the structure of a semantic field mirrors the structure of the conceptual field. The
notion of semantic fields can be found in Alexander von Humboldt (1836) and it was
developed among German scholars (in particular Trier 1931; Porzig 1950; Weisgerber 1950;
and Geckeler 1971). John Lyons (1963) examined the meanings that can be ascribed to words
such as téchnē (“skill”), epistḗmē (“knowledge”), sophía (“wisdom”), aretḗ (“virtue”) (all of
these translations are oversimplified) in the semantic fields of knowledge and skill in Plato’s
works. Lyons was motivated by Jost Trier’s survey of the shifting field of High German
wîsheit, kunst and list. Around 1200 these three words meant approximately: knowledge,
which subsumed ‘courtly skill’ and ‘technical skill’. By 1300 wîsheit had narrowed to
‘mystical knowledge’; kunst shifted to ‘artistic skill’; while list was effectively replaced by
wizzen meaning ‘technical skill’. But unlike Trier’s subjective speculations, Lyons presents a
rigorous analysis using techniques derived from the works by Zellig Harris (1951) and Noam
Chomsky (1957).
Few scholars have undertaken extensive analysis of a semantic field, but Edward Bendix
(1966) analyzed the field of have and its counterparts in Hindi and Japanese, Adrienne Lehrer
(1974) studied the field of cooking and sounds, and Anthony Backhouse (1994) did an
extensive study of taste terms in Japanese. A conceptual field such as color, kinship, or
cooking terms is covered by a number of lexemes in a language, each denoting a part of the
field. Different languages and, any one language at different times in its history, may divide
the field differently among lexemes. Although the sensory data in the color spectrum are the
same for all human beings, languages name parts of the field differently. The differential
value (“valeur” in Saussure 1931 [1916]) of any lexeme is that part of the conceptual field
that it denotes in contrast with the part denoted by other lexemes in the same semantic field.
In the Papuan language Western Dani, laambu divides the color spectrum in half; the other
half is mili. The differential value of laambu is very different from English yellow, even
though it is a typical translation for English yellow, because laambu implies not-mili (not
cool-dark), whereas yellow implies (not-white, not-red, not-green, not-blue, not-black, not-
brown, not-pink, not-purple, not-orange, not-grey).4
Unlike the field of color terms, the field of cooking terms is not neatly circumscribed.
Since the analysis by Lehrer (1974), microwave ovens have become ubiquitous; and because
one can boil, roast, and poach in a microwave, the semantic field has been revised with the
4 This assumes that these color terms are ‘basic’ in the sense of Berlin and Kay (1969) – which is not
uncontroversial; see MacLaury (1997). For further discussion of basic color terms, see chapter 27.
6/38
advent of this new form of cooking. To generalize: when new objects and new ways of doing
things come into existence, there is a change in the conceptual field that usually leads to a
change in the semantic field and the addition or semantic extension of lexemes. (This, of
course, is what Trier was trying to show.) Seemingly closed fields such as case inflections or
kin terms should permit exhaustive componential analysis in which every term within the
field is characterized by a unique subset of the universal set of semantic components defining
the field. However, these systems invariably leak into other fields when meaning extensions
and figurative usage are considered. Furthermore an exhaustive componential analysis of the
entire vocabulary of a language is probably unachievable, because it proves impossible to
define the boundaries – and hence all the components – of every field.
Semantic primes are primitive symbols that with their interpretations constitute the
vocabulary of the semantic metalanguage that a linguist may use to describe and analyze a
particular language (object language). We may suppose that semantic components are, or are
composed from, semantic primes, but what are they and how many primes are there? A
number of seventeenth century seekers after a universal language, including Dalgarno (1661),
Lodwick (1652) and Wilkins (1668), proposed primitive semantic components (see chapter
7). Their contemporary, Arnauld (Arnauld and Nicole 1996) recognized that the meanings of
most words can be defined in terms of others but that, ultimately, there are some undefinable
semantically primitive words.
In the American tradition of the 1940s-1960s, Morris Swadesh sought to establish a list of
basic vocabulary created to plot diachronic relationships between unwritten languages in
Africa, the Americas, and elsewhere (see also chapter 16). Words in the Swadesh-list are
‘basic’ in the sense that they name things likely to be common to the experience of all human
communities (the sun, human body parts and functions, etc.). The purpose of the Swadesh-list
was to take a pair of languages and compare 100–215 basic lexemes to discover how many
are cognates (see Swadesh 1955); hence one name for the program is lexico-statistics (see
Embleton 1986). The scale of vocabulary differentiation derives from studies of Indo-
European languages for which there are historical records. For related languages, the time of
divergence from a common mother language is estimated from the proportion of vocabulary
common to both, a procedure sometimes called glottochronology.
In more recent times, Uriel Weinreich (1962: 36) identified a discovery procedure for a
semantic metalanguage built upon natural language. This was to (a) stratify the language into
a central core of semantic primes whose members are definable only circularly and by
7/38
ostensive definition such as “color of the sky” in the entry for blue. (b) The next stratum out
uses items whose definitions contain only core items without (further) circularity. (c) Each
more peripheral stratum uses items from the preceding strata without circularity. Since 1972
Anna Wierzbicka has been carrying out this program in a cross-language context, seeking a
universal set of semantic ‘primes’ (originally named ‘primitives’), based on a “natural
semantic metalanguage” (NSM). The proponents of NSM believe that semantic primes and
their elementary syntax exist as a minimal subset of ordinary natural language (Goddard
1994: 10). It is claimed (e.g. in Goddard 1994: 12) that “any simple proposition” expressed in
NSM using any one natural language (e.g. English) will be expressible in NSM using any
other language (e.g. Japanese). This embodies a claim that, like predicate logic, NSM is
linguistically and culturally unbiased and that there is a heuristic or algorithm for translation.
The number of semantic primes has grown from 14 (in Wierzbicka 1972) to 63 (in Goddard
2009). However, there is a distinct NSM for every language, and primes are often not
isomorphic across languages as the figures 1, 2, 3 are. NSM primes are compositionally and
often semantically different across languages; their meanings show partial overlap rather than
complete identity: English SOME corresponds in part to French IL Y A … QUI, and English
THERE IS to French IL Y A. There is a professed need for allolexes (different variants of a
single prime) which makes the “semantic primes” far more like meaning clusters than unit
primes, for example: English I and ME; DO, DOES, DID; Italian TU (singular ‘you’) is a prime,
but VOI, LEI (plural, polite ‘you’) are semantically complex and defined in terms of TU. At the
end of the 20th century, the characteristics of NSM syntax were only beginning to be
addressed (see Wierzbicka 1996: 19–22, Goddard 1998: 329–36). NSM is described as
“elementary” although it is unclear what differentiates a well-formed semantic definition or
description from an ill-formed one (Allan 2001).
The expressions used in a semantic representation in NSM are supposed to match those
that children acquire early. They are deliberately anthropocentric and subjective, referring to
the natural world of sensory experience rather than intellectualized abstractions. Thus, red is
the color of blood or fire (Wierzbicka 1980; 1990; 1992), not an electromagnetic wave focally
around 695 nanometers in length. There are important questions about the playoff between the
effectiveness of a definition and its accuracy. What is the purpose of the semantic analysis?
For whom or what is the resulting semantic specification designed? (These questions apply to
any semantic theory, of course.) NSM semantic definitions are not designed to be used by
machines that simulate language understanding; they are intended to be easily accessible to a
non-native speaker of the language (see Allan 2001). Where a brief description would be
sufficient for some objects (e.g., dog), terms for emotions, being culture-specific, are much
8/38
more difficult to define. This raises the question of just how specific a definition should be.
Cruse (1990: 396) claims that “For dictionary purposes, the concept has only to be identified,
not fully specified”. This, of course, should apply to the whole lexicographical endeavor and
not just to NSM (see also chapter 21).
Prototype and stereotype semantics are alternatives to theories of meaning which postulate a
“checklist” of properties to be satisfied for the correct use of the object language expression
‘e’ (Fillmore 1975: 123). For example, the default denotatum of bird is bipedal, has feathers,
and is capable of flight. But there are several species of flightless birds (e.g. emus, penguins);
a downy chick and a plucked chicken are featherless, but nonetheless birds; and a one-legged
owl and a mutant three-legged hen are also birds. So, many have pointed out that the notion of
a checklist of essential properties for the denotatum of e is problematic.
Ludwig Wittgenstein (1953: §§66–71) wrote of some categories being defined not by a
checklist of properties but by “family resemblances” (Familienähnlichkeit); e.g. games may
be expressions of amusement, play, competition, skill, luck, and so forth. Such subcategories
exhibit chains of similarity. Take the example of the word mother and the category ‘Mother’.
The prototypical mother is the woman who produces the ovum, conceives, gestates, gives
birth to and then nurtures a child (giving rise to the traditional definition of mother).
Radiating from this are more peripheral attributes of a mother. The natural or biological
mother produces the ovum, conceives, gestates, and gives birth to the child. The genetic or
donor mother supplies the ovum to a surrogate mother in whose womb the genetic mother’s
ovum is implanted and in whose body the fetus develops through to birth. The nurturant
mother may be the genetic mother, a surrogate mother, adoptive mother, or foster mother. In
addition there is a stepmother, a mother-in-law, while polygamous societies and certain
kinship systems (like many in Australia) offer additional complexities. Figurative extensions
arise: the prototypical or natural mother is the source for necessity is the mother of invention.
A mother’s status is recognized in the convention of referring to mother nodes in a
mathematical tree structure. The nurturant mother is the source for house-mother and also
mother superior in a religious order. By contrast, descriptions like single mother or working
mother can connote challenges to the individual’s capacity as a nurturant mother. What we
see here is a set of identifiable resemblances among these uses and meanings of the word
mother, but no set of properties common to all of them. As Wittgenstein pointed out, the
boundaries of a category can be extended and some of its members are more peripheral than
others.
9/38
This last point corresponds to the prototype hypothesis that some denotata are better
exemplars of the meaning of a lexeme than others, therefore members of the category denoted
by the lexeme are graded with respect to one another. For example a bird that flies, such as a
pigeon, is a better exemplar of the category Birds than a penguin, which doesn’t. How are
prototypes discovered? The psychologists Battig and Montague (1969) asked students to list
as many Vegetables, or Fruits, or Diseases, or Toys, etc. as they could in 30 seconds based on
the hypothesis that the most salient members in each category would be (a) frequently listed
and (b) high on the list. Thus, for instance, a carrot is the prototype for Vegetable, i.e. the best
exemplar of that category, because it was listed frequently and early. A tomato belongs to two
categories: it is a Vegetable in folk belief and technically a Fruit. On the Battig and Montague
(ibid.) scale, a tomato ranked 6th as a Vegetable and 15th as a Fruit. Using their figures for
salience, the tomato’s degree of membership of the category Vegetable is 68% and of the
category Fruit is only 14%.
George Lakoff (1972a) interprets such rankings in terms of fuzzy sets of objects with a
continuum of grades of category membership between 0.0 and 1.0. The carrot is the best
instance with a value 1.0 and a pickle only 0.006. A tomato has the value 0.68 and 0.14
membership in the fuzzy set Fruit. Any entity assigned a value greater than 0.0 is a member of
the category, i.e. the pickle is a Vegetable no less than the carrot. What the fuzzy set
membership value indicates is how good or bad an exemplar of the category a certain
population of speakers perceives that entity to be. A tomato is vegetable-like because it is
eaten, often with other vegetables, as part of an hors d’oeuvre or main course or in a salad. It
is not eaten, alone or with other fruits, for dessert. A tomato is fruit-like because it grows as a
fruit well above the ground and not on or below it. Also, it is often eaten raw and the
extracted juice is drunk like fruit juices. And, whereas flowers are cultivated for
ornamentation, tomatoes are cultivated for food. Conclusion: it is our practice of eating
tomatoes as if they are vegetables rather than fruit that explains the relative ranking in each
category.
The most influential work in this domain of prototype semantics is that of the psychologist
Eleanor Rosch, who carried out a series of experiments summarized in Rosch (1978); see
chapters 18 and 27 for further discussion). Rosch (1973) found that the common cold is a very
poor exemplar of Disease – which conflicts with the Battig and Montague (1969) finding. The
discrepancy between the two findings is explained by the fact that Rosch only gave her
subjects six diseases to rank (cancer, measles, malaria, muscular dystrophy, rheumatism, cold)
and a cold is the mildest of them. The salience would also be affected by the number of
people suffering from colds at the time of the experiment. Obviously, then, establishing the
10/38
prototype depends upon the experiences and beliefs of the population investigated.
Consequently, the claimed prototypicality ranking is valid for the community surveyed, but
not for all speakers of the language, or even for the same subjects on a different occasion.
William Labov (1978) reported on various kinds of labeling tasks such as applying the
terms cup, mug, bowl, glass, goblet and vase to line drawings of containers of different shapes
and configurations such as those in Figure 1, where those in the left column in Figure 1 are
close to the prototypical cup. Some subjects were asked to label a picture without any
particular context being mentioned, others where it was to be imagined that someone was
drinking coffee from it, or it contained mashed potatoes, soup or flowers. Sometimes the
vessel was said to be made of china, glass or aluminum. The results leave no doubt that the
term chosen is based on the perceived characteristics of the referent. For a given container,
naming depended on the shape and configuration; what it is made from; the purpose to which
it is put; sometimes its location.
Hilary Putnam (1975) proposed that the meaning of an object language expression e
(typically a lexeme) is a minimum set of stereotypical facts about its typical denotatum,
including ‘connotations’. Connotations of e arise from encyclopedic knowledge about the
denotation of e and also from experiences, beliefs and prejudices about the context in which e
is typically used. For example, the connotations of cat could include ‘cute’, ‘easy to take care
of’, ‘causes allergies’, ‘good for catching mice’, etc., although they may be different from one
11/38
person or speech community to another (for some, cats may be domestic animals that live
with their owners, for others they are farm or even semi-wild animals that live in a barn.)
Connotations are especially obvious with tabooed terms such as the difference between nigger
and African American or between shit and feces: the denotations of these pairs are (almost)
identical, but their connotations are very different (see below). Connotations vary between
contexts and speech communities independently of sense and denotation: a male chauvinist
and a radical feminist might have quite different stereotypes and connotations for man and
woman, but under normal circumstances will have no difficulty picking the denotatum of one
vs. the other. Putnam expressly allows for experts to have considerably more knowledge at
their command than other members of their speech community – which raises the interesting
question: Do the words elm and beech have the same stereotype and meaning for a botanist as
they do for an inner city dweller who can’t distinguish an elm from a beech? Presumably not.
However, if the botanist were to point out and name an elm, the inner city dweller would
know that referent is not a beech, even if s/he could still not recognize another elm thereafter.
Katz sought to establish a theory of meaning that would do all of the following:
define what meaning (i.e. sense) is; define the form of lexical entries; relate semantics to
syntax and phonology by postulating semantic theory as an integral component of a theory of
grammar; establish a metalanguage in which semantic representations, properties, and
relations are expressed; ensure that the metalanguage is universal by correlating it with the
human ability to conceptualize; identify the components of meaning and show how they
combine to project meaning onto structurally complex expressions. Essentially, these are
goals that should be met by any semantic theory – though what is meant by “component of
meaning” and the integration of semantics with phonology and syntax may be radically
different within different theories. Missing from Katz’s conditions is the requirement that the
meaning of language expressions needs to be related to the real and imaginary worlds people
speak and write of. Furthermore, Katz’s theory offered no account of utterance or speaker
meaning.
Katz’s semantic theory is interpretative. The earliest version (Katz and Fodor 1963), was
geared to the Chomsky’s (1957) syntactic model and was quickly abandoned when Dwight
Bolinger (1965) and Uriel Weinreich (1966) objected that its recursively conjoining meaning
components destroys input from syntactic structure. In later versions Katz’s theory was
designed to assign meanings to the output of autonomous syntactic rules of a transformational
generative grammar of the kind described by Chomsky (1965, Aspects of the Theory of
Syntax), but it was not updated to accommodate later developments in generative syntax (see
chapter 17). Katz never properly justified the vocabulary and syntax of his theory, and we can
only learn to interpret his metalanguage by abduction from his examples, among which there
is little consistency, and so his semantic markers remain only partially comprehensible. The
rules for constructing semantic markers were never specified, and there were at least five
13/38
differently structured semantic readings for chase given by Katz himself (Katz 1966; 1967;
1972; 1977b; Katz and Nagel 1974) and an additional two in Janet Fodor (1977; see also the
detailed discussion, explication, and exemplification in Allan (1986; 2010). Briefly: there is
no consistent set of relations in any semantic marker tree, which is not the case for those of
syntactic phrase markers (trees).
Katz and Paul Postal (1964: 16) proposed a set of semantic redundancy rules to reduce the
number of semantic markers in a dictionary entry. For instance, from the rule (Human) →
(Physical Object) ⋀ (Sentient) ⋀ (Capable of Movement), for every occurrence of (Human) the
redundancy rule adduces the entailed markers to give a full semantic specification. Most
semantic theories propose some counterpart to this.
Katz claimed that his theory directly captures all the subtleties of natural language, which
is a better instrument for semantic analysis than the metalanguages of standard logic because
it is a formal language that maps knowledge of language without confusing it with use of
language (Katz 1975a; b; 1977a; 1981). In fact we can only interpret Katz’s semantic
markers for chase, for instance, because it uses English words whose meanings we combine
to match up with our existing knowledge of the meaning of chase. If we reword his various
semantic markers for chase into more or less normal English, they will read something like X
is quickly following the moving object Y with the intention of catching it. Katz has claimed
(as have others, e.g., Lakoff 1972b and McCawley 1972) that the English used in the semantic
metalanguage is not English, which is used only as a mnemonic device. However, the only
way to make any sense of the metalanguage is to translate it into a natural language. That is
why analyzing bachelor into {(Human), (Adult), (Male), (Single)}, as did Katz and Nagel
(1974: 324), is a more enlightening semantic analysis than, say, {(48), (41), (4D), (53)}.
Formalism, especially unconventional formalism, can only be justified if it increases
explicitness of statement, rigor of analysis, and promotes clarity of expression.
Katz’s semantic theory has been discussed at some length because it was the first
comprehensive theory of linguistic semantics linked with generative grammar. For reasons
that have been given, it was not successful, but it did identify the parameters that other
theories needed to engage with. Another major limitation was no proper treatment of
pragmatics and no obvious extension beyond sentences to texts. These faults are also to be
found in many of its rivals, reviewed in this chapter.
Language combines the meaning encapsulated in lexemes into the complex meanings of
phrases, sentences, and longer texts. Such combination is conditioned by the rules of syntax
14/38
and at least four kinds of selectional restrictions (see Chomsky 1965). There are category
features (Noun, Verb, …), which determine different morphological and collocational
possibilities, e.g. of flyNoun and flyVerb in A fly flew into the room. Strict subcategorization
identifies other syntactic categories that collocate with the lexeme. Syntactically transitive
verbs, for instance, are defined by some notational variant of the strict subcategorization
feature “[+ ___NP]” “takes a 1st object”: open (as in Fred opened the box) has this feature,
whereas the intransitive verb in The door opened easily has the feature “[– ___NP]”. Inherent
features such as [+ human, + female, ...] for woman or [+ active, ...] for go have a semantic
basis. The selectional features of one lexeme refer to the inherent features of collocated
lexemes (e.g. for a verb “[+ [+ animate]___[+ abstract]]” “has an animate subject NP and an
abstract 1st object NP”). Originally, syntactic selectional features were postulated to
constrain what was assumed to be a purely syntactic process of lexical insertion into syntactic
phrase markers. Later, it was appreciated that the procedure is semantically conditioned, as
shown by meaningful sentences like Shakespeare’s Grace me no grace, nor uncle me no
uncle or Scott’s But me no buts. What governs the co-occurrence of lexemes is that the
collocation has some possible denotation (be it substance, object, state, event, process,
quality, metalinguistic statement, or whatever). The most celebrated example of a supposedly
impossible sentence, “Colorless green ideas sleep furiously” (Chomsky 1957: 15), was, in
1971, included in a coherent story by Yuen Ren Chao (1997 [1971]). Or one could take an
example marked anomalous in McCawley (1968a: 265): “*That electron is green”, which is
judged to be anomalous because electrons are theoretical constructs that cannot absorb or
reflect light, and therefore cannot be felicitously predicated as green. However, an
explanatory model of an atom could be constructed in which an electron is represented by a
green flash: there would be no anomaly stating That electron is green with respect to such a
model (see Allan 1986; 2006b).
Generative semantics
Noam Chomsky, the founder of generative grammar (see chapter 17), was educated in the
Bloomfieldian school, which, as said above (see also chapter 16), eschewed semantic theory
as speculative. For Chomsky semantics was at best an add-on for the syntactic base, a position
affirmed by Katz and Fodor (1963) and in subsequent work by Katz as explained above, and a
decade later by Jackendoff (see below). The Aspects theory developed in Chomsky (1965)
had a level of deep structure at which the meaning of each sentence constituent was specified
and the meaning “projected” upwards through nodes in the phrase marker to develop a
reading for the sentence. Deep structure was separate from a level of surface structure at
which the form of the sentence (as used in everyday utterance) was specified. This conception
of grammar leads naturally to the view that pairs of formally distinct but semantically
equivalent expressions arise from the same deep structure by different transformations, e.g.
(a) X caused Y to die and X killed Y or (b) X reminds me of Y and X strikes me as similar to Y
or (c) my mother and the woman who bore me. The next theoretical development, generative
semantics, proposed that the initial structures in a grammar are semantic rather than solely
syntactic (see also chapter 17). Despite its name, generative semantics was always primarily a
theory of syntax which focused exclusively on the structuring of meaningful elements. It grew
directly from reaction to the “standard theory” of Katz and Postal (1964) and Chomsky
(1965) with its emphasis on syntactic justification.
One of the earliest works in generative semantics was Lakoff (1965), published slightly
revised as Lakoff (1970), originally conceived as an extension of standard theory. Lakoff
postulated phrase markers that terminate in feature bundles like those in Chomsky (1965); he
differed from Chomsky in proposing that lexemes be inserted into only some of these
terminal nodes, the rest functioning as well-formedness conditions on lexical insertion and
semantic interpretation. Lakoff (1965) assumed, as did Chomsky, that lexical insertion
preceded all other transformations. Gruber (1965) contained lexical structures that have most
of the syntactic characteristics of standard theory trees, but some terminal nodes are semantic
components. Gruber argued that some transformations must operate on “prelexical syntax”
(prior to lexical insertion). For instance, from the prelexical structure VP[V[MOTIONAL,
POSITIONAL] PrepP[Prep[ACROSS] …]] lexical insertion will put either the verb go under the V
node and the lexeme across under the Prep node, or alternatively map the single verb cross
into a combination of both the V and Prep nodes. The latter was a radical innovation: because
semantic interpretation was made before transformations such as passive applied, semantics
and syntax were interdependent. A similar conclusion was reached by others (Postal 1966;
1970; 1972; Lakoff and Ross 1976, circulated from 1967).
16/38
Weinreich (1966) showed that lexical insertion is semantically governed and that syntactic
structure is just the skeletal structure for semantics. James McCawley (1968b) assumed that
all natural language syntax can be represented by the symbols S (sentence), V (verb
functioning as predicate) and one or more NPs (noun phrases functioning as logical
arguments). In initial structure, V consists of a semantic component or “atom” and NP, which
is either a recursive S node (if it’s an embedded sentence, such as in John said that he was
coming), or a variable (an index) designating the referent (John or the cat). Thus, in
generative semantics, meaning is determined directly from the initial semantic structure:
initial symbols represent semantic components set into structures that are a hybrid of
predicate logic and natural language syntax – both well-established conventional systems.
These structures can be rearranged in various ways by transformations before lexical forms
are mapped onto them. Then transformations may rearrange or delete nodes until the final
derived phrase marker gives a surface form for the sentence together with its structural
description.
The problem for generative semanticists was to give consistent semantic descriptions for
morphemes, lexemes, phrases, etc., as they occur in different sentence environments in such a
way that the meaning of any sentence constituent could be determined from the initial
sentence structure. The semantic metalanguage was based on a natural language, and both
Lakoff (1972b) and McCawley (1972) claimed that a semantic component such as CAUSE is
distinct from the English verb cause but they didn’t explain how this can be so. No rules
governing the insertion of semantic predicates under V were ever specified. Either selectional
restrictions must apply to constrain insertion or there will be unrestricted insertion subject to
output conditions (see Weinreich 1966). In practice no such constraints have ever been
systematically identified. There was also the problem identified by Fodor (1970): it was
proposed that a simple sentence like X killed Y derives from the complex X caused Y to die.
However, kill and cause to die cannot always be used in the same contexts. In a sentence like
(1) the adverbial on Sunday seems to block the insertion of kill; however the adverb in (2) has
no such effect.
Die is supposedly based on BECOME NOT ALIVE or cease to be alive. The fact that the
sentences in (3) are acceptable but those in (4) are not suggests that DIE is a semantic
component (atom, prime).
X died emaciated.
Allan (1986) argued against semantic decomposition of most lexemes in favor of recognizing
entailment relations such as those in (5).
This certainly seems to be justified from a psycholinguistic point of view (see Fodor, Garrett,
Walker et al. 1980).
Conceptual semantics
chapters 9 and 10; 1990, chapter 2). He made no attempt to decompose nouns semantically,
treating them as semantic primitives. In his view, only kin terms and geometric figures
admitted of satisfactory semantic decomposition. By contrast, he found that verbs decompose
into comparatively few classes (as also in Role and Reference Grammar, see below).
Jackendoff’s vocabulary of semantic primitives (1983, 1990) is very much larger than the
set used by NSM researchers (see discussion above). The syntax of his “lexical conceptual
structure” (LCS) is a configuration of functions ranging over arguments. For instance,
Conceptual semantics shows that a semantic decomposition of verbs making extensive use
of just a few primitives is a feasible project. The syntax of LCS is a function-argument
structure similar to that of predicate calculus (see below and chapter 20), so that someone
acquainted with predicate calculus can construct a lexical conceptual structure despite the fact
that Jackendoff (1983, 1990) did not employ standard logical formulae. Although LCS made
no use of logical connectives, some of the more complex formulae imply conjunction between
the function-argument structures in a lexical conceptual structure. There is a score of
primitive verbs so far identified, so although the set of functions is restricted, the vocabulary
of primitive arguments is unbounded. Conceptual semantics was designed to integrate with a
dominant syntactic theory in late twentieth century linguistics: A-marking links the semantic
interpretation to a node in the syntactic phrase marker (see chapter 17). Jackendoff (ibid.)
suggested that “argument binding” in LCS (using Greek superscripts) does away with the
need for the level of “logical form” (LF) in syntax (the level of representation which fully
determines the semantics of a sentence – see chapter 17). LF has not yet been abandoned in
favor of conceptual structure; but Jackendoff’s conceptual semantics has been a real force
within the development of grammatical theory.
19/38
According to Van Valin (1993) and Van Valin and LaPolla (1997), the definition of
thematic roles in grammar is so unsatisfactory that we should admit just two macroroles,
‘actor’ and ‘undergoer’, in the grammar. The macroroles of Van Valin’s Role and Reference
Grammar are similar to the proto-roles in Dowty (1991); they are defined on the logical
structures of verbs. The maximum number is 2, the minimum is 0 (in sentences like Latin
pluit ‘[it’s] raining’ and English It’s raining). ‘Actor’ and ‘undergoer’ roughly correspond to
‘logical subject’ and ‘logical object’ respectively. They are called macroroles because they
subsume a number of thematic roles. They are properly dependent on hierarchies such as the
actor hierarchy, DO(x,... ≺ do′(x,... ≺ PRED(x,… ;6 the undergoer hierarchy without an actor,
PRED(x,... ≺PRED(...,y) ≺ PRED(x) (where A ≺ B means “A outranks B in the hierarchy”). In
contrast to the uncertainty of assigning thematic roles, assigning macroroles to a clause
predicate is well-defined.
5 Entailment is a relation such that if A entails B then whenever A is true, B is necessarily true. A
nonmonotonic inference is one that is not necessarily true, though it might be: if most nurses are women
and Pat is a nurse it does not follow that Pat is necessarily a woman; by contrast, being a natural mother
entails being a woman, thus if Pat is a (natural) mother, Pat is a woman.
6. DO only appears in the few logical structures that necessarily take an agent e.g. for murder as against kill.
20/38
Role and Reference Grammar, RRG (Foley & Van Valin Jr 1984; Van Valin & LaPolla
1997; Van Valin 1993, 2001, 2005; Van Valin n.d.), is a functionalist theory that does not
posit underlying and surface representations as different strata but integrates morphology,
syntax, semantics, pragmatics, and information structure in a readily accessible monostratal
representation (see chapter 17 for a discussion of multistratal representations, e.g., the
difference between deep structure and surface structure). RRG has been specifically
developed to apply to every natural language and seeks to show how language expressions are
used to communicate effectively. The basic clause structure consists of a predicate, which
together with arguments (if any), forms the Core. Other constituents are peripheral; they may
be located in different places in different languages, and can be omitted. The structures are
more like nets than like trees.
“Frames” (Goffman 1974; Fillmore 1982; 2006; Fillmore and Atkins 1992) identify the
characteristic features, attributes, and functions of a denotatum, and its characteristic
interactions with things necessarily or typically associated with it (see also chapter 27 for a
discussion of ‘frames’). For instance, a restaurant is a public eating-place; its attributes are:
(1) business premises where, in exchange for payment, food is served to be eaten on the
premises; consequently, (2) a restaurant has a kitchen for food preparation, and tables and
chairs to accommodate customers during their meal. Barsalou (1992: 28) described
“attributes” as slots in the frame that are to be filled with the appropriate values. The frame
for people registers the fact that, being living creatures, people have the attributes of age and
sex. The attribute sex has the values male and female. It can be represented formally by a
function BE SEXED applied to the domain D={x: x is a person} to yield a value from the set
{male, female}. The function BE AGED applies to the same domain to yield a value from a
much larger set.
21/38
Frames interconnect in complicated ways. For instance, the social status and the
appearance of a person are usually partly dependent upon their age and sex, but not
necessarily so. Knowledge of frames is called upon in the proper use of language. Part of the
frame for bird is that birds are FEATHERED, BEAKED and BIPEDAL. Most birds CAN FLY; applied
to an owl this is true, applied to a penguin it is false. Birds are sexed, and a (normal adult)
female bird has the attribute CAN LAY EGGS with the value true. Attributes for events include
participants, location, and time of occurrence, e.g. the verb buy has slots for the attributes
buyer, seller, merchandise, payment: these give rise to the “thematic structure” (see above) of
the verb. An act of buying occurs in a certain place at a certain time (a world~time pair with
values relevant to evaluation of truth, see below). To sum up, frames provide a structured
background derived from experience, beliefs, or practices, constituting a conceptual
prerequisite for understanding meaning. The meaning of a language expression relies on the
frames, and it is these that relate lexemes one to another.
“Construction grammar” (Fillmore and Kay 1987; Goldberg 1995; 2006) is a development
based on frame semantics. Unlike the semantic theories of Katz and Jackendoff, it does not
project meaning onto syntactic structures from lexemes. A projection would require the verbs
italicized in (9)–(12) to be distinct from default meanings: pant is not normally a motion verb;
bark and sneeze are not normally causative; knit is not normally ditransitive (i.e. a three place
22/38
verb like give, which has a giver, the person given to and the thing given, as in Ed gave Eli
the book).
(9) All in a sweat, Marlow panted up to the door and rapped on it loudly.
(10) The prison warder barked them back to work.
(11) Adele sneezed the bill off the table.
(12) Elaine knitted George a sweater for his birthday.
The additional verb meanings result from the construction in which the verb occurs.
Construction grammar proposes various integration types. For instance in (9) and (11) the
construction indicates the motion, the verbs pant and sneeze the manner of motion; in (10)
and (11) the constructions are causative, indicating a theme and result; in (12) the valence of
knit is augmented to make it a verb of transfer by mentioning the recipient/beneficiary
(compare buy). The construction coerces an appropriate interpretation by imposing the
appropriate meaning. This is exactly what happens with apparent violations of selectional
restrictions discussed earlier; also in interpreting variable countability constructions such as
(13)–(15) (see Allan 1980).
The principal motivation for countability is to identify the individual from the mass; typically,
uncountable referents are perceived as an undifferentiated unity, whereas countables are
perceived as discrete but similar entities. Thus (14) offers as alternatives an individual potato
or a quantity of, say, mashed potatoes; (15) compares an unspecified quantity (mass) of sugar
with three individual spoonfuls or lumps of sugar. (16) compares the herd as a single
collection of animals against the herd as a set of individual animals (not all dialects of English
allow for this).
Linguistic theories of meaning that go beyond lexis to account for the meaning of syntactic
constructions necessarily incorporate aspects of lexical semantics. For many centuries certain
philosophers have discussed aspects of lexical and propositional meaning. From the term
logic of Aristotle and the propositional logic of the Stoics developed the strands of inquiry
dealt with in the next part of this chapter.
23/38
Davidson (1967b: 310) said that “to give truth conditions is a way of giving the meaning of a
sentence.” But truth is dependent on worlds and times: Marilyn Monroe would have been 74
on June 1, 2000 is true: although MM died in 1962 we can imagine a possible world of June
1, 2000 at which she was still alive, and given that she was born June 1, 1926, she would
indeed be 74. McCawley (1968b; c) was one of the first linguists to adopt and adapt truth
conditions and predicate logic (a common way of studying truth conditions, see chapter 20)
into grammar, most popularly in his book Everything that Linguists Have Always Wanted to
Know about Logic (McCawley 1993 [1981]). The importance of truth conditions had often
been overlooked by linguists, most especially those focusing on lexical semantics. Hjelmslev
(1943); Lyons (1968) and Lehrer (1974) suggest that the nine lexemes bull, calf, cow, ewe,
foal, lamb, mare, ram, stallion – which constitute a fragment of a semantic field (see above) –
can be contrasted with one another in such a way as to reveal the semantic components in
Table 1.
How can we determine that the analysis is correct? The basis for claiming that BOVINE or
MALE is a semantic component of bull cannot be a matter of language pure and simple. It is a
relation speakers believe exists between the denotata of the terms bull and male and bovine
(i.e. things in a world that they may be felicitously used to refer to). Doing semantic analysis
of lexemes, it is not enough to claim that (16) is linguistic evidence for the claim that MALE is
a semantic component of bull, because (17) is equally good until a basis for the semantic (and
therefore grammatical) anomaly has been established that is independent of what we are
seeking to establish – namely the justification for the semantic components identified in
Table 1.
The only language-independent device available is an appeal to truth conditions, and this
takes us to the denotata of bull and male. In fact what we need to say is something like (18).
24/38
(18) In every admissible possible world and time an entity which is a bull is male and in no
such world is an entity which is a bull a female.
Note that the semantic component MALE of Table 1 must be equivalent to the relevant
sense of the English word male. Thus, the assumption is that semantic components reflect
characteristics of typical denotata as revealed through their intensions across worlds and
times. Intensions are what ‘senses’ describe. Some people think of them as concepts, others
as the content of concepts (see below). In any case, they provide the justification for
postulating the semantic components in Table 1 as a set of inferences such as those in (19).
(19) For any entity x that is properly called a bull, it is the case that x is adult ⋀ x is male ⋀ x
is bovine.
In fact it is not part of a general semantic characterization of bull that it typically denotes
adults; one can, without contradiction, refer to a bull calf. Rather, it is part of the general
naming practice for complementary sets of male and female animals. Nor is bull restricted to
bovines, it is also used of male elephants, male whales, male seals, male alligators, etc. The
initial plausibility of Table 1 and (19) is due to the fact that it describes the prototypical or
stereotypical bull (see above). The world of the English speaker is such that bull is much
more likely to denote a bovine than any other species of animal, which is why bull elephant is
usual, but bull bovine is not. This reduces (19) to something more like (20).
(20) For any entity x that is properly called a bull, it is the case that x is male and probably
bovine.
What is uncovered here is that even lexical semantics is necessarily dependent on truth
conditions together with the probability conditions that are nonmonotonic inferences
sometimes equated with implicature (see below).
Formal semantics
Since about the time of Cresswell (1973) and Keenan (1975) there have been many linguists
working in formal semantics. Formal semantics interprets formal systems, in particular those
that arise from the coalescence of set theory, model theory, and lambda calculus (models and
lambda calculus are briefly illustrated below)7 with philosophical logic – especially the work
of Richard Montague (Montague 1974; see also Dowty, Wall and Peters 1981), and the tense
logic and modal logic of such as Prior (1957) and Kripke (1963; 1972). By and large, formal
7. See also Chapter 20 in this volume, and Gamut (1991); McCawley (1993); Allan (2001) for explanations
of these terms. (L.T.F. Gamut is a collective pseudonym for Johan F. A. K. van Benthem, Jeroen A. G.
Groenendijk, Dick H. J. de Jongh, Martin J. B. Stokhof, and Henk J. Verkuyl.)
25/38
semantics has ignored the semantics of lexemes such as nouns, verbs, and adjectives – which
are typically used as semantic primes (but see Dowty 1979). It does, however, offer insightful
analyses of secondary grammatical categories like number and quantification, tense, and
modals.
“Event-based semantics” was initiated by Davidson (1967a). The idea is to quantify over
events; thus Ed lifts the chair is represented in terms of “there is an event such that Ed lifts the
chair”. In Ed hears Jo call out there is a complex of two events as shown in (21), where there
is the event e of Jo’s calling out and the event e′ of Ed hearing e; ∃ is the existential
quantifier “there is”.
Following a suggestion by Parsons (1980; 1990) thematic roles can be incorporated as in (22),
Max drinks the beer.
There is always the question of how the meanings of complex expressions are related to
the simpler expressions they are constructed from: this aspect of ‘composition’ is determined
by model theory in Montague semantics, which is truth conditional with respect to possible
worlds. Truth is evaluated with respect to a particular model of a state of affairs. For instance
if a model consists of Harry, Jack, and Ed, and Harry and Ed are bald but Jack is not, and if
Harry loathes Ed, then we can evaluate the truth of such statements as Not everyone is bald,
Someone loathes someone who is bald, and so forth. Where traditional predicate (and
propositional) logic is concerned only with extension (existence) in the (real) world,
intensional logic allows for existence in a possible (hypothetical) world. Just as intensions are
comparable with ‘sense’, extensions are comparable with ‘reference’ or, better, denoting
something within a particular model (or set of models). In Montague semantics, semantic
structure is more or less identical with syntactic structure.
In later developments (see Gamut 1991; Chierchia and McConnell-Ginet 2000) valuation
functions were proposed. Suppose there is set of men a, b, and c (Arnie, Bob, Clive) who
constitute the domain of a model world at a particular time, M, in which a and c are bald. The
M
extension of baldness in M is represented〚bald〛. Let x stand for any member of {a, b, c}. A
valuation function takes a sentence x is bald as its domain and assigns to it a value in the
M
range {0,1} where 1 is true and 0 is false. So the function〚bald〛 applies in turn to every
member of the domain X in model M to assign a truth value. The extension of being bald in
M
M is〚bald〛 = {a, c}. Put another way: in M, bald(x)=1 ↔ x∈{a, c} “x is bald is true if, and
26/38
only if, x is a member of the set {a, c}”. To evaluate Someone is not bald in M, a variable
assignment function would check all assignments of x until one instance of x is not bald is
found to be true (in our model, when x is assigned to b).
Although Ross’s hypothesis was principally a syntactic device, it opened the gate to
pragmatic constraints on pronouns relevant in exophora, i.e., when referring to something in
the outside world (Just look at her! said of a passing woman) and recognizing the most likely
actor (the ‘biter’) in I took my dog to the vet and she bit her. Huang (2000) has argued that
pragmatics accounts for what Chomsky (1981) identified as syntactic binding conditions on
anaphors. Intuitively, argument binding is a matter of semantics and pragmatics rather than
syntax, e.g. the pronoun her appropriately refers to, say, Amy for semantic not syntactic
reasons. In German, das Mädchen “girl” is rendered neuter by its diminutive suffix -chen but
is normally pronominalized in colloquial speech on a pragmatic basis by the feminine sie and
not the neuter es, although the matter is hotly debated and in the written language the use of
the neuter pronoun is the norm. The choice of what are generally referred to as anaphoric
forms in texts has been discussed under: the “familiarity hierarchy” (Prince 1981); “centering
theory” (Grosz 1977; Sidner 1979); “topic continuity” (Givón 1983); and the “accessibility
27/38
theory” (Ariel 1988; 1990). These all emphasize the importance of context (see below) in
selecting what form of anaphor to use.
As a rule, any two successive references to an entity involve some kind of change on the
second reference, see (24).
(24) Catch [a chicken1 ]. Kill [it2 ]. Pluck [it3 ]. Draw [it4 ]. Cut [it5 ] up. Marinade [it6 ]. Roast
[it7 ]. When you’ve eaten [it8 ], put [the bones9 ] in the compost.
All nine subscripted NPs refer to the creature identified in ‘a chicken1’, which refers to a live
chicken. By 2 it is dead, by 3 featherless, by 5 dismembered, by 7 roasted, and by 8 eaten. 9
refers to the chicken’s bones after the flesh has been stripped from them. Thus 7, for instance,
refers not to the chicken in 1, but to the caught, killed, plucked, drawn, cut up, and marinaded
pieces of chicken. Heim (1983; 1988) described this as “updating the file” on a referent.
These successive states of the chicken are presented as changes in the world–time pair spoken
of: although the world stays constant throughout (24), each clause corresponds to a temporal
change: time1, time2, ... time9. The aim of Heim’s file change semantics has much in common
with that of “Discourse Representation Theory” (DRT, Kamp 1981; Kamp and Reyle 1993)
where the interpretation of one in a sequence of utterances (a discourse) is dependent on co-
text such that the next utterance is an update of it. DRT has been especially successful in
capturing the complex semantics of so-called ‘donkey sentences’, originating in Walter
Burley’s “Omnis homo habens asinum videt illum”,’Every man who has a donkey sees it’ of
1324 (Burley 2000). Consider, for instance, (25) – which paraphrases as (26).
First take (26): its discourse representation structure (DRS) is (27). The arrow indicates that
the second box is a consequence of the first. The left-hand box is interpreted first, then the
right-hand box. Because it shows movement from one state to another, (27) can be thought of
as a dynamic model.
(27) x y z
girl(x)
z=y
pony(y)
x loves z
x owns y
Notice that the anaphor for a-pony-loved-by-the-girl-who-owns-it is ‘z’, and it does not occur
in the left-hand box. We now turn to (25) for which the DRS is (28).
28/38
(28) x y z
girl(x)
pony(y) every z=y
x owns y x x loves z
DRT is undergoing extensions in the twenty-first century, see Asher and Lascarides (2003);
Jaszczolt (2005).
Aspects of Pragmatics
This section focuses on the importance of context for any account of meaning in language and
for the understanding of indexicals (e.g. now, you, here).
There are at least two ways in which the meaning of a new word can be revealed by
whoever coins it: it may be formally defined – a rare procedure in everyday language use -- or
the hearer or reader is left to figure out the meaning from its use in the prevailing context. The
term ‘context’ denotes any or all of at least four things: the world and time spoken of; the
co-text (i.e. the text that precedes and succeeds a given language expression); the situation of
utterance; and the situation of interpretation. The meaning ascribed by use in particular
contexts will take precedence over any formally defined meaning. As Wittgenstein (1953: 43)
famously wrote: “the meaning of a word is its use in the language.” Assignment of meaning
by ordinary use is phylogenetically and ontogenetically prior to defined meaning – but for
words (lexemes) not sentences, because at any one time the set of lexemes is bounded, but the
set of sentences is not. However the ways in which the meanings of sentences are constructed
are determined by use, so although no speaker could literally and truthfully say I’ve just been
decapitated (in other words it has no extension), the meaning is readily interpretable via its
intension (including a metaphorical interpretation). One problem with describing meaning in
terms of usage is that it risks confusing denotation with connotation: the denotations are the
same of urine and piss or my mum and the woman who bore me, but the connotations are
different. Each of these two pairs can be used of the same referent, but the contexts of use are
normally different. Roughly speaking, denotation is what a lexeme is normally used to refer
to, whereas the connotations of a language expression are pragmatic effects that arise from
encyclopedic knowledge about its denotation (or reference) and also from experiences,
beliefs, and prejudices about the contexts in which the expression is typically used (see Allan
2007).
The anthropologist Malinowski (1923) coined the term “phatic communion” to refer to the
social-interactive aspects of language (greeting, gossip, etc.) and to focus on the importance
of the “context of situation” in representing meaning. This view was adopted by J. R. Firth,
29/38
the most celebrated British linguist of his generation (see chapter 15). Firth, who was also
influenced by the Prague school functionalists (Mathesius 1964 [1936]; Vachek 1964; see
chapter 14, Part two), emphasized the importance of studying meaning in the context of use,
taking into account the contribution of prosody (Firth 1957, 1968). His own work in these
areas is less significant than the effect it had on one of his students, Michael A. K. Halliday,
who worked on prosody and also developed a polysystemic grammatical theory on Firthian
lines, originally called System-Structure Grammar, then Scale and Category Grammar, and
now Systemic Functional Grammar (SFG) (see The Collected Works of M.A.K. Halliday,
Halliday 2002-2009, see chapter 18). Halliday and his school have always been interested in
the grammatical analysis of text and discourse and the Hallidayan approach has been taken up
in Rhetorical Structure Theory (RST), which offers an account of narrative structure (see
Halliday & Hasan 1976, 1989 [1985]; Mann et al. 1992; Mann & Thompson 1986;
Matthiessen & Thompson 1988). Hallidayan theory has also been adopted by critical
discourse analysts, who focus on the fact that all language is socio-culturally and
ideologically situated (e.g. Hodge & Kress 1988; Kress & van Leeuwen 2001, see chapter
18). An important Hallidayan contribution to linguistic terminology is the labeling of
“metafunctions” (see also chapters 15 and 18)..
Indexicals
The situations of utterance and interpretation provide anchors for deictic or indexical
categories such as tense, personal pronouns (like I, you, we), deictic locatives (here, there)
and demonstratives (this, that) (see also chapters 19 and 20). The term ‘deixis’ derives from
the Stoic δειξις (demonstration, indicated referent); “indexical”, in this sense, was introduced
by Peirce (1931: Chapter 2). Although study of these grammatical categories had been
proceeding for more than two millennia, there was an upsurge of interest after World War II
(see Benveniste 1971 [1956], 1971 [1958]; Jakobson 1962; Lyons 1977; Levinson 1983;
Fillmore 1966, 1997 [1971-75]). Many languages have, corresponding to their personal
pronoun systems, the speaker as first person (I), the hearer as second person (you), all others
as third person (he, she, it), as well as parallel locatives meaning roughly “near speaker”
(here), “near hearer” (there), “not-near either speaker or hearer” (yonder). The situations of
utterance and interpretation may determine choices of adverbials and directional verbs
relative to the location of speaker and hearer; e.g. the choice among the verbs come, go, bring,
come up, come down, come over, etc. Situation of utterance and assumptions about the hearer
also play a role in determining the topic and the linguistic register or jargon – that is, the
variety of language associated with a particular occupational, institutional, or recreational
30/38
group: for instance, legalese, medicalese, cricketese, linguisticalese, and so forth (Biber &
Finegan 1989; Allan & Burridge 1991, 2006). They influence politeness factors such as
terms of address and reference to others (see Brown & Gilman 1972 [1960]; Ervin-Tripp
1972; Geertz 1972; Shibatani 2006); and kinesic acts such as gesture, facial expression, and
the positions and postures of interlocutors (Hall 1959; Argyle 1988; Clark 1996; Danesi
2006).
Scripts
Beyond earliest childhood, very little we encounter is totally new in all its aspects. Most of
what we hear and read can be interpreted wholly or partially in relation to structured
knowledge arranged into modules of information8. A speaker presupposes this common
ground when constructing a text so that understanding (30) is to invoke the restaurant script
(Schank and Abelson 1977; Schank 1982; 1984; 1986) as a set of inferences, some of which
are defeasible (can be cancelled without contradiction).
From (29) we infer that, most probably: (a) Sue intended to eat at the restaurant with her
boyfriend; (b) Sue entered the restaurant with her boyfriend; (c) Sue and her boyfriend sat
down; (d) They ordered food; (e) The food was brought to them; (f) They ate it; (g) Either Sue
or her boyfriend or both of them paid the bill; (h) Then they left the restaurant. In (29) many
of the inferences in the first clause are cancelled by what follows:
(30) Sue went to a restaurant last night with her boyfriend, but as soon as they’d got inside
the door they had a huge fight and left before even sitting down.
It is confirmed that they entered and exited but it implicitly denied that they sat down,
ordered, ate, and paid.
As can be seen from the example above, scripts contain structured information about
dynamic event sequences. Regular components of a script are predictable and deviations from
a script are potentially newsworthy. Scripts have personae, props, and action sequences. A
restaurant script has customers, servers, cooks, etc. The props include tables, chairs, menus,
cutlery, plates, food. The events include the customer entering the restaurant, ordering food,
the food being brought by the server, the eating of the food, the requesting, presentation, and
paying of the bill, and the customer leaving the restaurant. The vocabulary used in the script
evoked by going to a restaurant indicates its semantic associations and their relationships.
8. For a discussion of frames, scripts, and schema(ta) in the context of psychology, see chapter 27.
31/38
Conversational implicature
Once the meaningful interpretation of a language expression makes recourse to context,
pragmatics comes into play (see Gazdar 1979; Levinson 1983). The boundary between
semantics and pragmatics was specifically defined by Grice (1975: 43) as a distinction
between “what is said” – the truth-conditional aspects of meaning – and “what is implied,
suggested, meant” – the non-truth-conditional pragmatic overlay that is implicated. Grice
writes of the small conversation in (31), “B implicates that Smith has, or may have, a
girlfriend in New York” (ibid. 51).
The implicature is inferred from what B actually says given the cooperative assumption that it
is a rational response to A’s remark, i.e. that it is relevant to the co-text. Implicatures (more
precisely, conversational implicatures) are, for instance, the defeasible inferences discussed in
respect of (29) and (30) above.
32/38
Grice (1975) described the cooperative principle in terms of four categories of “maxims”:
“Quantity”, “Quality”, “Relation”, and “Manner” (? see also chapters 18, 20, 25). Quantity
enjoins the speaker/writer to make the strongest claim possible consistent with his/her
perception of the facts while giving no more and no less information than is required to make
his/her message clear to the audience. Quality enjoins the speaker/writer to be genuine and
sincere. The maxim of relation requires that an utterance should not be irrelevant to the
context in which it is uttered, because that would make it difficult for the audience to
comprehend. The maxim of manner requires that, where possible, the speaker/writer’s
meaning should be presented in a clear, concise manner that avoids ambiguity, and avoids
misleading or confusing the audience through stylistic ineptitude. Such maxims are not laws
to be obeyed, but reference points for language interchange – much as the points of the
compass are conventional reference points for identifying locations on the surface of the
earth. The cooperative maxims are fundamental to a proper account of meaning in natural
language; even though they are pragmatic entities, to build a semantic theory that makes no
reference to the implicatures that arise from cooperative maxims would be like building a car
with square wheels. The perceptiveness of Grice’s observations cannot be denied; much
criticism has been leveled against various maxims but they fail if we interpret Grice
charitably. One frequent objection is that Grice mistook the conventions of his own society to
be universal; this is a common enough mistake and not fatal to the theory.
The four Gricean (categories of) maxims were reduced to three (Manner, Quantity, and
Informativeness) in Levinson (1995; 2000), two (Relation and Quantity) in Horn (1984), and
one (Relevance) in Sperber and Wilson (1995 [1986]); see the comparison in Table 2.
It has become a matter of controversy whether or not there is a clear distinction between
what is said and what is meant. Horn (1972) identified sets of scalar implicatures, e.g. Ed has
three children implicates Ed has exactly three children; Some felines don’t have retractile
claws implicates Not all felines have retractile claws; I think he’ll apply implicates I don’t
know it for a fact that he’ll apply. Grice (1978) accepted these as “generalized conversational
implicatures” because they do not rely on a particular context unlike the particularized
implicature in (31). Grice (1978: 117) famously wrote: “Senses are not to be multiplied
beyond necessity”, described as “Modified Occam’s Razor”’. This became known as “radical
pragmatics” (see Cole 1981); it favors the underspecification of semantics, the additional
meanings being supplied from pragmatics. For instance, in (32) we assume that the light came
33/38
on as a result of Sue flicking the switch, whereas in (33) Sue’s flicking the switch seems to
follow the light coming on.
(32) Sue flicked the switch and the light came on.
(33) The light came on and Sue flicked the switch.
These are pragmatic inferences that can be cancelled and, certainly out of context, (32) and
(33) have the same truth conditions (because by the rules of propositional logic (p ⋀ q) ↔ (q ⋀
p)). However, according to Carston (1988), although they have the same logical form they
have different truth conditions which are determined on the basis of pragmatic enrichment
(e.g. (32) is true just in case Sue’s flicking the switch caused the light to come on).
The fact that It’s raining is necessarily understood as referring to rain in some particular
locality has led Relevance Theorists (Sperber and Wilson 1995 [1986]; Carston 1988; 2002)
to name this an “explicature” on the basis that it enriches logical form by making it more
explicit. Bach (1994) calls it an “impliciture” because it results from expansion of what is
implicit in the semantic content. Bach (2004) claims that semantics concerns the meanings of
sentences, which may often fail to determine the meaning of propositions completely but
these are semantic properties independent of anybody’s act of uttering them. Pragmatics is
concerned, not with sentences themselves, but with utterances of sentences in the course of
communicating, and truth conditions apply to these utterances rather than to sentences out of
context. The debate over the semantics~pragmatics interface continues (see, e.g. Allan and
Jaszczolt 2012).
Summary Remarks
This chapter has reviewed developments in semantics and pragmatics from 1960 to the end of
the twentieth century. Part I on lexical semantics began with a review of componential
analysis in semantics which has a history dating back to the seventeenth century but was
rediscovered in the mid-twentieth century inspired by the analyses of kinship systems in
anthropology, distinctive feature analysis in Prague school phonology, and the extension of
features to morphosyntactic analysis. Componential analysis was also encouraged by the
recognition of semantic fields and the differential values of lexemes within them. The
semantic components of lexemes derive from properties of the denotata (referents) of
lexemes. Recognition of the fact that some denotata are more typical and better exemplars of
a lexeme than others led to developments in prototype and stereotype semantics. Theories of
semantics seek to identify the components of meanings, relate them to one another and to
their formal representation in lexemes (more properly, the items listed in dictionaries), to
34/38
identify the network of meanings among them and combinations of them that give body to
syntactic structures.
The difference in meaning between The hunter killed a rhino and The rhino killed a hunter
arises from the syntactic differences and it is that aspect of meaning which was discussed in
Part II. Katz’s semantic theory was examined at some length because, for all its faults, it was
the first comprehensive theory of linguistic semantics in generative grammar; Katz
adumbrated the goals that should be met by any semantic theory and attempted to marry
semantic interpretation to syntactic structure. Alternatives to Katz’s theory such as generative
semantics, conceptual semantics, the semantics and pragmatics of the functionalist Role and
Reference Grammar, and then frame semantics and construction grammar were reviewed.
One major problem facing attempts to predict the meaning of a syntactic structure is the
difficulty of identifying the constraints on combining lexemes; it remains an unsolved
problem, although the best proposal so far seems to be the identification of semantic frames
and the recognition that interpretations are coerced by co-textual lexemes.
Part III looked at the contribution of truth conditional and other accounts of linguistic
meaning derived from philosophical logic. It was shown that even lexical semantics is
necessarily dependent on truth conditions and the probability conditions expressed as
nonmonotonic inferences that are sometimes equated with implicature. Event-based semantics
and Montague semantics, which use model theory, possible worlds semantics, and lambda
calculus were briefly reviewed. There was a short discussion of the fact that much anaphora
can be accounted for on semantic or pragmatic bases. This led into file-update semantics and
the dynamic semantics of discourse representation theory.
Part IV took up aspects of pragmatics, starting with the importance of context to the proper
interpretation of linguistic meaning. We dallied briefly with the contribution from Halliday’s
systemic functional grammar, then switched to the discussion of indexicals. Scripts that
capture recurrent dynamic event sequences are indispensable to the understanding of how
linguistic meaning is normally very much underspecified in natural language discourse: a
huge amount is left to be inferred (monotonically or nonmonotonically) by hearers and
readers. That topic leads directly to a consideration of implicature and the controversy over
the terms to be used to distinguish what is said from what is meant.
Of course much has been omitted in this short excursus: Lyons’ book on semantics (1977)
is unsurpassed as a handbook on many aspects of semantics: it presents a background for
much of what has been discussed (see also Cresswell 1979 for an excellent review). Stephen
Ullmann (1951; 1962; 1975) wrote on etymology and semantic relations among lexemes and
35/38
some grammatical morphemes. D. Alan Cruse’s book (1986) became a standard introduction
to lexical semantics. There are several histories of some aspects of semantics in late twentieth
century America (the introduction to McCawley 1976; Newmeyer 1986; R. Harris 1993;
Huck and Goldsmith 1995). Other works (such as Allan 2009, 2010; Horn and Ward 2004;
Mey 1998) give surveys of semantics and pragmatics at or just after the end of the twentieth
century.
36/38
Figures
Tables
Table 2. Sources for implicatures in Grice, Levinson, Horn, and Sperber and Wilson
GRICE LEVINSON HORN S. AND W.
Quality: Be truthful.
Relation: Don’t be irrelevant Be optimally
without cause. R: Make your contribution necessary; relevant.
say no more than you must (given Q).
Quantity: Say no more and no Q: What isn’t said, isn’t.
less than is necessary to get
the point across. I: What is expressed simply is Q: Make your contribution sufficient;
stereotypically exemplified. say as much as you can (given R).
Manner: Don’t be stylistically M: What’s said in an
inept. abnormal way isn’t normal.