Semantics and Grammar

Grammar and semantics are often thought of as separate levels of linguistics. But grammatical
categories often have meaning. Most of the traditional grammarians assumed that grammatical
categories are essentially semantic. Nouns were defined as names of things, gender was
concerned with sex, while plural simply meant 'more than one'. On the other hand, many
linguists have argued that grammar must be kept distinct from semantics and that grammatical
categories must be wholly defined in terms of the FORM of the language, the actually observable
features. One of the earliest statements is Sapir, but he believed in his theory of linguistic
relativity which maintained that since each language had a different formal structure, it presented
a different world. Likewise, Bloomfield argued that we must be 'scientific' and the study of
meaning was a weak point in linguistic theory; hence formal features, not meaning, should be the
starting point of a linguistic discussion.

There are two good arguments for FORMAL GRAMMAR, i.e., excluding meaning from
grammar. First argument: meaning is often very vague and meaning categories are not easily
delineated. What might seem to be obvious semantic categories are often in fact definable only
in terms of formal features of a language (to this extent Whorf may have been right). If then,
grammatical categories are given semantic definitions, the definitions are circular. An excellent
example: Noun = word used for naming anything. What is anything? The things that can be
named, include, in English, fire, speed, place, intelligence, suffering, etc. Why is "redness" a
name of a thing, whereas "red" is not? Similarly, why does "rain" refer to a thing, while "It's
raining" does not? It is reported that there are languages in which words for 'river', 'spring', etc.,
are essentially verbs, so a literal translation would be "It's rivering" rather than "There's a river".
So the painfully simple answer is that 'things' are what are designated by Nouns.

Second argument: Even where we can define semantic and grammatical categories
independently, they often do not coincide. One of the best known examples is that of wheat
and oats, where there is a clear lack of correspondence between grammatical number with
numerical quantity. In terms of 'one' and 'more than one' these can't be distinguished, but "wheat
is in the barn" "oats are on the table" --> grammatical. Similarly, hair is singular in English,
but French (cheveux) and Italian have plural nouns. Similarly, gender and sex are independent
in German [also Hindi], German for young woman, “Mädchen” and “Fräulein” are neuter, while
the feminine “la sentinelle” in French may refer to a strapping male. In English, tense is not
directly related to time since past tense is used for future e.g. If he came tomorrow...

Hence, the basic grammatical categories of a language have to be established independently of

their meaning. However, once we have defined the formal categories of grammar, we find that
there is some correlation between gender and sex, tense and time, grammatical number and
enumeration, though this correlation will never be exact. In French, nouns referring to feminine
 are always feminine, though some feminine nouns may refer to males, and in English ONE of
the functions of tense is to refer to time. Indeed it is only because of these correlations that the
labels 'gender', 'tense', etc, have any usefulness at all.

However, it is hard to draw a distinguishing clear line between grammar and semantics; since the
more detail we consider grammar, the closer the correlation with semantics. For instance, in *
John is seeming happy, we could say this is ungrammatical on the grounds that the verb seem
does not occur in the progressive (continuous) form is seeming. But is this in fact a grammatical
rule or is it the case that for semantic reasons that John cannot be in a continuous state of
seeming? Actually, there is no clear answer. Similarly,
* John is having gone there
* John continued having gone there
Little doubted, the first is ruled out grammatically by a simple rule that puts the auxiliary have
before the auxiliary be and so allows only _John has been going there_. But it is not clear
whether we ought to say that there is a grammatical rule that prevents continue from being
followed by the auxiliary have, or whether we should say it makes little sense semantically.
Hence,
1. Though we can, and must, set up formal categories, they will be found to have some
correlation, but not one-to-one with semantics.
2. There is a difficult borderline area.
also:
3. Some of the categories seem to be found in all languages; as far as we know, there is no
language that does not distinguish in some way between N and V, even though some may not
have different word-classes (parts of speech). Some of the more imp semantic categories, e.g.
relating to sex, quantity, time, are found in many grammatical systems. Similarity between
languages support some kind of universalist view.
The controversy abt semantics and formal grammar was revived between in terms
of 'interpretive' and 'generative' semantics within
tranformnational-generative-grammar. Chomsky 1965p.16 had argued that there
is a syntactic [deep structure] and that it is at this level we can relate
active and passive sentences, or Question forms. The diff in the order of
the words (as well as other differences) is a matter of their [surface
structure].

deep structure also distinguishes John is eager to please from

John is easy to please : John is the 'deep subject' of please in 1 and
the object of please in 2; also while J is the subject of is in 1, the
subj of is in 2 is "please John"

Part of the syntax is concerned w transformation rules that convert deep

structures into surface structures. Essential that given the DS, these rules
automatically generate the correct SS. The DS is generated by the BASE which
consists of two components: CATEGORIAL COMPONENT (the entire grammatical
apparatus) and the LEXICON (inventory of lexical items).

DS: - enables generation of surface structure, e.g. To oversimplify,

[Passive] marker will generate Piano played by J while its absence
generates J played the piano, all else being same. same DS with Q
will generate Was piano played by J?

- can generate the semantics

Generative Semantics 122

However, others argued that if there is a deep structure, it must be much

deeper, so deep in fact that it is essentially semantic and not syntactic.
In this sense semantics is not interpretive, but generative.
One of the many complex arguments: the active/passive relationship seems to
break down with
Many men read few books vs
Few books are read by many men
which are clearly diff in meaning. To anal such sentences as having same DS
except the [Passive] marker, is clearly not satisfactory. Lakoff 1971
p.238-45 argues that the DS must be the sem structures, which specify the
meanings. A rather diff argument [Lakoff 1968] suggests that
Seymour sliced the salami with a knife
shd be related to
Seymour used a knife to slice the salami
and that they have essentially the same DSs. Fillmore's case theory is also
within generative semantics. 123

Gen Semantics was doomed to failure, because of the impossibility of using

semantics as a basis for grammar. But Chomsky is still maintains that syntax
is AUTONOMOUS - i.e. indep of semantics [Chomsky 1977 p. 36-9]. 124

6.2 Grammatical categories 124

GENDER: Fr and German may have historical reasons for some idiosyncracies
that were conventionalized... English has no gender except he,
 she, it. Latter can be applied to a dog of unknown gender, but
not to indeterminate-sex humans. they, them, their can be used
- e.g. has anyone lost their hat? If anyone comes tell them to go away._
This is frowned upon by some grammarians. For babies,
sometimes it is used, but not politically wise.

Swahili: animates, small things, and big things, each marked by

appropriate prefix and requiring agreement with adjs and verbs.
(etymologyically, gender is not related to sex, but merely means
'kind'; as sex took on erotic qualities in 20th c. English,
gender came to be used, often in feminist writing, for biol
qualities etc., 1963+) `

But of course, semantic distinctions are impure for all such classes
- e.g. Bloomfield: Algonquin lg have animate/inanim distinction, but
"kettle" and "raspberry" are animate, though "strawberry" is
inanimate.

NUMBER: Many lgs have number system, but many in SE Asia or Africa do not.

Similarly for number. What is the semantic importance of "one" vs

"more than one" not clear. Also classic Sanskrit, Greek Arabic had
also "dual". Fijian and Tigre (Ethiopia) also distinguish little
plurals and big plurrals.
No "natural" numerical classes to be found that may be expected to
crop up in most languages.

COUNT/MASS NOUNS: English clearly makes this distinction, though grammar

books often ignore it.
count nouns: cat, book; take "a __" in singular
mass nouns: butter, petrol. "*a butter"; "some butter" " butter is"
cake, fish --> belong to both classes.

no good justification why butter is a mass noun but jelly can be

count as well as mass. Why can we refer to a single mass of jelly as
a jelly but not to a mass of butter as a butter?
Again, while cake can be both mass and count, bread is only
mass. [in olden times, no sliced bread?]

can interchange:
a butter, a petrol = a type of butter or petrol;
 a coffee, a beer = cup of coffee; glass of beer
chicken is white meat; The chinese eat dog : mass N

PLURAL: mass Ns are closer to plurals, e.g. in Bilin, the word for water is a
plural.
English: scissors, trousers, tongs etc - formally plural, but
countable.
pronouns - I, we, 2P you, 3P he she it they: combination of deictic
categ, gender, and 1,2, or 3P = speaker, hearer, non-participant.

problematic: 1st person plural: (We in English)

rare for there to be several speakers. - more often refers to both
speaker and hearer, or speaker + non-participant
Similarly, 2P pl - you - may refer to hearer pl or to
hearer+non-participant.
rule for the pron plural: determined by the highest ranking person
involved, and 1p > 2p > 3p
some lgs: distinctive forms for "I and you" or "I and he/she"
sometimes pl forms used for politeness [German sie?]

ARTICLE: (mainly for English)

definite article the: single identifiable item in either lg or
non-lg context; or the most familiar ("the govt" = of our country),
"the garden" = our garden. But in context, may refer to other govts
or gardens or moons.

if item becomes uniquely identifiable, article may be dropped.

E.g. Parliament not the parliament; similarly bank rate.

idiosyncracies: rivers take articles, but cities do not.

Fries 1952: proposes grammar w four parts of speech: N V Adj Adv (though he
doesn't use these terms) - along with 145 sets of "function words".
The Fn words are exemplified by :
the, may, not, very, and, at, do, there, why, although, oh,
yes, listen! please, let's

The Engl conjunctions after, when, while, if, if translated into

Bilin (Ethiopian Cushitic lg) will be inflected on the verb.
Finnish has complex case system; not only nom, acc, ablative etc, but also
elative (out of), illative (into), adessive (on), essive (as).
6.3 Grammar vs Lexicon 130

GRAMMAR vs LEXICON : often posed in terms of whether deviant sentences are

ruled out because of grammar or lexicon. e.g.
* The boys is in the garden - violates only one grammar rule
* Been a when I tomato - violates what?
In contrast:
* The water is fragile : semantic violation

Semantically ill-formed sentences

arguments for separation: some sentences can be clearly grammatical, but

lexically meaningless (e.g. Chomskys C G I S F). Incidentally,
Carnap had made a similar point using Engl syntax but nonsense words:
Pirots karulize elastically (1937:p.2).

Grammar must also argue for illegitimacy of sentences like C G I S F.

Some linguists believe that a formal grammar can rule out such COLLOCATIONAL
possibilities also; [Joos 1950, 1958:356] argues that all the collocationjal
possibilities of a word would be suff to characterize it linguistically. 132

Even Chomsky (1965:95ff) attempts to handle collocational possibilities

within grammar, ruling out sentences like
*The idea cut the tree,
*I drank the bread, etc.
These are diff from other invalid Ss like
*He elapsed the man
(elapse is intransitive, takes no object). Chomsky proposes that a
SELECTIONAL RESTRICTION mechanism similar to the latter class can be used
for the former - cut would require a 'concrete' subject, and drink a
'liquid' object.

But this argument is unsatisfactory.

1. limitless num of components needed - all relevant information must be
included in the classes --> infinite
2. fails to account for legimtimate uses where selectional restrictions are
violated - e.g. hypothetical : John thought we could drink bread,
 or with negation: You can't drink bread.

Questions the justification for attempting this even, by introspection "as

native speakers", our reactions to these classes of errors is different.
But there are Ss which are illegit, but the violation cannot be clearly
marked as grammatical or lexical.

e.g. The dog scattered : scatter normally occurs w plurals or collective

Ns: (The dogs/herd scattered). So the violation appears
grammatical. However, imagine a dog that
can break itself up magically and scatter over an area? This would
be permitted, but this time on lexical grounds. But should we not
say The dog scattered itself even in this situation?

6.4 Grammatical Relations

"sleep in" functions as a unit - The bed's not been slept in.

Is the [deep subject] the 'doer' or 'actor and the deep object the 'sufferer'
(or patient)?
problem w verbs that are not of action but of state: I like ice cream or
I saw the boys. More so, e.g. from Halliday 1970:147
General Leathwell won the battle
but what did GL do? fire guns, kill enemy, etc? Or did he quietly sit in HQ
and let the troops do the work? Was he an 'actor' or 'supervisor'?

RELATIONAL GRAMMAR: linguists have argued for such structures (not using the
term DS/SS) as universal.

Passive transformation:
Some lgs do not move the NPs at all - e.g. Hindi [Johnson 1974:271)
rAm ne moTer chAlAi (Ram drove the car)
rAm se moTar chAlAi gai (The car was driven by Ram)
It is only in terms of Subj and Obj that we can make any general, universal
statement about passive and active; in all cases, the obj in the active voice
becomes the subj in the passive

So notions such as subj and obj appear to be useful in many lgs.

Problem: Ergative lgs like Basque or Eskimo.
'Subj' of intransitive and 'object' of transitive are in the same
 (nominative or absolutive) case, while 'subj' of transitive is
ergative. (Subj and obj in quotes because they may not be approp to
these lgs). example from Basque

Transitive as CAUSATIVE:
marked in some lgs like Tigrinya (Ethiopia)
Engl trans and intrans cook are handled by cuire and faire cuire.
Engl has some historical causatives - e.g. He felled the tree --> caused it
to fall.

Some linguists have argued for the transitive distinction being looked at in
terms of causativity. John rang the bell = 'J caused the bell to ring'.
1. But diff between semantic analysis of Engl and formal features in Tigrinya
or French
2. Lgs have causatives for transitive vs as well - e.g. Tigrinya: sabbare
he broke (something) and asbare he cause (someone) to break (sthing).
Both Fr and Ti [and Hindi, but not Beng] wd translate Engl show as
'caused (someone) to see'.
also - what can be a causative analysis for John hit Bill? (John caused
Bill to be hit) 139

6.5 Components and the sentence

DERIVING SEMANTICS (Katz/Fodor) 140

worth looking at because no one else has made such a clear and detailed
proposal [pre-Montague]
set of PROJECTION RULES that combine (AMALGAMATE) meanings (PATHS)
example of K/F:
the man hit the colorful ball
"colorful ball":
colorful has paths for marking an actual colour, or for the colourful
nature of an aesthetic object (evaluative)
ball has three paths - one as the social activity [dance?], the other with
as (phys object) but distinguished by [having globular shape) and
[solid missile for projection by engine of war] (cannon ball).

But all amalgamations wd not work, e.g. aesth object marker would not bind w
phys object (Palmer washes his hands off with the factual accuracy of this
argument, saying let's go with the argument). So there are only four
combinations - social dance is colourful, or has a colour, or phys ballx2 has
a colour.

hit two paths: colliding with, or striking phys object.

Again this rules out the dance, so we are left w four readings:
[colliding / striking] a [ball / cannonball] that has colour

Problems w such rules. E.g. pregnant would be marked [-male] but how to
permit pregnant horse where the latter is not marked for gender?
Similarly, pretty child, buxom neighbour, etc.
can resolve with what [Weinrich 1966:429-32] calls "transfer feature" -
transferring a feature to another word. But not simple.

6.6 Predicates and arguments

In a S the V is often best seen as a relational feature - active/passive can

be treated as relational opposites [5.5].

Analysis in relational terms seems to offer a far more satisf solution to the
problem of sentence meaning than componential analysis. In essence such
analysis will have much in common w predicate calculus.

walk: W(x)
love: L(x,y)
give: G(x,y,z)

Then can define relations e.g. father (x,y) as 'parent of' and +male, as
father (x,y) := parent(x,y) ^ male (x)

Predicate calculus permits SUBORDINATION - where a whole phrase acts as an

argument - e.g. Fred thinks that John loves Mary where J l M is acting as a
phrasal argument to [think]:
[think] (x, [love](y,z))

causal interpretation of transitives:

Killed (J,M) --> [cause](x, [become] (y, [~alive](y)))

These can be mapped to trees.

Generative semanticists argued that such trees were in fact its deep
structure. e.g. I almost killed him - almost can qualify cause (shot
at, but missed) become (recovered) or ~alive (shot him, and he was near
dead). (last two: distinguishes "nearly became dead" from "became near dead")

However, John killed Mary is not identical w _J caused M to become not

alive_.

6.7 Case Grammar

Case grammar was proposed by Fillmore as one of the arguments in favour of

generative semantics, but is best understood as a version of an analysis in
terms of predicates and arguments, w emphasis on the function of the
arguments.

three Ss with open:

J opened the door with a key :
The key opened the door
The door opened
J = [agentive]; key = [instrumental] and door = [objective]

[Fillmore 1968] suggests six cases:

- AGENTIVE ('typically animate perceived instigator')
- INSTRUMENTAL ('inanimate force or object causally involved')
- DATIVE ('animate being affected') EXPERIENCER (Fillmore [1971a])
- FACTITIVE ('object or being resulting from the action or state') RESULT
- LOCATIVE ('location or spatial orientation')
- OBJECTIVE ('the semantically most neutral case')
Later, Fillmore 1971a uses EXPERIENCER for dative and RESULT for factitive,
and adds:
- COUNTER-AGENT added as 'the force or resistance against which the
action is carried out
- SOURCE ('the place from which something moves')
- GOAL
and talks of AGENTS and OBJECTS instead of agentive and objective.

Case grammar can capture the inverse relation of teach and learn; and also
account for the difference between J ruined the table and J built the table.

hierarchical ordering of cases: Agent > Experiencer > instrument > object;
i.e. if agent is present, it will be subj, else ...
can't say: * John and the key opened the door or
* The key opened the door by John
* The door opened w a key [can't see why not]
Only if the object is alone can it be the subject
The door opened

In later work [Fillmore 1977] suggests that J smeared the wall w paint or
J smeared paint on the wall; or J sold a book to H or H bought a book from
J; these are changes in PERSPECTIVE and not case.

strange cases:
I saw Helen and a football match
(both are objects, but diff kinds; ?ZUEGMA?)
The smoke rose (smoke is object)
The smoke rose and blotted out the sun (smoke = both object and agent)

What is agency?
The wind blew and opened the door
(is the wind an agent or an instrument?)
agents must be animate and also deliberative? What of
the virus destroyed the cabbages vs
the slugs destroyed the cabbages

AMBIGUITY in case assignments, particularly difficult example:

My ear is twitching
ear could be the agent (since it is doing the twitching) or the experiencer
or even the locative? 149

--6.8: Interrogative / Imperative

Declarative: J shut the door.

Interrogative: Did J shut the door?
Imperative: Shut the door.

But forms can be deceptive:

1. I want to know where you have been

I insist you stay the night
2. Have you heard the news that we won?
Haven't I told you not to do that?
 3. Understand that I can't do that
Tell me what you have done.

Instead of Decl, Intr, and Imp, maybe we should use statement, question
and command for these functions. (1) are decl but not statements etc.
(cf [Bar-Hillel 1970:365]).

intonation as a formal mark of the question. e.g. John's coming? (with

rising tone). Trying to incorporate intonational features into grammar is
attractive but problematic.