Transitivity, Argument Structure and Syntax

Transitivity, Argument Structure and Syntax
The current trend in linguistics is to focus on argument structure, with the
hope that if syntactic behaviour of arguments of the verb (e.g. arity, case
marking, subcategorisation) can be systematically related, we would see
the effects of that in full-blown syntax.
CCG attempts to address the same question, but from the other end: once
we understand the combinatory limitations of syntactic and semantic de-
pendencies, then we can attempt to narrow down the systematicity in ar-
gument structure which would be consistent with that behaviour. That’s
because CCG shows how phrasal (sentential) logical forms and phonolog-
ical forms are systematically related without extranous devices.
It might also be possible to explain the argument structure if the combi-
natory limitations can narrow down all and only the argument structures
attested in human languages.
1

---

“All” part is an empirical question, showing the modeling potential and de-
scriptive adequacy of the theory.
“Only” part is where linguistics become a science with predictive power,
leading to explanatory adequacy. If this way of thinking is right, it’ll be a
computational science.
We are nowhere near that goal, but the important thing is to put forward
explicit, testable, and falsifiable hypotheses, and to be able to reveal more
data to look at, and new ways of looking at it.
The following is one way of doing this in CCG.∗
∗From a very early draft, Bozsahin and Steedman 2006; please DO NOT CITE.
2

---

Let us consider the knowledge of transitivity to be part of Universal Gram-
mar, as proposed by Hopper and Thompson (1980, 1984):
There is gradience in transitivity.
The core cases are the determinants of lexical categorization.
Lexical categories such as N and V are not part of universal grammar
(by implication, neither is VP nor NP).
We also follow Edmund Husserl in assuming that all categories are seman-
tical in nature (cf. Rosch 1978 for psychological basis of categorization).
3

---

Their origin may be further traced back to a discourse basis, as Hopper
and Thompson argued, but their combinatory specialization can safely
start with the assumption that in a core transitive activity there are two
participants, which must be distinguished linguistically, which to us means
distinguishability that starts at the level of representation, viz. the logical
form.
We claim that all other verb classes can be derived from a representation of
core transitivity, along with relative markedness of functors and arguments.
4

---

The semantic basis of transitivity can be captured by the universal seman-
tic type es → (eo → t).
es is a thing high in potency and eo is a thing high in affectedness, in Hopper
and Thompson’s (1980) terminology.
The syntactic category of this type, viz. T −1(es → (eo → t)) according to
the principle of categorial type transparency, is expected to show cross-
linguistic variation due to the nature of T −1.
We do not assume that T −1e is NP universally; it can be N, or anything
that can be distinguished from syntactic types of propositions.
The types e and t are not universal either; all that matters is that there be at
least two primitive types a and b such that a = b if and only if T −1a = T −1b.
5

---

We can now conceive the possible function-argument asymmetries in es →
(eo → t) to be all combinations of the three syntactic objects and their as-
sociated semantics in (1a), where TV is the syntactic category of a basic
transitive verb, and pred is its normalized semantics.
There are exactly three ways, enumerated in (1b–d), in which we can es-
tablish a functor-argument relation syntactically on these three objects.
NB. such a relation is required by the principle of combination, by which all
functional categories must make functor-argument distinction because of
typing.∗
∗Notice that many of the purported LFs are impossible due to typing, e.g. o pred a .
Moreover, Condition C effects discussed earlier allow one LF among the well-typed can-
didates: pred o a .
6

---

(1)a. TV: pred
NPA: a
NPO: o
b. S{|NPA, |NPO} = TV: λ{a ,o }.pred o a
c. S{|TV, |NPA}: λ{pred ,a }.pred o a
d. S{|TV, |NPO}: λ{pred ,o }.pred o a

---

The first alternative (1b) leads to active systems.
The second to accusative systems (1c).
And the third to ergative systems (1d).
This conjecture relies on observations of Hopper and Thompson (1984)
over a large inventory of languages and a wide array of constructions:
They report that when a noun takes a noun as an argument (e.g. in noun
compounding and nominal predication), the argument noun is “categorially
deficient”, that is, it tends to lose things associated with the category, for
example inflections.
7

---

Similarly, when a verb takes a verb or a clause as an argument (e.g.
purpose clauses, predicate adjectives, clausal complements, verbal com-
pounds), the argument verb (or clause) is categorially deficient. For ex-
ample purpose clauses are universally very low on categoriality; in most
languages they are infinitives and lack tense-aspect-person morphology,
or otherwise marked as irrealis (ibid., p.739).

---

In (1b), the transitive verb takes two nouns as arguments, therefore there
is no function-argument asymmetry among participants, and the nouns are
not low in categoriality with respect to each other.
This is witnessed in active languages where both arguments are usually
inflected on case.
In (1c), the object takes the transitive verb and the agent NP as arguments,
therefore there is a function-argument asymmetry among nouns, and the
agent noun is low on categoriality.
This is witnessed in accusative languages where the syntactic subject
(the argument that the agent maps to) is usually unmarked (nominative),
whereas the object is high on categoriality; it has an overt case (accusative).
8

---