WHY THE NOTION OF LEXICAL TEMPLATE?
Ricardo Mairal Usón,
UNED - Madrid (Spain)
2 An overview of the
most relevant approaches to lexical representation
3 The format of a
lexical template
4 Lexical
templates in the primary lexicon
4.5 Cessative
phase of verbs of existence
5 Constraining
the generative power of a lexical template
Lexical
knowledge representation has become one of the crucial issues in linguistic
theory today. Generalizing a bit, one finds two major schools of thought: (i)
the syntacticocentric approaches[1] and (ii) the constructional
approach. Within the former, subclasses can be differentiated on whether they
employ natural language phrases (Dik, 1997), a metalanguage (Jackendoff, 1990;
Rappaport/Levin (1998); Tenny (1994), Van Valin and LaPolla (1997), Wierzbicka,
(1992, 1996, 1999)), or a set of thematic roles (Fillmore, 1968, Chomsky, 1981,
etc.).
Most theories on lexical representation, with the
exception of the work done in Construction Grammar, propose a set of
grammatically salient features as the central and unique components in a
lexical representation theory, based on the assumption that only those aspects
which are grammatically relevant serve as the input for grammatical processes.
In consonance with this, Pesetsky (1995: 14), in his analysis of speech act
verbs, and Grimshaw (1993: 3), in her study of verbs of color, claim that the
parameters “loud” and “soft” or the color parameters, though important in some
other respects, do not have any role in grammatical processes. Seemingly,
Levin/Rappaport (1996a) contend that the best way to isolate those properties
of the lexical class of sound emission is to test the behaviour of these
predicates against the Unaccussative Hypothesis. However, I believe that this
is an oversimplification in the sense that although there are certain semantic
patterns which do not actively interact in the different structural
realizations, there are others which highly constrain the different syntactic
configurations, e.g. the manner vs. result constants. This is not an excuse to
overgeneralize and exclude all the semantic parameters in one blow.
Consequently, the approach presented in this paper
diverges from these proposals in that in my opinion, lexical representations
should be enriched with a much more robust and powerful semantic component. As
is claimed in the most recent linguistic literature, lexical representations
should capture those aspects of meaning which are grammatically relevant.
However, this is done at the cost of sacrificing the meaning potential of a
predicate. In this respect, a notable exception is the work of Construction
Grammar, which formulates an encyclopedic representation of the meaning of
predicates. Yet, this is accomplished at the cost of designing a linking
algorithm which loses a lot of the explanatory force as represented in the more
syntacticocentric models.
Then, the aim of this paper is to present an
alternative formalism which accounts for the full set of parameters which
constitute the meaning of a predicate. We use the term lexical template to
refer to this new lexical notational device. For the purposes of
exemplification, this paper focuses on the internal structure of lexical
templates by analyzing five lexical classes of verbs, viz. contact-by-impact
verbs, break verbs, consumption verbs, sound verbs and existence
verbs. The internal lexical complexities which revolve around these five
classes are discussed. Finally, the generative power of lexical templates is
brought to the fore.
In order to
evaluate the extent to which a lexical template is a departure from most
current approaches to lexical representation, let me reproduce the format
proposed for some of the predicates under analysis in this research monograph.
I shall begin with Fillmore’s (1968; 1970: 126-ff) representations, which
illustrate the view that the meaning of a predicate can be reduced to a set of
unanalyzable semantic notions called thematic roles, which are defined
independently of the meaning of a predicate. These constructs indicate the type
of relationship each of the arguments bear with respect to the predicate.
Suffice the representation of the predicates hit and break:
(1) hit: [Agent, Instrument, Place]
(2) break: [Agent, Instrument, Object]
Role-centered
representations have received abundant and well observed criticisms since its
first formulation[2]. Although thematic roles are
implicit in practically all linguistic models (with the exception of Ravin
(1990)), their explanatory potential has been diminished since they are no
longer regarded as primitive notions. In fact, in order to circumvent the basic
problems inherent in thematic roles, it was suggested that these should be
defined over predicate semantic decompositions. This means the development of well-articulated
theories of lexical representation.
One of the
central features common to many theories of lexical representation is the
notion of event. The basic idea is that since verbs are taken to denote events,
it is assumed that the principles underlying the lexical semantic
representation of verbs derive from the type of event structure. In connection
with this, there have been several proposals which take as the central
corollary the notion of event and develop a lexical representation theory in terms
of the conceptualization of this notion. For example, Jackendoff’s (1983, 1990)
logical structures are based on a localist conception, such that all events
involving location and motion are central for the construal of events. He uses
the predicates GO, BE, STAY, and CAUSE to encode the underlying properties of
motion events, and the two types of location events, viz. stative and eventive,
and their corresponding causatives respectively. He posits the Thematic
Relations Hypothesis in order to account for certain instances of systematic or
regular polysemy; hence the polysemic nature of a predicate like keep is
accounted for by the fact that all of the different configurations can be
explained by the functions CAUSE and STAY and their differences emerge from the
different semantic fields involved:
(3) keep: [CAUSE, (x, (STAY y, z))]
A second line of research which strongly emphasizes
the notion of event is that group of theories which make an extensive use of
the notion of Aktionsart – the internal temporal properties of a predicate – to
determine the event structure representation of each predicate; for example,
Dowty (1991), Tenny (1994), Van Valin/LaPolla (1997); Levin/Rappaport (1995,
1996a), etc. Drawing on the pioneering work of Vendler (1967), predicates are
classified according to the type of aspectual properties the event designates.
In this regard, Van Valin/LaPolla (1997: chapter 3) propose an inventory of
logical structures based on the type of event designated by the predicate. For
example, predicates like break, kill or destroy designate a
causative accomplishment whereas a predicate like drink can designate an
activity or an active accomplishment structure depending on the referential
nature of the NP:
(4) kill [do’ (x, f)] CAUSE [BECOME dead’ (y)]
(5) break: [do’ (x, f)] CAUSE [BECOME broken’ (y)]
(6) destroy: [do’ (x, f)] CAUSE [[do’ (y,f)] CAUSE [BECOME destroyed’ (z)]]
(7) drink: [do’ (x, [pred’ (x, y)]) & BECOME consumed’ (y)]
In much the same vein, Levin and Rappaport (1995, 1996a)
and Rappaport/Levin (1998) propose an event structure representation for each
predicate; hence a predicate like drink is assigned an activity event
structure representation, while break a causative accomplishment
interpretation:
(8) drink: [x ACT <DRINK> y ]
(9) break: [x ACT] CAUSE [BECOME [y <BROKEN> ]]
Within the
group of those theories which posit a system of lexical decomposition, it is
worth noting another proposal which, unlike the preceding ones, does not resort
to any sort of metalanguage but instead it constructs the argument structure of
a predicate using natural language phrases. A case in point is Dik’s (1978)
procedure of stepwise lexical decomposition:
(10) hack
[V] (x1: animate)Ag (x2: object)Go
df = cut [V] (x1)Ag
(x2)Go (x3: pieces
: uneven)Result?
(s1:way [N]: rough [A]: violent [A])Manner
If there is
a factor shared by all these representations, it is the fact that all of them
postulate notions – thematic roles, argument positions, causal chains, etc. –
to which mapping rules can make reference to. In this regard, Tenny (1994:2)
makes this point clear in his Aspectual Interface Hypothesis, when he affirms
that “Only the aspectual part of thematic structure is visible to the
universal linking principles”.
As
previously mentioned, this is not necessarily incompatible with the claim that
those factors which do not have a direct role in the formulation of mapping
rules should be part of the lexical representation. In this regard, one of my central
claims is that these structures still need further semantic decomposition, and
this entails the inclusion of an enhanced semantic component, which necessarily
goes beyond the present scope of the various lexical representations presented
thus far. Furthermore, the fact that these theories postulate a separate
lexical entry for each syntactic configuration signifies that it makes no
allowance for information shared by sets of predicates, and as a result, is
unable to account for regularities such as the distinctive sets of syntactic
alternations that characterize certain classes of predicates.
As things
stand, from these structure it would be desirable to work out a more complete
representation in such a way that the set of semantic and syntactic factors
that hold within a lexical class could be easily be represented into one
unified structure. This claim has been echoed from some linguistic paradigms
like RRG:
“Many aspects of the meaning of a verb [the specific requirements that a verb imposes on one or more of its arguments] would be represented in a full decomposition, but given that no such representation exists at present, they will have to be stipulated for the time being” (my own emphasis) (Van Valin and LaPolla, 1997:156)
In much the same way, Levin and Rappaport (1996b: 58)
also recognize as an urgent need to analyse the way constants (or that part of
the meaning of a predicate which is idiosyncratic) constraint the event
structure representations[3]. As the
authors affirm, this observation is also applicable to more encyclopedic
approaches like that of Construction Grammar:
“Hardly any attention has been paid to the
nature of what we have called the “constant” and the constraints on the
association of constants with event structures. Since the multiple association
of a constant with event structures is taken to be constrained by compatibility
between the two, only a more careful study of the nature of the constant and
the ways in which this determines the event structure it is associated with
will help answer the many open questions related to multiple argument
expression.” (my own emphasis)
Seemingly,
Jackendoff (1996: 118-119) explicitly admits this sort of compromise solution
in his logical structure although he claims that:
(…) and all the extra stuff beyond the
perceptual-motor differences must be represented in the meaning of these words.
But how? In what format? I don’t have a formalizable theory of these aspects
of meaning yet (….) [my own emphasis]
As becomes
clear from the preceding passages, it is necessary to investigate the way
semantic aspects (thus far excluded from lexical representations) interact in
the realization of structural configurations. In contrast to a deeply-rooted
view that these aspects did not have a grammatical impact, a new conception has
arisen such that some linguists have found out that these aspects of meaning
can be a useful tool to provide answers to many of the unresolved issues which
follow from a purely syntacticocentric approach in the design of a semantics-to-syntax
linking algorithm[4].
With this
theoretical credo in mind, the notion of lexical template grows out as an
attempt to formulate a proposal along these lines. Such a template means a
fully semantic decomposition of the meaning of a predicate together with the
inclusion of those syntactically salient aspects. Moreover, each lexical
template is not meant to represent the lexical properties of individual
predicates but conversely it contains and captures the full set of linguistic
features as encoded within a lexical class. This makes it possible to come to
grips with the identification of those semantic parameters which define a set
of predicates.
By way of
example, after an analysis of the lexical entries proposed for hit
above, I conclude that a full set of semantic parameters which play an active
role in the meaning of this predicate are absent in these representations;
manner, result, type of blow, instrument, reason, etc. I have extracted these
parameters by looking at the lexical class to which this predicate belongs as a
whole, viz. contact-by-impact verbs. Furthermore, following
Levin/Rappaport’s quotation, the notion of lexical template is not only
applicable to the more syntactically-driven type of representations but it also
comes to complement constructionist representations where constants, though
implicit, have not been fully developed.
In the previous
section, I have discussed some of the unresolved issues in relation to the internal
structure of a lexical representation, namely, the lack of a more fine-grained
semantic decompositional system. Let me focus on more examples I have come
across in the analysis of the corpus.
Firstly, if one compares the structure for cooking
verbs, destroy verbs and verbs of killing, viz. a causative
accomplishment, we have the same representation with the only difference being
stated in terms of the resultative predicate, dead’, cooked’ and destroyed’
respectively:
(11) kill: [do’ (x,f)] CAUSE [BECOME dead’ (y)]
(12) cook: [do’ (x,f)] CAUSE [BECOME baked’ (y)]
(13) destroy: [do’ (x,f)] CAUSE [[do’ (y,f)] CAUSE [BECOME destroyed’ (z)]]
The question that arises is the following; to what
extent do these representations tell us something about the multifaceted nature
of these predicate meanings? How can one capture the difference both in meaning
and syntactic productivity of these two classes of verbs? The easy answer is to
affirm that both differ in the type of primitive involved. However, that does
not tell us anything new at all. Furthermore, an analysis of the
configurational patterns of these predicates is very revealing in the sense
that their syntactic behavior differs substantially. Cooking verbs can
occur in the following constructions: causative / inchoative and middle
constructions, while these are excluded in the other two lexical classes.
Moreover, destroy verbs cannot occur in a resultative construction,
while verbs of cooking and killing can. Thus, how can we account for these syntactic
contrasts? Can one provide any theoretical principle which explains this
striking syntactic behavior?
By looking at the semantic class as a whole, one can
figure out some tentative and initial answers to some of these queries. For
example, one could argue that both destroy and kill are verbs of
existence in contrast to cook, a verb of change of state, an assumption
which explains why the first two group of verbs cannot occur with the middle,
while cooking verbs can. What seems to be clear is that a glance at the
semantic potential of a predicate, regardless of the fact of whether the
semantic parameters encoded are syntactically relevant or not, turns out as a
useful tool to elucidate some syntactic differences. Then, why shouldn’t one
exploit this line of research any further? This is precisely the bottom
theoretical line behind the notion of lexical template; lexical templates are
bound to lexical classes.
The notion of lexical template proposed in this paper
feeds upon some of the major proposals in lexical semantics; hence it is an
eventive aspectual theory in that the notion of event structure is represented
in terms of the temporal internal properties of the predicate. The internal
structure representation of the event is further fragmented into several causal
chains which explain and justify the existence of the whole range of
concatenated events which hold within a single event. As shown elsewhere, the
bracketing which signals the different phases of an event is not accidental
since it has important syntactic consequences. In this regard, a lexical
template bears some resemblances with some of the axioms postulated in a causal
approach (cf. Croft, 1991). Finally, one of the points of divergence with
respect to the great bulk of theories of lexical representation concern the
inclusion of an enhanced semantic component, as I have made it abundantly clear
in the first part of this paper. This semantic component is the result of the
research work presented in Faber/Mairal (1999), who divide the lexicon into
coherent hierarchical semantic classes, each being represented with a full set
of semantic parameters.
If I claim that lexical templates are bound to
specific lexical classes, this presupposes a twofold distinction in the
lexicon; one corresponding to the lexical template – which encodes the
conceptual substance common to the members within a lexical class – and the
actual forms of the lexical entries themselves – which inherit part(s) of the
information in the template[5].
It is important to emphasize that, unlike some other
functionally-oriented proposals such as Dik (1997a: 78-103), the notion of
lexical template uses abstract predicates in its semantic decomposition. In
order to obtain important cross-linguistic generalizations it is thus necessary
to resort to some sort of universal metalanguage which can account for the
actual similarities which hold among different lexical classes. The use of
natural language phrases precludes the establishments of such generalizations.
In this regard, one of the most problematic issues is the right-grain size to
use in the identification of the type of primitives involved in a lexical
representation. In order to solve part of this problem, I resort to the use of
an ontological module. The ontological model, independently of how this is
finally conceived – in computational or more lexicographic terms - converges on
the use of a metalanguage as a descriptive device for lexical representations.
In the ontology framework, it is claimed that natural language sentence can be
reduced to a formalized, language-neutral representation, a Text Meaning
Representation.
In sum, the
format of a lexical template lies somewhere between the Aktionsart characterization of lexical units as proposed in RRG and
to a lesser degree in Rappaport/Levin (1998) on the one hand, and the richer
semantic description as postulated in the Functional Lexematic Model (FLM)
(Faber/Mairal, 1999).
In
reference to its internal structure, a lexical template consists of two major
types of variables: internal and external variables[6]. The notational device adopted to
distinguish one from the other is that internal variables are marked in Greek
letters while external in Roman letters[7]. This is methodologically coherent
with my claim that a lexical representation should contain those aspects of the
meaning of a word which are grammatically relevant (external variables) and
those aspects of the meaning of a word which reflect the idiosyncratic features
relevant to each predicate (internal variables).
These two
basic building blocks coincide with the distinction introduced in Rappaport and
Levin (1998) and Grimshaw (1993). The first two authors argue that a verb
meaning is composed of a set of primitive predicates and constants. The
primitive elements encode the core meaning and the event type of the verb and
thus define the broad lexical semantic classes of verbs. The constant, which
has an ontological status (e.g. stuff, place, manner, etc.), specifies the
idiosyncratic information to each individual predicate. Thus, verbs which share
the same structural elements (that is primitives) are differentiated from each
other in terms of the type of constant specific to each class member. In much
the same vein, Grimshaw (1993) makes a distinction between semantic structure
and semantic content. This author further suggests that certain verb arguments
are structure arguments by virtue of the association with one of the open
positions in the event structure representation and in turn these serve as the
input for linking rules and for the determination of argument expression.
Content arguments are identified with the type of constants. Grimshaw
hypothesizes that this division may explain why certain arguments are assigned
semantic roles while others cannot; the latter correspond to the content
arguments.
Following
Grimshaw, external variables are those aspects of the meaning of a word which
are syntactically realized and are identified with a configurational position
in the syntactic core. Internal variables are the product of extracting the
semantic parameters which define the meaning potential of a whole lexical
class. This distinction is also very similar in many respects to the
constructional view which proclaims a constructional meaning – equivalent to
our external variables and event structure – and a core verb meaning –
equivalent to the set of internal variables.
Internal variables are classified along a scale of
accessibility to argument realization. At one extreme, we do find a group of internal
variables which are not strictly classified as non-projectable features. Often,
it might be the case that an internal variable can also be bound to an explicit
position in argument structure, viz. the instrument variable which
characterizes a whole range of verbs may or may not be bound to give rise to an
Instrument Subject Alternation:
(14) The gun hit at her face, … (H85 0836)[8]
At the other pole of the scale, one finds internal
variables which are never syntactically realized, viz. Reason and purpose which
are not bound to any external variable because they never have a syntactic
impact. These only form part of the meaning definition of the predicate, but
they are never lexicalized in the argument structure. This possibility of
lexicalization of an internal variable is not haphazard: internal variables are
instantiated lexically in external variables if there is a construction
triggered by the binding. This fact is explicitly expressed in the Lexical
Mapping rules as developed in Mairal (fc).
In sum, the idea behind this work is that the full
range of complements can be predicted from the semantic representation in its
lexical entry together with an independently motivated set of morphosyntactic
principles. Thus, the different interpretations of a predicate (as a
propositional (knowledge)/believe, perceptual, intentional predicate, etc.)
follow from the content of the internal variables in the semantic
representation. A further issue is to correlate these different interpretations
with the RRG Interclausal Relations Hierarchy, an issue which I have explored
elsewhere (Mairal, (fc)).
In what
follows, I shall like to propose a lexical template for the lexical classes
under analysis, each representing a particular facet which comes to illustrate
different features of a lexical template’s internal make-up. As advanced above,
I have selected five lexical classes, some closer in meaning and syntactic
encoding, while others distant in either meaning or syntax but closer in some
other respects.
The lexical
class of contact-by-impact verbs belongs to the general semantic domain
of action verbs. The generic term of contact-by-impact is hit, the central unit that defines the syntactic
and semantic universe of discourse; the rest of the members of this class are: strike, knock, tap, rap, cuff, slap, smack,
spank, whack, swat, bash, bump, thump, punch, sock, jab, club, clout, butt,
kick, dribble, hammer, crown, brain, box, beat, batter, clobber, whip, lash,
flog, flagellate, birch and cane.
After examining the whole set of both semantic and
syntactic regularities which converge within this lexical class, I should like
to propose the following lexical template:
(15) [[do´ (w, [use.tool.(a).in.(b).manner.for.(d)´ (w, x)]) CAUSE [do´ (x, [move.toward´ (x, y) & INGR be.in.contact.with´ (y, x)], a = x)]
The representation in (15) contains an effector (w)
who carries out the action of hitting upon an affected entity (y) using a tool (x).
More specifically, an effector uses a tool in a certain manner and with a
certain purpose causing an activity such that the instrument moves towards the
affected entity and x becomes in contact with y. It is not a
problem that the predicate underlying the conative with hit verbs is
move.toward´ while it is BECOME be-at´ with some other predicates (e.g. cut),
because hit at, etc. does not necessarily entail contact, while cut
at does.
Unlike manner-of-cutting verbs, contact-by-impact verbs do not conceptualise
an intermediate subevent that makes reference to an intermediate activity
(‘make-a-cut-on’ in the case of cutting
verbs). This difference stems from the fact that cutting actions are
conceptualized as having more duration than hitting actions, so that it is
possible to elaborate internally the cutting event; hitting, on the other hand,
involves a briefer contact: you may be cutting a piece of bread for a certain
time, but if you are hitting an object for a period of time, there is not just
one single blow, but a succession of hitting actions. That is, hitting is a
punctual event, unlike cutting.
As for the format, the internal make-up of a lexical
template consists of internal and external variables. The semantic content of
the lexical class of contact-by-impact permeates
a rich set of semantic parameters: (1) an agent that effects the blow (w); (2) a blow measured in terms of
force (lightly: tap, cuff); hard (strike, punch, thump,
knock, clobber, clout, batter, cane, whip, lash, birch); very hard (sock,
knock, clobber, bash, etc.); sound (sharp, punctual, loud (crack,
smack, whack, etc.); dull, punctual (thump, bump); and
movement (quick, (tap, rap, swat, etc.); swinging (swat), iterative
(rap, jab, batter, clobber (b); (3) an instrument with which the blow is delivered
(body part, object) (hand: hit, cuff, strike, etc.); fist: punch,
box, sock, etc.); head (butt), object (whip (whip, slash),
newspaper /fly swatter (swat) (a); (4) the entity
affected by the blow (person, body part, object, ball) (head: crown, brain;
buttocks: spank; ear: box; etc.) (y);
(5) the reason for the blow (anger (strike), punishment, causation of
movement, desire to hurt (spank, cane, whip, lash, flog, etc.; desire to
move something (strike, hit)) (d) (Faber/Mairal
1999: 185-186).
This rich set of internal variables again justifies my
claim that further decomposition is required if one’s theory aspires to account
for what could be termed lexical competence.
This class
of verbs marks the result of an action and are thus called result verbs. These
include the following: break, smash, shatter, splinter, snap, crack,
fracture, decompose, destroy, split, burst, explode, etc.
I believe
that the internal variables for English, especially the manner variable, should
be eliminated because English break (unlike its Lakhota counterparts)
has no specification of the nature of the causing activity. This in fact
suggests that the intermediate activity predicate should likewise be unspecified
as well. This would lead to the following lexical template:
(16) [[do´ (x, [use´ (x, y)] CAUSE [do´ (y, Ø)])] CAUSE [BECOME/INGR pred´ (z)]]
This
structure, in its maximal projection, designates a causative accomplishment such
that an effector initiates an action using an instrument such that the affected
entity comes to have a new state – that
of coming apart. Note that the only two internal variables relevant for this
class of verbs, instrument and result, are codified in terms of a use predicate
and a result subevent respectively, which is given by the resultative
construction itself. Thus, there is not need, in contrast to the other two
approaches, in postulating more internal variables. The important issue here is
that these verbs lexicalize a result. There are a number of comments to be made
about this representation:
1.
It treats the two causing actions, the one of the
instigator x and the instrument y as unspecified, which, as noted
above, seems to be correct for English.
2.
Having use´ as the first activity predicate is neutral
with respect to the nature of the causing action but it introduces an implement
argument, which is a potential instrument.
3.
There’s no ‘BECOME be-at´’ component, because I am not
really sure that that is not a fact about the world we attribute to the verb
rather than a true property of the verb. If it is neutral with respect to the
causing activity and specifies only that some unspecified action brought about
a specific result state, then attributing BECOME be-at´ to its semantic
structure is unjustified.
4.
There is no problem with the y argument being
the actor when the x argument is unspecified; it is an
implement-effector and the highest ranking remaining argument.
With regard to the resultative construction, one can
not posit a structure like:
(17) [LS2 [state’ (z)]]
since that is a property of the resultative
construction and not a lexical property
of break verbs. The result state comes from the semantic representation of
the construction, not from the lexical template of the main verb in the
construction.
Thus, in contrast to the contact-by-impact
verbs, break verbs only lexicalize the instrument and the result,
without giving any glues as to the manner, the means, or the nature of the
causing activity. This poor semantic description explains why this predicate
subcategorizes certain syntactic constructions which are very much in
consonance with this semantic description, viz. the causative / inchoative, the
middle construction, and its ill-formedness with an unspecified object
alternation, which presupposes an activity reading or interpretation, something
which is absent in this lexical template.
According
to Levin (1993:213), these verbs –also called verbs of ingesting – are related
to the process of ingesting food or drink. I will be examining the latter group
which contains the following class members: drink, imbibe, gulp, quaff,
swig, swill, guzzle, tipple and sip. In Van Valin/LaPolla
(1997: chapter 3), the following representation is proposed:
(18) do’ (x, [pred’ (x, y)]) & BECOME consumed’ (y)
I believe
that this structure only captures those facets of the meaning of the predicate
which are grammatically relevant, and consequently this does not do justice to
the internal semantic parameters which define this lexical class, viz. manner
and quantity. Furthermore, one could question whether the predicate drink,
which is the prototype of this lexical class, is in itself a primitive. If one
looks at the definition given for drink in the lexicon, this is stated
as follows:
(19) drink: to consume liquid, taking it into one’s mouth and swallowing it.
A possible
formalization of this definition could be stated as follows:
(20) do’ (x, [take’.(a).into one’s mouth.(b).in.(d).manner´] (x,y)) & BECOME consumed’ (y) a= y
Initially,
this representation accounts for the fact that this classes of verbs designate
an activity, and furthermore they can also designate an active accomplishment
structure depending on the referential nature of the NP. However, two issues arise; first, to analyze to which
extent take’ could not be further decomposed; second, to represent the internal
nominal structure which signals the goal of the movement described in the action
(into one’s mouth).
In relation
to the first query, as a first approximation one could argue that take
could be the primitive since this predicate also encodes movement, something
which is essential to account for certain syntactic constructions, viz. the
conative. However, take’ is too complex a predicate to be used, and thus
I shall like to formulate a further
decomposition along the lines of ‘cause
become be-in’ here. With regard to the second query, the representation of the nominal
clause, possession could be expressed by have.as.part’ (a,b). Depending
on which argument becomes the head of this structure, I shall account for both
the saxon genitive or a postmodification. In the case that concerns us here, mouth
(which stands for b) is interpreted as the head of the nominal clause, yielding
the structure, one’s mouth.
With regard to the
internal variables, it is no surprising that these predicates, unlike contact-by-impact
verbs, only codify a manner component given that the prototypical
interpretation of these verbs is that of an activity. Then, I will represent
the manner of the activity by means of a (b) variable.
Then, with all this
in mind I shall like to propose the following lexical template for this class
of verbs:
(21) do´ (x, [CAUSE.BECOME.be-in´.([have.as.part´.(x, mouth)], a).in.(b).manner´] (x, y)) & BECOME consumed’ (y) a = y
The reading
of this representation would go as follows; x, an effector, carries out
an activity such that causes y to become into x’mouth and y
becomes consumed. Unlike the other two previous interpretations, it is
interesting to observe that the internal variable is in turn modified and
further decomposed by a more elaborate logical event structure. This is a
significant step forward in relation to the hotly debated issue of the
dichotomy between internal and external variables.
Finally,
one could even discuss the internal nature of the CAUSE predicate, and
causatives in general. That is a very interesting and delicate issue to tackle.
In this regard, I think that they are an inevitable part of some activity
decomposition. However, this is an issue that I shall like to come back to in
relation to the codification of the next lexical template, verbs of existence,
where the causative predicate has almost become a classic in the linguistic
literature since the emergence of the debate between generative and
interpretative semantics at the beginning of the seventies.
Within this
lexical domain, I shall concentrate on the lexical class “to make a sound
indicating happiness”, which includes the following predicates: laugh,
chuckle, giggle, titter, snigger, snicker, cackle, guffaw, howl, roar.
One of the
recurrent semantic parameters of these verbs is that of manner. This is no surprising
given that this group of verbs designate a prototypical activity event
structure.
(22) do’ (x, [express. (a).in.(b).manner (x, y)] a = y
This structure would
be interpreted as follows; x carries out an activity such that x
expresses a sound y in a certain manner. The lexicalization of the
manner component will ultimately lead to the actual choice of one lexeme over
others in the lexical class.
It is interesting to
observe that these predicates subcategorize a cognate object, as illustrated in
the y variable, although their prototypical instantiation would be that
of an intransitive verb. However, as shall be seen later, things are not that
straightforward since these predicates can also occur with reaction objects or
even in a caused motion construction like the following examples illustrate:
(23) … he just laughed me straight out of the room (HMH 085).
(24) The audience were slower to laugh themselves into such a roaring myth (HTN 0682).
(25) Hillary threw back his head and laughed his great, frank, hearty laugh (H9D 2219).
(26) He laughed his big , round, comfortable but
oddly high-pitched laugh … (FB0 1159).
(27) Margaret laughed her goat laugh at Shildon’s stupidity (KRM 0096).
In principle, cognate
objects are independent of reactions, as shown by the fact that both can
cooccur together. In the case of a reaction object and a caused motion
construction, I claim that both arguments are given by the constructions
themselves. For example, in the case of the caused-motion construction I would
posit a structure like the one below:
(28) [laugh LS] CAUSE [BECOME NOT be-in’ (y,z)]
Within the
larger domain of EXISTENCE, I shall like to concentrate on the causative
cessative phase, to cause somebody to die. The hierarchical structure of
this lexical class includes a vast number of predicates which designate
different means and manners of encoding the cessation of existence: kill,
murder, assassinate, eliminate, do in, bump off, do away with, liquidate,
exterminate, massacre, butcher, slaughter, execute, behead, decapitate,
guillotine, hang, crucify, electrocute, starve, strangle, suffocate, smother,
gas, drown, slay.
If one looks at the
logical structures proposed for this type of verbs, I again observe that only
those aspects which are grammatically relevant are encoded, without making any
type of explicit mention to the range of semantic parameters encoded in this
representation. In this line, one could even argue that there is almost no
difference between verbs of cooking and verbs of killing since
both make use of the same event structure, an assumption which leads us to
detect that something is wrong with these two representations:
(29) [do’ (x,f)] CAUSE [BECOME dead’ (y)]
(30) [do’ (x,f)] CAUSE [BECOME baked’ (y)]
In order to circumvent this problem, I resort to the
semantic parameters encoded in this lexical class and postulate the following:
instrument, manner, purpose or reason. The next step, as has been the usual
practice thus far, consists of expanding the logical structure above and
convert it into a lexical template. The resulting representation has the
following format:
(31) [[do’ (w, [use.(a). in.(b).manner.for.(d)´ (w, x)] CAUSE [do’ (x, Ø)])] CAUSE [BECOME dead’ (y)] a= x.
However, a closer look at some of these predicates
reveals that the parameter ‘agency’ is a salient and distinguishing property,
thus the oddity of the following instances:
(32) * Malaria murdered Fred (VV).
(33) * The explosion murdered Larry’s neighbor (Van Valin/Wilkins, 1996: 310).
(34) * The dagger murdered Julius Cesar (Levin, 1993: 231).
In order to account for this particular feature, I
shall like to reformulate the lexical template above and add a DO operator,
which signals agency as an inherent feature of the verb (cf. Van Valin and
LaPolla, 1997:119). Furthermore, this operator explains why these verbs cannot
participate in the Instrument Subject construction. Then, the revised lexical
template is the following:
(35) [DO’ [[do’ (w, [use.(a). in.(b).manner.for.(d)´ (w, x)]) CAUSE [do’ (x, Ø)]] CAUSE [BECOME dead’ (y)] a = x.
Then, this representation would go as follows; and
effector (w), using an instrument, intentionally carries out an action in
a certain manner and with a certain purpose such that this causes that the
instrument instigates another action causing that a patient (y) becomes
dead. Note that I have identified the a variable
with the external instrument variable. The rest of the internal variables are
not bound to any specific construction.
As noted above in the discussion of consumption verbs,
an intricate issue that arises concerns the semantic decomposition of the
causative operator CAUSE. In this regard, compare the following two structures:
(36) John killed Bill.
(37) John caused Bill to die.
The first represent a lexical causative while the
second is a clear example of a core juncture, and consequently the causality is
much less direct than in a nuclear juncture or in a lexical causative. Someone
may object that the representation for both should be the same. Recall that
this is one of the arguments against the use of semantic decomposition.
However, following Van Valin/LaPolla (1997: 671, fn. 17), the logical structure
for a predicate like cause would have the following format:
(38) [do’ (x, f) ] CAUSE [undergo’ (y,z)]
where the variable z stands for the complement
structure of the verb:
(39) [do’ (John, f) ] CAUSE [undergo’ (Bill, [BECOME dead’ (Bill)])]
This structure clearly differs from that proposed for
a lexical causative such as kill. Despite this tentative analysis, it is
unquestionable that one of the most intriguing and fascinating lines of
research is to explore the universality and primitive nature of some of the
predicates. In this regard, either an ontology or a set of culture-bound
universals can serve as an alternative to detect where the chain of
decomposition actually ends. However, this is an issue which is beyond the
scope of this paper.
Finally, it is necessary to analyze whether there is
any type of constraint on the directionality of the lexical template. In
consonance with this, there are some authors who proclaim that lexical
templates should be minimal representations, while others claim just the opposite.
My analysis has shown that both processes are possible and this justifies the
fact that the general linking principle – the Lexical Template Modelling
Process (LTMP) - is in itself a modelling process, without invoking neither a
reductionist nor a maximalist orientation.
In the most recent literature the notion of template
has been adopted by some linguistic models. In this regard, there are two
questions that merit a brief comment:
·
The format of a template should be a minimal
expression or else a maximal representation.
·
The directionality of the template, that is, whether
the template should be constrained by an expansion mechanism, and accordingly,
no reductions are possible, or else, whether a template should be maximal in
its form and both reduction and expansion processes are possible.
Our approach formulates a maximal representation since
this intends to capture the whole conceptual substance of a lexical class,
though without invoking that reduction processes hold to account for those
syntactic configurations which do not fit the model in full. This entails a
very different conception of what basic templates are in comparison with other
proposals: Rappaport and Levin’s (1998) account of these phenomena is based on
an incremental process on templates, their Template
Augmentation Process, which is in turn inspired in the monotonic nature of
verb meaning:
Template
Augmentation: Event structure templates may be freely augmented up to other
possible templates in the basic inventory of event structure templates.
Thus, if we consider a prototypical activity verb like
sweep its basic template would be:
(40) [x ACT <SWEEP> y ]
However,
the use of this predicate in a resultative construction entails the expansion
of its basic activity template resulting in the following:
(41) [x ACT <SWEEP> y ] CAUSE [BECOME [y <STATE>]]
This procedure seems to work properly for these verbs,
and makes in principle hard to decide whether a reduction or an expansion
process is the right choice. I have found compelling evidence, both from the
primary and the affixal lexicon, that shows that it is desirable to formulate
maximal representations, and thus unrestricted in nature, to minimal
representations, which serve as input for expansion processes. This kind of
solution runs the risk of being unrestrictive and unconstrained: it leaves open
the possibility of yielding impossible or non-lexicalized structures, like the
following:
(42) fight: *[[x ACT <FIGHT> y ] CAUSE [BECOME [y <STATE>]]]
Fight, as any activity verb would be an optimal
candidate for augmentation to yield a causative accomplishment reading where
the activity becomes the causing subevent for an accomplishment; but there is no
such meaning available for this verb. Rappaport/Levin (1998:112) apparently
filter out this possibility by imposing the so-called Subevent Identification Condition, according to which each subevent
in the event structure must be identified by a lexical head in the syntax
(either a verb, and adjective or a preposition). Thus, nothing prevents from
interpreting the sentence “The villagers
fought against the invaders” as a causative accomplishment since the
prepositional complement would be associated to the second subevent in the
causal chain; but this is obviously not the case. In much the same way, if we
consider the template proposed for externally caused change of state verbs (break, destroy, smash, decompose, etc.)
in Rappaport/Levin (1998: 116) we observe that such template is just the
maximal projection after the Template Augmentation Principle has been applied
resulting in the following structure:
(43) [[ X ACT <MANNER> ] CAUSE [BECOME [ y <BROKEN>]]]
However, since this structure cannot be further
augmented because it is already a maximal projection, this makes it difficult
to account for the following instances:
(44) The first question can be broken down into two further questions (LOB G8.H83)
(45) At a particular depth in the sea, (it) resolves itself into two components (LOB J1H55)
(46) The last three compounds decomposed to oxides on heat treatment (LOB J7.H94)
(47) The shadow resolved itself into a large brown ...... (LOB 11.H58)
(48) ....was not fit to go under water and then he had burst into tears (LOB N1H.39)
According to Rappaport/Levin (1998: 122-123), the
resultative phrase is a further specification of the change lexicalized in the
predicate. Though certainly true, Levin and Rappaport’s structure cannot
account for this phrase without having to stipulate a new expansion since the
added subevent, according to the Subevent Identification Condition, can be
satisfied given that this is headed by a preposition. In contrast, as noted
above, our proposal already codifies this resultative phrase by means of the
variable (z) in the expanded lexical template. This, again, is one more
argument in favour of proposing maximal representations for the lexical
properties of lexical classes.
Lexical templates, on the other hand, in being
maximalist avoid the danger of overgenerating non-realized alternations; this
involves a more economical kind of explanation. At the same time, maximalist
templates are further motivated if one considers certain word-formation
phenomena, such as what we might call “inherited polysemy”: it is not unusual
to find derived words with more than one meaning, each of which corresponds
with different meanings of the base word. Suffice the following example: descendant can refer to either an entity
involved in a motion activity (“someone who descends”) or may designate one
participant in a stative locative-temporal connection (“someone posterior in a
temporal line”). It is clear that these two possible meanings are bound
directly to the polysemous nature of the verbal base: to descend can refer to the action of going down along a vertical
path, in which case its event structure is that of an activity:
(49) As we descend down the valley the views became more and more fascinating
or it may express just the localization in a temporal
line:
(50) We all descend of a forest-living ape
It is evident that the stative reading is part of a
complex event structure such that it is possible to conflate under one single
meta-entry the two possible interpretations without having to stipulate two
different lexical entries, one for descend
as an activity:
(51) do’ (x, [descend’ (x, y)])
and a second augmented one as an active accomplishment
where the state subevent is encapsulated:
(52) do’ (x, [descend’ (x, y)]) & BECOME be-loc’ (z, x)
A further complication of this approach is its
inability to explain why the temporal meaning of the verb focalizes only on the
stative resultative subpart of the augmented template (be-loc’ (z, x));
Rappaport/Levin (1998) make no allowance for any reduction process, which is
necessary if descend (“be after in
succession”) is taken as a base word for the suffix –ant; that is, one faces the following disadvantages:
(a)
if we want to correctly predict the two possible
meanings of the derived word, and one makes no use of maximal representations,
the base word would have to be represented at least twice.
(b)
even if this approach were adopted, the stative
interpretation of the base word would have to be arrived at by an event
pruning process from its active accomplishment interpretation, a mechanism
disfavored by Rappaport/Levin (1998).
The maximizing approach is more economical in having
to establish only one representation for the verb, something that is in accord
with my aim of designing a minimalist lexicon. The affix will have then access
to this meta-entry and will select all or part of it for the elaboration of the
representation underlying the derived word. That is, affixes are sensitive to
the semantic parameter of degree of event
elaboration (D.E.E.) (cf. Kemmer 1993, and Wee 1999): affixes can opt to
choose the event represented in the meta-entry as an undifferentiated whole:
(53) φin: do’ (xi, [descend’ (xi, y)]) & BECOME be-loc’ (z, xi) [D.E.E. = Æ]
or it may select one of its substructures or component
parts:
(54) φin: be-loc’ (z, xi) [D.E.E. = ³ 1]
Of course,
I claim that the D.E.E. corresponds to alternative conceptualizations by the
speaker. This parameter also refers to the extent to which participants in an
event are distinguished. In the case of agentive formations the D.E.E. explains
also the “potential polysemy” of units like cleaner,
that can either refer to the Agent or to the Implement of the event, depending
on whether the first participant is specified or not.
Role and
Reference Grammar posits both directions, either reduction or expansions. For
example, the lexical rules in section 4.6 go ‘both ways’ in the sense that they
can be reduction or expansion rules. This is because languages do both; there
are languages exemplified in Van Valin/LaPolla (1997) by Tepehua, Mparntwe
Arrente, in which inchoatives are formed by the addition of an inchoative
morpheme, and causatives are formed by the addition of a causative morpheme.
These would clearly be expansions. However, in Romance and Slavic languages,
the base form of a verb like break is a causative accomplishment, and
the added morphology (reflexive morphemes) reduces the lexical template from
causative accomplishment to an accomplishment (Van Valin/LaPolla, 1997: 98-99;
178-183). The tricky case is languages like English, in which there is little
morphology signaling the alternations (Rappaport/Levin, 1998: 118).
Initially,
when I began my research, it seemed to me more reasonable to propose a maximal
form underlying the class and then to argue that the different verbs left
different aspects of the maximal lexical template out. The advantage of this is
that it constraints the system; the maximal lexical template would correspond
either to a causal chain lexical template (or an extended motion event with
activity + source + path + goal specified for motion verbs like descend).
As things stand, I firmly believe that both expansion and reduction must be
allowed but tightly constrained.
In order to
make things clear, the format of a lexical template is a maximal expression but
a different issue is the way this interacts with specific syntactic
configurations. In this regard, I claim that the accommodation of a lexical
template to a specific construction can take a maximized or reduced format. That
is, I will be talking about modelling processes. In much the same way as
predicates share a semantic area as encoded in their respective lexical
subdomains, predicates select /licence/ focalize one part, or parts of the
lexical template. This avoids the issue of whether lexical representations
should be minimal or maximal representations, a theoretical issue which is
highly controversial.
This paper
has been mainly concerned with the notion of lexical template. Firstly, the major
current approaches to lexical representations have been addressed with a view
to showing that these primarily proclaim that the identification of those
elements which form part of a lexical representation should be equated with
those aspects which are grammatically salient. In connection with this, I
advance the claim that a full semantic decomposition should be posited in order
to explain the full meaning potential of a predicate. With this assumption in
mind, I explain why the notion of lexical template grows out. In the second
part of this paper, an explicit account of the nature of the lexical
representation as well as the format of a lexical template is given. In order
to prove the explanatory value of the notion of lexical template, I bring to the
forefront the lexical templates proposed for a wide variety of lexical classes.
Bach, E./Harms, R.T. (eds.) (1968): Universals in Linguistic Theory. New York: Holt, Rinehart and Winston.
Butler, Ch. (2002): “Catching a glimpse of linguistic reality:
modelling the complexities of verb meanings in the Functional Lexematic Model”.
In Mairal/Pérez Quintero (eds.),
241-272.
Butt, M./Geuder, W. (eds) (1998): The projection of arguments: Lexical and compositional factors. Stanford: CSLI.
Chomsky, N. (1981): Lectures on Government and Binding. Dordrecht: Foris.
Croft, W. (1991): Syntactic Categories and Grammatical Relations. Chicago: The University of Chicago Press.
Dik, S.C.
(1978): Stepwise Lexical Decomposition. Lisse: The Peter de Ridder
Press.
Dik, S.C. (1997): The theory
of Functional Grammar: The structure of the clause. Berlin: Mouton de Gruyter.
Dowty, D.R. (1991): “Thematic Proto-Roles and Argument Selection”. Language 67, 547-619.
Durán, P./Aguado, G. (eds.) (2001): La
investigación en lenguas aplicadas: un enfoque multidisciplinar. Madrid:
Universidad Politécnica de Madrid,
Emonds, J.E. (2000): Lexicon and Grammar: The English Syntacticon. Berlin /New York: Mouton de Gruyter.
Faber, P./Mairal, R. (1999): Constructing a lexicon of English verbs. Berlin /New York: Mouton de Gruyter.
Fillmore, Ch. (1968): “The Case for Case”. In E. Bach y R.T. Harms (eds.), 1-88.
Fillmore, Ch. (1988): “The Mechanisms of Construction Grammar”. BLS 14, 35-55.
Goldberg, A. (1995): A Construction Grammar approach to argument structure. Chicago: University of Chicago Press.
Goldberg, A. (1998): “Semantic Principles of Predication”. En J.P. Koenig (ed.) Discourse and Cognition. Bridging the Gap. Stanford: CSLI Publications.
Grimshaw, J.
(1993): “Semantic Structure and Semantic Content in Lexical Representation”.
Unpublished manuscript. Rutgers University.
Jackendoff, R. (1983): Semantics and Cognition. Cambridge, Mass.: The MIT Press.
Jackendoff, R. (1990): Semantic Structures. Cambridge, Mass.: The MIT Press.
Jackendoff, R. (1996): “Conceptual semantics and cognitive linguistics”. Cognitive Linguistics, 7-1, 93-129.
Jackendoff, R. (1997): The Architecture of the Language Faculty. Cambridge, Mass.: The MIT Press.
Kemmer, S. (1993): The Middle Voice. Amsterdam: John Benjamins.
Jackendoff, R. (2002): Foundations of Language. Cambridge: Cambridge University Press.
Lappin, S. (ed.) (1996): The Handbook of Contemporary Semantic Theory. London: Blackwell.
Levin, B. (1993): English Verb Classes and Alternations. A Preliminary Investigation. Chicago: University of Chicago Press.
Levin, B./Rappaport Hovav, M. (1995): Unaccusatives: At the syntax-lexical semantics interface. Cambridge, MA: MIT Press.
Levin, B./Rappaport Hovav, M. (1996a): “Lexical semantics and syntactic structure”. In S. Lappin. (ed.), 487-507.
Levin, B./Rappaport Hovav, M. (1996b): “From Lexical Semantics to
Argument Realization”. Unpublished
manuscript.
Mairal, R. (2001): “En torno a la interficie
léxico-gramática en los modelos gramaticales”. In Durán, P./ Aguado, G. (eds.),
115-151.
Mairal, R. (fc): Lexical templates and syntactic structures: The design of a Lexical Grammar Model. Unpublished manuscript.
Mairal, R./Faber, P. (2002): “Lexical Templates and Functional Grammar”. In R. Mairal/Pérez, M.J. (eds), 39-94.
Mairal Usón, R./Pérez Quintero, M. J. (eds.) (2002): New perspectives on predicate argument structure in Functional Grammar. Berlin and New York: Mouton de Gruyter.
Mohanan, T./Wee, L. (eds.) (1999): Grammatical Semantics. Evidence for Structure in Meaning. Standford: CSLI Publications.
Nirenburg, S./Levin L. (1992): “Syntax-Driven and Ontology-Driven Lexical Semantics”. In J. Pustejovsky/ Bergler, S. (eds.), 5-20.
Nirenburg, S./Raskin, V. (2002): Ontological Semantics. New Mexico: CRL.
Pesetsky, D. (1995): Zero Syntax. Cambridge, Mass.: The MIT Press.
Pustejovsky, J./Bergler, S. (eds.) (1992): Lexical Semantics and
Knowledge Representations. Heidelberg: Springler Verlag.
Rappaport Hovav, M./Levin, B. (1998): “Building verb meanings”. In M. Butt/Geuder, W. (eds), 97-134.
Shibatani, M./Thompson, S.A. (eds.) (1996): Grammatical Constructions. Their Form and Meaning. Oxford: Clarendon Press.
Tenny, C.
(1994): Aspectual Roles and the Syntax-Semantics Interface. Dordrecht: Kluwer.
Van Valin, R.D. (ed.) (1993): Advances in Role and Reference Grammar. Amsterdam: John Benjamins.
Van Valin, R.D./Wilkins, D. (1993): “Predicting syntactic structure
from semantic representations: Remember
in English and its equivalents in Mparntwe Arrernte”. In Van Valin (ed.). (1993),
499-534.
Van Valin, R. D./Wilkins, D. (1996): “The Case for ‘Effector’: Case Roles, Agents, and Agency Revisited”. In M. Shibatani/Thompson, S.A. (eds.), 289-322.
Van Valin, R. D./LaPolla, R. (1997): Syntax: structure, meaning and function. Cambridge: Cambridge University Press.
Vendler, Z.
(1967): Linguistics in Philosophy. New York: Cornell University Press.
Wee, L. (1999): “Explaining Grammatical Polysemy”. In T. Mohanan/Wee, L. (eds.).
Wierzbicka, A. (1992): Semantics, Primes and Universals. New York & Oxford: Oxford University Press.
Wierzbicka, A. (1996): Semantics: Primes and Universals. Oxford: Oxford University Press.
Wierzbicka, A. (1997): Understanding Cultures Through their Key Words: English, Russian, Polish, German and Japanese. New York & Oxford: Oxford University Press.
Wierzbicka, A. (1999): Emotions across Cultures and Languages: Diversity and Universals. Cambridge: Cambridge University Press.
[1] I take the term used in Jackendoff (1997) to describe a particular group of linguistic theories and apply it to lexical representations. Also some other labels are used to describe this trend of research, viz. “syntactically-driven theories” (cf. Nirenburg and Levin, 1992).
[2] For a survey of the major critiques, see Jackendoff (1990), Croft (1991), Dowty (1991), Levin and Rappaport (1996b), Van Valin and Wilkins (1996), among others.
[3] In fact, in a more recent study Rappaport and Levin (1998) argue that result constants constrain the subcategorization frame of a predicate more than manner constants.
[4] Nonetheless, there are some
tentative proposal to provide richer semantic decompositions, e.g. Van Valin
and Wilkins’ (1993:517) analysis of the predicate remember and Van Valin
and LaPolla’s (1997:117) analysis of speech act verbs.
[5] This twofold distinction is very similar in many respects to the lexical conceptual structures and predicate argument structures as proposed in some other approaches like Jackendoff (1990) or Levin and Rappaport (1995, 1996a). This comes to reinforce my claim that my proposal is not exclusive of one particular model but can be extended and accommodated into other theoretical frameworks.
[6] In a similar vein, Fillmore (1988:36), within the realms of Construction Grammar, affirms that a construction is constituted of external and internal properties. The external syntax of a construction encapsulates the properties of the construction as a whole, that is, “anything speakers know about the construction that is relevant to the larger syntactic contexts in which it is welcome”. In contrast, the internal syntax of a construction refers to the construction’s make-up.
[7] I have adopted this notational device from Role and Reference Grammar (cf. Van Valin and LaPolla, 1997 chapters 3 and 4).
[8] I use corpus data from the BNC and the LOB so that a more complete version of the full gamut of syntactic alternations is obtained. Each example is accompanied by a code giving a reference to the exact place in the corpus where the example may be found.