To illustrate the potential benefits of our approach, we have provided below a detailed analysis of grasp, characterized in terms of the meaning components defined above. We describe the unique properties associated with grasp, and identify commonalities with other verbs which point to productive linguistic generalizations.
The transitive verb grasp
SUMMARY OF FEATURES
We choose to consider the basic meaning of this verb to be an event of exerting force on an object, where the object is the denotation of the direct grammatical object of the verb. If the subject of the verb denotes an agent, such as a human being, the default assumption is that this agent exerts force on the object by closing his/her hand or hands around it. This definition corresponds more or less directly with the animation primitive GRASP.
The basic action denoted by grasp under this definition is essentially punctual; the action of closing one's hand around an object is momentary and does not extend over an interval of time in the way that, for example, a carry event may. (The distinction between punctual events and other kinds of events, such as a carry events, is finally an issue of granularity, since no event is truly instantaneous. However, at the level of granularity required for this type of instruction understanding, the distinction is meaningful.)
However, it is clear that many types of events that may be described by the verb grasp involve more than just this simple hand-closing. Still, no event of grasping (except when grasp occurs in the conative alternation, as discussed below) can fail to include this action. Therefore, we identify this as the core meaning of the verb when applied to an agent subject.
The meaning of an instruction such as (1), though, is not exhausted by this basic definition for grasp:
(1) Grasp handles.
Such an instruction may be intended to provoke a behavior on the part of the agent that includes reaching for the handles in question, and then also holding them in a certain way and for a certain length of time.
The extended meaning of grasp in cases like this may be productively viewed within the framework of an aspectual system such as that proposed by Moens and Steedman (1988). They note that an event in general consists of three parts: a preparatory process, a culmination, and a consequent state. A reference to an event, such as the reference in (1), may include some or all of these three parts, depending on contextual and other factors.
The preparatory process may involve bringing about the preconditions for the execution of the core action ( culmination). In the case of (1), this means that the agent will reach for the handles, and thereby gets his/her hands in a position such that closing them will cause force to be exerted on the handles.
The consequent state refers to the desired post-conditions of the action, in this case, that the agent is holding the handles. Notice that the preparatory process is inherently bounded; it is over when the culmination of the event occurs. However, the consequent state is not bounded in this way. It is useful to speak of bounded processes as telic and unbounded processes as atelic. For example, (2) expresses a telic process:
(2) and carefully slide unit out from equipment bay.
since there will come a natural point at which this action has been completed, namely, when the unit is free from the equipment bay (as judged by either the agent or the instruction-giver). Verbs such as slide are not inherently telic because they could continue indefinitely. That is why the end point of the sliding event, being out of the equipment bay, must be specified.
In the absence of such an explicitly specified termination condition for a process, the agent carrying out the instruction in (1) will have to determine when the consequent state of the grasp is over. The default solution may be simply to wait until explicitly told to let go of the handles. Such an explicit instruction may take the form of (3).
(3) Stop grasping the handles.
In this case, the use of the aspectual verb stop, which applies only to processes, signals that grasping is referring to the consequent state of the grasp event. Another possible instruction to signal the end of the consequent state would be (4):
(4) Grasp the handles until I give the signal.
The until clause again refers to the consequent state of grasp, not the actual hand-closing action that is the core meaning.
However, it is more likely that an intelligent agent will interpret an instruction like (1) not as an end in itself, but as a step in a plan to do something with the handles. This reasoning about the larger context of a given instruction requires knowledge about the typical kinds of things that are done with handles, and involves expectations about subsequent instructions that may be given. An agent who reasons in this way will know something about how long to maintain the hold on the handles, particularly if holding the handles is a precondition to performing another subsequent instruction such as (2). Moreover, such knowledge is independently necessary to constrain the many variations of grasp that an agent can perform, depending on the specific object to be grasped and the larger goal to which the grasping may contribute (Levison 1994).
The verb grasp can also occur in a syntactic frame containing a with adjunct, as is seen in (5).
(5) Grasp the handles with a monkey wrench.
Though the with phrase is syntactically optional, it does not pattern like other adjuncts, whose semantic contribution to the core phrase is an elaboration or specification. It more closely resembles the direct object of eat, which, though omissible in many contexts, is filling an essential role in the verb's predicate-argument structure (Palmer 1988). To see this, consider
(6) Grasp the handles.
This variant of (5), in its default reading, implies that the agent is intended to exert force on the handles directly by closing his/her hands around them. However, it is clear that in (5) the agent is not instructed to touch the handles with his/her hands (particularly not if the handles are very hot). This means either that the with adjunct has negated one part of the meaning of the core phrase it has attached to, which is not typical of an optional modifier; or else that the with adjunct is specifying the filler of an essential role in the meaning of grasp, which in the default case in (6) was taken to be filled by the agent's hands. This latter possibility resembles the situation with the optional object of eat; the default interpretation, when this object is missing, is that the role of the thing eaten is filled with some unspecified kind of food, but this may be overridden by an explicit object.
Under this analysis, both (1) and (6) have a syntactically unrealized argument position, corresponding to the semantic role of intermediary (Palmer 1993). The intermediary may also be viewed as the object of an implicit ``manipulation subscene'' (Gawron 1988); the agent manipulates the intermediary (the hands, or the monkey wrench) and this results in the object of the verb being affected in the desired way.
It may seem counterintuitive to view an agent as manipulating his/her own hands, but in the animation environment, this has some validity, since instructions must be sent to the system to simulate hand motions as well as other kinds of actions. Also, non-human agents may have detachable end-effectors, so that the distinction between a hand and an instrument becomes hazier.
The alternative would be to view grasp and grasp with as two different verbs with distinct but related meanings, much as the transitive and intransitive forms of eat are sometimes considered to be distinct entries in the lexicon, in some theories of grammar. However, note that one may say
(7) Grasp the handles with your hands.
and this is basically synonymous with (1). (Synonymy in the case of instructions may be loosely defined in terms of the behaviors they provoke in otherwise equal circumstances.) This suggests that there is an implicit argument in forms like (1) and (6) that can be made explicit if desired, for clarity, without altering the meaning. The synonymy of (1) and (7) is problematic for an account where (1) is interpreted as not involving an intermediary at all. Possibly then (7) would receive an interpretation akin to that of
(8) The car hit the wall with its bumper.
but it is not clear that this differs from the intermediary interpretation; (8) could be interpreted as an example of non-agentive manipulation of an intermediary. The fact that examples such as (8) are only acceptable when the with phrase has as its object an inalienably possessed part of the subject, could be due to the fact that non-agentive manipulators, being unable to actively gain control over objects that they manipulate, must already be in a relationship of control with respect to those objects, for example, by virtue of being inalienably attached to them.
It is clear however that grasp is an agentive verb, requiring an intentional agent as its subject. In fact, it is more agentive than a verb like hit, since someone can be said to accidentally hit something, but it's odd to say that someone accidentally grasped something. This seems to derive from the semantics of the verb; while hitting typically requires only coarse-grained body manipulations, grasping involves carefully coordinated finger motions, and thus is seen to take more conscious deliberation. On the other hand, a falling person may instinctively grasp a tree branch without much intentionality.
The subject of grasp however is not an agent in the instrument subject alternation (Levin 1993), as in (9).
(9) The monkey wrench grasped the handles.
In an appropriate context, (9) could describe a situation where someone is manipulating monkey wrenches in such a way as to cause them to exert force on the handles. Not all instruments occurring in with phrases can be promoted to subject position in this way, however; (11) is not a valid paraphrase of (10).
(10) I grasped the handles with the thick towel.
(11) * The thick towel grasped the handles.
However, when an instrument does occur in subject position, this may be interpreted as a situation in which the true subject, the agent, has been omitted syntactically, and thus the instrument has been promoted to subject as the highest-ranked thematic entity remaining in the sentence (Fillmore 1968). We hypothesize that such omissions of the subject can only occur in very constrained circumstances, namely, where the omitted argument is immediately recoverable from discourse context.