The verbs that change the status of only one object include as a subset the class of motion verbs, where the changed state of the object is its position or orientation. All motion verbs require a spatio-temporal component describing the position of the object over time, and may include additional components (e.g., culminating conditions and manner of motion) that are specific to the particular verb or subclasses of verbs. For example, the motion verbs used in the F16 technical orders corpus can be partitioned into two main subclasses:
Although in general many of these verbs can refer to either an agent's volitional change of location ( I walked home, I turned around (primitive LOCOMOTE) or an agent changing the location of an object, I turned the chair around (primitive MOVE), in this domain the usages almost always correspond to the latter.
The syntactic and semantic behavior of these two subclasses of motion verbs are
documented in (Levin 1993), see Fig 
, and this allows us to
make generalizations about each subclass -- the arguments required as well as
how these arguments may or may not be realized syntactically. The semantic
components of a verb or verb class aid in searching for the proper lexical item
to express an action, and the syntactic frames associated with the verb or verb
class provide guidelines as to how to describe the action linguistically.
Thus, knowing which Levin class a verb belongs to helps in both lexical choice
and sentence planning.
Figure: Portion of verb lattice involving Levin's motion verb classes
As a verb of inherently directed motion, enter has an underlying representation with two arguments: an object (object1) that is the actor undergoing motion, and an object (object2) that is the location that is entered. (When more than one object is involved in the motion event, the first object is understood to be the one undergoing motion, and other objects provide constraints on the motion of the first object.) The second argument may be expressed as an object of the preposition into or as a direct object (in the so-called preposition drop alternation), or it may not be expressed at all, if the context makes the location self-evident so that an explicit syntactic realization of the location would be redundant. The action is known to be completed once the proper geometric relation between the object and location is achieved (as in the case of enter) or terminated (as in the case of escape). Most verbs of inherently directed motion disallow an additional agent who causes the object to move ("sleeve will enter aerial refuel receptacle bore" is permissible, but not "enter sleeve into aerial refuel receptacle bore"). Consequently, these verbs are used in descriptions rather than instructions.
enter (actor, location)
- -----------------------
enter (L-actor, L-object)
enter (L-actor, into(L-object))
enter (L-actor)
(agent = <none>;
object1 = L-actor,
object2 = L-object
OR
object2 = <inferred>;
spatiotemporal =
(goal = (execution-type = establish;
relation = in(object1, object2))))
In contrast, verbs specifying manner of motion usually exhibit the causative - inchoative alternation in which the object undergoing the motion can be realized syntactically as either a direct object with an Agent causing the object to move ("slide bracket on shank") or as the subject of a sentence with no Agent specified ("bracket slides on shank"). In the absence of a prepositional phrase specifying direction, these verbs do not indicate any direction or goal of motion, although a particular verb may have additional arguments inherent in its meaning.
slide (agent, object1, object2) / slide (object1, object2)
- ----------------------------------------------------------
slide (L-object1)
slide (L-object1, on/over(L-object2))
slide (L-agent, L-object1)
slide (L-agent, L-object1, on/over(L-object2))
(agent = L-agent
OR
agent = <none>;
object1 = L-object1;
object2 = L-object2
OR
object2 = <inferred>;
spatiotemporal =
(goal = (execution-type = maintain;
relation = contact(object1, object2)));
kinematics = translational)
For example, the representation for slide includes an optional agent and an object1 that undergoes the sliding motion, plus a second object2 that provides the surface over which the sliding occurs. Although the direction of motion is not specified, any slide event must contain a translational motion component of object1 with respect to object2 and maintain contact between object1 and object2. The surface of sliding may be realized syntactically using an on-PP ("Slide sleeve on tube") or over-PP ("Slide coupling nut over packing and sealing ring"), or it may have to be inferred contextually ("Carefully slide matrix assembly forward to disengage two bushings"). Since "slide" is not an inherently telic verb (the sliding action may continue indefinitely), its end point should be specified explicitly ("sliding sleeve into housing"). The information from the goal clause ("into the housing") may also be used to supply the inferred sliding surface.
Another distinction that needs to made with respect to slide has to do with the amount of effort involved in moving the object, or the WEIGHT parameter. If the effort exceeds a certain threshold, then the action becomes a pushing action rather than a sliding action, and the choice of lexical item should change accordingly.
Some verbs used in task descriptions (reinstall, align, couple, coordinate, insert, replace, lubricate, refuel, discard, defuel, safety-wire, adjust, release, reposition) aren't in the Levin classes. One of our research issues will be determining appropriate classifications.