REFLEXIONES CRÍTICAS A LOS TRABAJOS SELECCIONADOS

As seen in the example in section 5.3.1, the semantic structures in VerbNet consist of various semantic predicates which are used to show the relations between events and the participants in those events [28, p. 37-41]. These predicates are joined together using the conjunct symbol & and may also be negated, by embedding the predicate inside another predicate called NOT. The arguments appearing inside the semantic predicates may include the thematic roles specific to each class, an assortment of various universal constant arguments, such as Forceful and Directedmotion, and finally a few arguments that are verb specific. The latter argument type depends on the given verb, so an argument like Form will, for instance, be Broken for the verb break and Bended for the verb bend. In other words, the Theme ends up having a broken form in a breaking event, and a bended form in a bending event. Predicates also typically take an argument which denotes the event in question, or one of its stages (see below). Events are, as we have seen, marked with the E symbol.

Every semantic predicate usually takes as one of its arguments a time function, which specifies when the predicate is true. These time functions are based on an approach to time established by Moens and Steedman [20], which is itself based on Vendler’s method of categorizing events [29]. By their accounts, events are divided into four categories, namely culminations, points, processes and culminated processes. These are defined depending on both their length and their outcome. In other words, the events are either atomic or extended, i. e. they are either instantaneous or last for some time, and either consequently result in a new state, or not. Following this principle, culminations are atomic events that lead to a new state, points are also atomic, but do not lead to a new state, and both processes and culminated processes are extended events, but whereas the former does not lead to a new state, the latter does. Separate from events are also the state category, which VerbNet does not refer to directly. In other words, there is no use of variables which directly denote states (like an S symbol), but they are indirectly represented using various semantic predicates which denote

states.

Events thereby have a general structure, called a nucleus. This consists of different stages, namely the starting point, the preparatory process, which may last for a given time, the culmination, which is an instantaneous point in time, and the consequent state. The time functions used by VerbNet returns the selected stages from a given event. They include the functions START, DURING, END and RESULT, which all take a single argument (the event variable). The illustration in (5.1) below shows the structure of an event in VerbNet, where the rectangle represents the whole event. Note that Moens and Steedman did not refer to a starting point in their original illustration, but it’s included here since it seems logical in the VerbNet framework, where the START function may refer to the start of an event.

Figure 5.1: The event time structure used in VerbNet

In addition, one can also say that an event is true at all times in the event, by not using any time function. Instead, the event variable is used directly as the time argument in the given semantic predicate. For an example of the use of these time functions, the structure for the sentence Brutus killed Caesar with a knife can be seen in formula (68), taken from the class murder-42.1-1. Note the use of subscripts on some of the predicate arguments, which indicate what specific role they occupy.

(68) CAUSE(BrutusAGENT, E) & ALIVE(START(E), CaesarPAT IENT) &

NOT(ALIVE(RESULT(E), CaesarPAT IENT)) &

USE(DURING(E), BrutusAGENT, KnifeI NSTRU MENT)

This structure uses the predicates CAUSE, ALIVE and USE, plus the NOT predicate to indicate a negation. CAUSE represents the fact that the sentence is causative, where an Agent is said to cause the given event. This predicate is very similar to the CAUSE function used by Jackendoff (chapter 3) and the CAUSE predicate used by Parsons (chapter 4). Note the use of just the E variable in the CAUSE predicate here. This indicates that the fact that Brutus is the causer is true at all times in the event. ALIVE indicates that some participant is alive at a given stage in the event, while USE says that some participant uses some Instrument at some stage in the event. The START time function used in the ALIVE predicate means that Caesar was alive at the start of the event, the DURING function inside the USE predicate means that Brutus used a knife in the preparatory stage of the event, while the RESULT function used in the ALIVE predicate inside the

NOT predicate indicates that Caesar was not alive in the consequent state of the event.

Some event types do not have access to all the stages. Processes do not have clear start or culmination points, nor a consequent result state, so they can only refer to the preparatory stage of the event, using the DURING function. Culminated processes, on the other hand, do have an end point, so both DURING and END functions can be used. Points can refer to the end points and the consequent result state (END and RESULT), while culminations can refer to all stages. It is often unclear when an event starts, so the START function is most often used when referring to the state of something at the exact moment an event begins to happen.

ALIVE is an example of a semantic predicate denoting a state, saying that at some stage in an event, the state of some thing is that of being alive. Another illustrative state example is shown in (69), where a predicate curiously called STATE is used to denote the end state of something (the Patient in this case). Note that the role EndState is a verb specific argument here, similar to the Form argument mentioned earlier.

(69) a. Bill dried the clothes. b. CAUSE(BillAGENT, E) &

STATE(RESULT(E), DriedENDSTATE, ClothesPAT IENT)

In other words, the result of the event is that the state of the clothes ends up being dried. The causer of the event is Bill.

There are close to 150 different semantic predicates in the VerbNet framework and they are all divided into four main categories [23]. These are the general predicates, the variable predicates, the specific predicates, and the predicates for multiple events. The general predicates are universal across all classes, and all languages, and examples include CAUSE and MOTION. The variable predicates, which include PREP, ADV and PRED, are all "in a one-to-one relation with a set of words in the language" [28, p. 39]. This means that they depend on the words used in the sentence in question. Specific predicates constitute specific verbal meanings, and these predicates are each shared across a few number of verbs. For example, the predicate SUFFOCATE involves suffocation, and it is used by a number a verbs, where suffocate is only one of them.

Finally, the predicates for multiple events relate two variables to each other. They essentially represent when the events happen according to one another, so a predicate like BEFORE(E0, E1) would indicate that event E0 happened before the event E1. Other examples include AFTER, which is the opposite of BEFORE, CONTAINS, which says that one event takes place inside the time interval of the other event, and EQUALS, which means that the two events start and end at the same points in time [28, p. 39-40].