The semantics of verbs

2.1. Comparisons between approaches
Comparisons between approaches
It is quite difficult to compare the three above approaches. They are based on very different assumptions. We can however indicate that classes constructed on syntactic criteria are of much interest from a theoretical point of view, in the study of the cooperation between syntax and semantics. They are certainly less useful in LKB design since they are far from complete and include many forms of exceptions.
The approach based on general semantic criteria is much more concrete and applicable. However, the classes which can be formed on this basis remain very general. Classes formed using the ontological criteria of WordNet, from that point of view, are more fine-grained, and they should be prefered (see section on WordNet 3.4). The LCS-based classification is also fine-grained, its main advantage is to base the classification on semantic frames, which can then be used for semantic representation. Frames may be felt to be more arbitrary, but, in fact, they share a lot with the previous classification method. LCS representations are more formal, they allow thus more precise classifications, but they may also be felt to be incomplete (because of their formalism based entirely on predicates and functions) and to be difficult to establish. Depending on the language, verbs may express grammatical tense, aspect, or modality. Grammatical tense is the use of auxiliary verbs or inflections to convey whether the action or state is before, simultaneous with, or after some reference point. The reference point could be the time of utterance, in which case the verb expresses absolute tense, or it could be a past, present, or future time of reference previously established in the sentence, in which case the verb expresses relative tense. Aspect expresses how the action or state occurs through time. Important examples include: perfective aspect, in which the action is viewed in its entirety through completion (as in "I saw the car") imperfective aspect, in which the action is viewed as ongoing; in some languages a verb could express imperfective aspect more narrowly as:
habitual aspect, in which the action occurs repeatedly (as in "I used to go there every day"), or continuous aspect, in which the action occurs without pause; continuous aspect can be further subdivided into
stative aspect, in which the situation is a fixed, unevolving state (as in "I know French"), and progressive aspect, in which the situation continuously evolves (as in "I am running") perfect, which combines elements of both aspect and tense and in which both a prior event and the state resulting from it are expressed (as in "he has gone there", i.e. "he went there and he is still there") discontinuous past, which combines elements of a past event and the implication that the state resulting from it was later reversed (as in "he did go there" or "he has been there", i.e. "he went there but has now come back") Aspect can either be lexical, in which case the aspect is embedded in the verb's meaning (as in "the sun shines," where "shines" is lexically stative), or it can be grammatically expressed, as in "I am running."
Modality[10] expresses the speaker's attitude toward the action or state given by the verb, especially with regard to degree of necessity, obligation, or permission ("You must go", "You should go", "You may go"), determination or willingness ("I will do this no matter what"), degree of probability ("It must be raining by now", "It may be raining", "It might be raining"), or ability ("I can speak French"). All languages can express modality with adverbs, but some also use verbal forms as in the given examples. If the verbal expression of modality involves the use of an auxiliary verb, that auxiliary is called a modal verb. If the verbal expression of modality involves inflection, we have the special case of mood; moods include the indicative (as in "I am there"), the subjunctive (as in "I wish I were there"), and the imperative ("Be there!"). The meaning of a verb can be represented by—and, thus, reduced to—a list of semantic role labels, each identifying the semantic relation an argument bears to it.
A REPRESENTATIVE SET OF SEMANTIC ROLES:
a. Agent (A), the instigator of the event
b. Counter-Agent (C), the force or resistance against which the action is carried out
c. Object (O), the entity that moves or changes or whose position or existence is in consideration
d. Result (R), the entity that comes into existence as a result of the action
e. Instrument (I), the stimulus or immediate physical cause of an event
f. Source (S), the place from which something moves
g. Goal (G), the place to which something moves
h. Experiencer (E), the entity which receives or accepts or experiences or undergoes the effect of an action ...
Why people like role-centered approaches
ADVANTAGES OF SEMANTIC ROLE LISTS:
— Capture relationships between sentences that are obscured by syntax
— Provide a basis for perspicuous statements of argument realization ‘... case structure descriptions of words and sentences offered a level of linguistic organization at which universal properties of lexical structure and clause organization are to be found, and, moreover, ... such descriptions were in some sense intuitively relatable to the ways people thought about the experiences and the events that they were able to express in the sentences of their language. ‘At the entry level at least anybody can do it [=case grammar], in the privacy of their own home, with no linguistic training required! ...
The reason that anyone can do this kind of case grammar is that football fans for example are very knowledgeable about Goals, aficionados of “General Hospital’ or “St. Elsewhere” know all about Patients, and anyone who has watched a James Bond movie can identify an Agent.’
Implementing a role-centered approach: The crucial issues
CRUCIAL QUESTIONS:
— How do you choose a set of semantic roles?
— How do you assign semantic roles to the nps in a sentence?
CHOOSING A SET OF SEMANTIC ROLES:
— Need to cover types of semantic relationships nouns bear to verbs.
— Choose roles that capture natural classes of arguments to bring out similarities and differences in verb meaning reflected in argument expression.
— Set shouldn’t be so big that it is impossible to capture generalizations, nor so small that it doesn’t allow for distinctions (the lumping/splitting dilemma).
FUNDAMENTAL ASSUMPTIONS FOUND IN MOST SEMANTIC ROLE LIST APPROACHES:
A. The semantic roles are taken to be semantically unanalyzable,
B. The semantic roles are defined independently of the meaning of the verb,
c. the set of semantic roles is small in size.
CHARACTERISTICS OF THEORIES OF SEMANTIC ROLES:
a. INDEPENDENCE: Each semantic role is given a consistent semantic definition that applies to all verbs and all situations, e.g., role definitions do not depend on the identity of a particular verb or on other roles it assigns.
b. COMPLETENESS: Every argument of every verb is assigned some semantic role.
c. UNIQUENESS: Every argument of every verb is assigned only one semantic role.
d. DISTINCTNESS: Every argument of every verb is distinguished from the other arguments of the same verb by the role(s) it is assigned.
STRONG (BIUNIQUE): when uniqueness also holds
WEAK: Uniqueness doesn’t hold, i.e., each argument is assigned a different set of roles from other arguments of same verb
THE EFFECT OF UNIQUENESSS: Precludes labelling Sam both a benefactive and a goal:
I bought Sam a car.
THE EFFECT OF DISTINCTNESS:
— Acknowledges the oddness of ‘One-per-sent’ violations
Pat sent the book to Tracy to Sandy.
— Precludes labelling both crane and hay as instruments:
The farmer loaded the truck with hay with a crane.
a. The farmer used the crane to load the truck with hay.
b. # The farmer used the hay to load the truck.
MORE GENERAL REQUIREMENTS ON A SET OF PRIMITIVE PREDICATES:
(Though these criteria can be adapted to semantic roles)
a. FINITUDE: set of primitives should be smaller than set of words whose meaning it encodes
b. COMPREHENSIVENESS: set should be adequate to express and distinguish the senses of the word meanings it is to encode
c. INDEPENDENCE: no primitive should be definable in terms of other primitives
d. NONCIRCULARITY: no two primitives should be mutually definable
e. PRIMITIVENESS: no subset of primitives should be such that it could be plausibly replaced by a smaller defining set
• ON THE DIFFICULTIES OF FINDING PRIMITIVES:
Hunting for semantic and lexical universals is not like pearl-fishing. Primitives do not present themselves glittering and unmistakable. Identifying them is an empirical endeavour but one that calls for much interpretative effort.
Semantic roles lose their shine
Semantic roles did not live up to their promise, as manifested in:
— Lack of consensus on number and set of roles.
— Lack of reliable diagnostics for isolating semantic roles.
– Diagnostics identify something else (e.g., syntactic, pragmatic notions). – Grammatical markers are often inadequate (e.g., with).
a. Instrument: slice with a knife
b. Comitative: work with Kim
c. Theme: spray a chair with white paint
d. Cause: shiver with cold
e. Manner: laugh with glee
— Disagreements about role assignments to particular NPs.
a. Instrument or theme: I sprayed the wall with paint.
b. Benefactive or goal: I bought Sam a car.
— Recourse to “wastebasket” patient or theme role or to one-off roles.
a. The engineer avoided the bridge.
b. The engineer studied the bridge.
c. The engineer confirmed the start date.
Problems with semantic roles:
— Different phenomena require different grain-size roles.
a. Pat came to the library/from the cafeteria/through the woods.
b. Pat came from the cafeteria to the library.
— Problem of near-paraphrases: capturing similarities and differences.
An assignment that captures the near-paraphrase relation:
. Terry loaded the truck (location) with hay (theme).
b. Terry loaded the hay (theme) on the truck (location).
But it fails to capture the perception that it is the object in each sentence that is “affected”.
— Inability to capture crossclassification: e.g., case syncretisms.
— Problems of overall explanatory effectiveness:
– No insight into argument realization generalizations.
– Why are verbs at most triadic?
– Why do only some sets of semantic roles cooccur?
a. ‘Agent, Patient, Instrument’: e.g., break, cut, mix
b. ‘Theme, Source, Goal’: e.g., run, swim, walk
c. ‘Location, Experiencer, Patient’: No verbs?
ONE SOLUTION: Semantic roles are labels for recurring sets of entailments of arguments that enter into generalizations.
THE BOTTOM LINE: A small universal set of semantic roles does not provide an articulated enough lexical semantic representation.
Semantic role representations reduce a verb’s meaning to a set of roles assigned to its arguments.
The set of semantic roles is taken to be independent of the verbs, and there is no organizational framework for determining or constraining what the overall set of roles should be, nor what set is appropriate for a given verb.
Event conceptualization and lexical semantic representation
Two dimensions of variation in lexical semantic representations:
— the nature of the representation, e.g., semantic roles, predicate decompositions — the model of event conceptualization, i.e., the organizing hypothesis
Why is a model of event conceptualization important?
Such a model embodies a hypothesis about the way in which events are organized in language.
The most successful theories of lexical semantic representation are organized around such a hypothesis, which helps to choose among various meaning components and semantic representations.
TWO COMPETING PROPOSALS FOR THE MODEL OF EVENT CONCEPTUALIZATION:
— in terms of the causal structure of the event: a causal approach — in terms of the time course of an event: an aspectual approach
Certain meaning components said to be relevant to semantic representations recur in various studies:
— Causation: CAUSE
— Change: BECOME, CHANGE, GO, T (transition)
— State: BE
— Measuring out
— Telicity: endpoint/goal/terminus/telos
— Affectedness: Patient role, AFF
— Agency: Agent role, ACT, DO
— Process/activity: ACT, DO, P (process)
The items listed reflect a debate as to whether an aspectual, a causal, or a hybrid model of event conceptualization is preferable.

2.2. Relations with other areas of lexical semantics

It is clear that verbs is one of the most central syntactic category in language. They have deep relations with the other categories: nouns because they select arguments which are often nominals, adverbs because adverbs modify verbs, prepositions, since they introduce PPs. Verbs assign thematic roles to their arguments and to prepositions, which, in turn assign thematic roles to NPs. Verbs associated with adverbs permit the computation of aspect.
Verb semantic classes in LKB
For the same reasons as above, verbs have a priviledge position in LKBs. The following resources have developed a quite extensive description of verbs, described in the next chapter of this report: 3.6, 3.9, 3.10, 3.10.2.
Verb semantic classes in Applications
Verbs are central in many applications, in particular in Machine Translations 4.1, 4.1.3. They are now becoming of much interest in automatic indexing 4.3, 4.2.3 and information retrieval 4.2, 4.2.3 where indexes, previously based key-words, are now formed of verbs and arguments, in order to describes concepts but also actions or states. Projects of much interest from that point of view are reported in.
Verbs vary by type, and each type is determined by the kinds of words that accompany it and the relationship those words have with the verb itself. Classified by the number of their valency arguments, usually three basic types are distinguished: intransitives, transitives, ditransitives and double transitive verbs. Some verbs have special grammatical uses and hence complements, such as copular verbs (i.e., be); the verb "do" used for do-support in questioning and negation, and tense or aspect auxiliaries, e.g., "be", "have" or "can". In addition, verbs can be nonfinite, namely, not inflected for tense, and have various special forms such as infinitives, participles or gerunds. Intransitive verbs
An intransitive verb is one that does not have a direct object. Intransitive verbs may be followed by an adverb (a word that addresses how, where, when, and how often) or end a sentence. For example: "The woman spoke softly." "The athlete ran faster than the official." "The boy wept."
Transitive verbs
A transitive verb is followed by a noun or noun phrase. These noun phrases are not called predicate nouns, but are instead called direct objects because they refer to the object that is being acted upon. For example: "My friend read the newspaper." "The teenager earned a speeding ticket."
A way to identify a transitive verb is to invert the sentence, making it passive. For example: "The newspaper was read by my friend." "A speeding ticket was earned by the teenager."
Ditransitive verbs
Ditransitive verbs (sometimes called Vg verbs after the verb give) precede either two noun phrases or a noun phrase and then a prepositional phrase often led by to or for. For example: "The players gave their teammates high fives." "The players gave high fives to their teammates."
When two noun phrases follow a transitive verb, the first is an indirect object, that which is receiving something, and the second is a direct object, that being acted upon. Indirect objects can be noun phrases or prepositional phrases.[2]
Double transitive verbs
Double transitive verbs (sometimes called Vc verbs after the verb consider) are followed by a noun phrase that serves as a direct object and then a second noun phrase, adjective, or infinitive phrase. The second element (noun phrase, adjective, or infinitive) is called a complement, which completes a clause that would not otherwise have the same meaning. For example: "The young couple considers the neighbors wealthy people." "Some students perceive adults quite inaccurately." "Sarah deemed her project to be the hardest she has ever completed."
Copular verbs
Copular verbs (a.k.a. linking verbs) can't be followed by an adverb or end a sentence, but instead must be followed by a noun or adjective, whether in a single word or phrase. Common copulae include be, seem, become, appear, look, and remain. For example: "His mother looked worried." "Josh remained a reliable friend." Copulae are thought to 'link' the adjective or noun to the subject.
The copular verb be is manifested in eight forms: be, is, am, are, was, were, been, and being in English. These verbs precede nouns or adjectives in a sentence, which become predicate nouns and predicate adjectives similar to those that function with a linking verb. They can also be followed by an adverb of place, which is sometimes referred to as a predicate adverb. For example: "Her daughter was a writing tutor." "The singers were very nervous." "My house is down the street."
Adjectives that come after copular verbs are predicate adjectives, and nouns that come after linking verbs are predicate nouns.
The origin of the field theory of semantics is the lexical field theory introduced by Jost Trier in the 1930s, although according to John Lyons it has historical roots in the ideas of Wilhelm von Humboldt and Johann Gottfried Herder. In the 1960s Stephen Ullmann saw semantic fields as crystallising and perpetuating the values of society. For John Lyons in the 1970s words related in any sense belonged to the same semantic field,[10]:32 and the semantic field was simply a lexical category, which he described as a lexical field. Lyons emphasised the distinction between semantic fields and semantic networks. In the 1980s Eva Kittay developed a semantic field theory of metaphor. This approach is based on the idea that the items in a semantic field have specific relations to other items in the same field, and that a metaphor works by re-ordering the relations of a field by mapping them on to the existing relations of another field. Sue Atkins and Charles J. Fillmore in the 1990s proposed frame semantics as an alternative to semantic field theory.