Case Grammar

Shortly after Aspects was published in 1965, Charles Fillmore began working on a new kind of semantically oriented grammar. For more than a year mimeographed copies of his article outlining this model were circulated among linguists. Finally in 1968, the article entitled The Case for Case, was published in Universals in Linguistic Theory edited by Emmon Bach and Robert Harmes. In this article, Fillmore proposed that all languages have deep cases like Agent, Instrument, Experiencer, Object, etc. (These will be capitalized to distinguish them from the standard parts of speech, some of which have the same name.) He defined the Agent as the animate actor, the Instrument as the inanimate cause, the Experiencer (this term was suggested in his later work) as the animate receiver, the Object as the thing manipulated, etc. And although he took note of a correlation between subjects and Agents, between direct objects and Objects, between indirect objects and Experiencers, etc., he pointed out that this correlation is by no means perfect since subject, direct object, and indirect object are surface structure designations, while Agent, Object, and Experiencer are deep structure designations. Thus, to use his example, the Agent is a subject in John broke the window with a hammer, the Instrument is a subject in A hammer broke the window and the Object is a subject in The window broke. In showing that there is a partial correlation between the surface structure functions and deep structure cases, Fillmore noted that whenever there is an Agent in an active English sentence it is automatically the subject of the sentence: the Instrument is next in line of eligibility as a subject, and the Object is eligible to become the subject only if there is no Agent or Instrument in the sentence.

Fillmore generalized that each deep case has a particular preposition associated with it. For example, the preposition for the Agent case is by as in Paris was captured by the German army. The preposition for the Experiencer case is to as in John told a story to Mary. The preposition for the Instrumental case is with as in John hit the window with a hammer. And the preposition for the Object case is also with as in John loaded the truck with hay. He feels that the preposition is always present in the deep structure, but is lost whenever the noun phrase becomes a subject, a direct object, or an indirect object in a surface structure, as in By the German army captured Paris, John told to Marry a story, With a hammer hit the window, and John loaded with hay onto the truck, respectively. In the deep case, Fillmore considers all noun phrases to be prepositional phrases. He pointed out that the noun phrases can be moved around in the sentence or be deleted as long as the preposition remains with it, but as soon as the preposition is lost, as it is when the noun phrase becomes a subject or direct object, the noun phrase can no longer be moved or deleted.

To this point, we have been discussing Fillmore’s treatment of noun phrases. He handles verbs not in terms of deep cases, but rather in terms of deep case frames. That is, he would classify verbs according to which deep cases they co-occur with. In the sentence John jumped from one side of the ditch to the other side, the verb jump has associated with it an Agent (John), a Source (one side of the ditch), and a Goal (the other side of the ditch). The fact that the sentence, John jumped, is grammatical in English while the sentence, Jumped, is not grammatical, shows that for the verb jump an Agent is necessary, whereas a Source and a Goal are possible, but not necessary. Fillmore would therefore say that the verb jump has the case frame [+– A(S) G]. In reading this kind of notation of case frames, the brackets indicate that the enclosed cases are members of the set of cases that occur with a particular verb. The plus sign means that the cases mentioned are present rather than absent. The blank space stands for the position that the verb would fill if it were there. This blank space can be read as “in the context of”. Next Fillmore lists the cases which go with the verb. In actual use, at least in English, one of these cases will precede the verb because it will be acting as subject. In the sentence, John enjoyed the movie, both the Experiencer (John) and the Object (the movie) are necessary; this verb therefore has the case frame [+– E 0]. As another example, consider the sentence, John hit the wall (with a hammer). In this sentence, the Agent (John), and the Object (the wall) are necessary to the sentence, but the Instrument (a hammer) is not necessary. The verb hit therefore has the case frame [+– A(I) 0]. A very important consequence of this type of analysis is that a classification of verbs according to their case frames corresponds very closely to a classification of verbs according to their real-world significance. Thus, the case frame [+– E 0] is appropriate not only for the verb enjoy, but for all verbs of psychological events. And the case frame [+ A (I) 0] is appropriate not only for hit, but for all verbs of contact. And even more exciting is the possibility that [+– E 0] is appropriate not only for verbs of psychological events in English, but for verbs of psychological events in all human languages. Because Fillmore is dealing with universal concepts, he cautions that the cases are unordered in these case frames, although it may be decided later that ordering is necessary so as to represent eligibility for such things as subject marking if this is shown to be a universal hierarchy.

Two important criticisms have been directed at the case model. The first question is whether deep cases are adequate as a universal linguistic base. The second question is related to the first; if it is shown that something more basic than deep cases, such as semantic features, is required as a universal linguistic base, then can deep cases be justified as an intermediate structure somewhere between deep structure and surface structure? Our own feeling is that although deep cases provide some very important insights into the workings of language, they are too gross to be semantic primes. Therefore we would make only the weaker claim that they represent an intermediate structure. And even with this claim, further work may show that semantic features by themselves are perfectly adequate and possibly even simpler than deep cases in explaining the fascinating facts which Fillmore brought to light.

In addition to his statements regarding deep cases and case frames, Fillmore made a number of important observations about what the generative semanticists call predicate raising. Fillmore handled such information by entailment rules in a kind of hierarchy, whereby require would be said to entail permit; certain to entail possible; steal and accuse to entail both criticize and responsible, etc. He would say that persuade entails cause and believe, while the generative semanticists would postulate these expressions as lower predicates. Fillmore is also very interested in figuring out presupposition, i. e., what a speaker assumes to be true and known before he says anything. For example, in the sentence, Harry realizes that John is President, it is presupposed that John is President. Since negation does not affect presupposition, the sentence Harry doesn’t realize that John is President, also presupposes that John is President. One aspect of presupposition which Fillmore is especially interested in is that resulting from various kinds of orientation. A cube in space has six sides. If this cube is sitting on the ground, it has a top, a bottom, and four sides. If in addition, one side is more ornate than the others, it has a top, a bottom, a front, a back and two sides. And finally, if there is some reason to distinguish one side from the other, it has a top, a bottom, a front, a back, a right side, and a left side. A dresser would be an example of this last situation, and it would be noteworthy that the right-hand side of a dresser is the same as the right-hand side of the person who is facing the dresser, rather than the right side of the dresser itself. Fillmore has been very concerned with such aspects of language, although it is not entirely clear how this fits into his Case Model grammar.

In this Case Model grammar Fillmore has been basically interested in justifying his deep cases by morphological and syntactic considerations. He uses several tests for deciding on the membership of a case. One such test is based on the assumption that only members of the same case can be conjoined. For example, with the acceptable John and Mary broke the window, there are two Agents joined as the subject. But the sentence, John and a hammer broke the window, is deviant because an Agent and an Instrument have been joined. Another test Fillmore uses is that there can be only a single use of a deep case per simple predicate. For example, John broke the window with a hammer, is acceptable, but The hammer broke the window with a chisel, is unacceptable because of the two different uses of the Instrument case. Fillmore’s third test for case membership has already been mentioned. It is the particular preposition used as a marker, i. e., John is an Agent in the sentence The cookie was stolen by John, because John can be marked with the preposition by. By is still the marker in the deep structure of John stole the cookie, but it has been deleted from the surface structure as happens when an Agent noun phrase becomes the subject of the sentence.

Some of the followers of Fillmore have not placed such a heavy reliance on syntactic and morphological considerations for determining deep cases. Although they have not contested Fillmore’s syntactic and morphological tests, they have supplemented these tests with semantic evidence stated in terms of semantic features. [...]

Two books based on the case model are John M. Anderson’s The Grammar of Case: Toward a Localistic Theory (1971) and D. Terence Langendoen’s Essentials of English Grammar (1970).

Wallace Chafe in his Meaning and the Structure of Language (1970) states that When introspection and surface evidence are contradictory, it is the former which is decisive, (page 122). His model of grammar differs from Fillmore’s in that it gives priority in assigning cases to semantic categories rather than to syntactic or morphological considerations. Chafe has a lengthy discussion of various kinds of semantic features: derivational, inflectional, and selectional. Chafe also postulates a direct relationship between case frames and semantic categories of verbs such as state, process, event, and action. He indicates that a state is a one-place predicate which is filled by a Patient (what Fillmore calls an Object) as in The wood is dry. He considers a process to be a change of state which is also a one-place predicate with a Patient filling that one place as in The wood dried. He considers an action also to be a one-place predicate with the one place being filled by an Agent as in Michael ran. An event is a combination of an action and a process. Thus Chafe would conceive of the sentence, Michael dried the wood, as an event consisting of an action (Michael acted) and a process (the wood dried). Since an event consists of an action, which has an Agent, and a process, which has a Patient, it is logical that an event such as Michael dried the wood, would be considered a two-place predicate consisting of an Agent (Michael), and a Patient (the wood).

It can be seen from the above examples that Chafe’s grammar makes use of deep cases. Unlike Fillmore, however, Chafe marks these deep cases as being hierarchical. For example in the sentence, John made Bill a tie clasp, Fillmore would consider the Agent (John), the Benefactive (Bill), and the Object (a tie clasp) all to be on the same level. But Chafe would diagram this sentence as follows so that it would show the subject-predicate dichotomy:

Chafe’s diagramming tends to superordinate those cases likely to become subjects and to subordinate the others.

Chafe is also very interested in discourse analysis, i. e., the grammar of items larger than the sentence. He points out, for example, that such a sentence as I bought a car yesterday, but the fender was bashed in, is not ungrammatical, even though the definite article the is marking a previously unidentified noun. His point is that when a speaker introduces a concept like car he is at the same time introducing all of the parts that one assumes are automatically a part of all (or most) cars. When someone mentions a car, it can be assumed that the hearer has conceptualized a car in his mind with many of its parts, and it is therefore possible to refer to the windshield, the tires, the hand brake, etc, even though the windshield, tires, and hand brake have not been previously designated independently. This observation has important ramifications in considering the nature of semantic features.

One of the most important considerations in discourse analysis is the pronoun system. Postal and others have shown that pronouns have both deep structure and surface structure constraints. Relative pronouns, personal pronouns, and reflexive pronouns must all have antecedents, but Paul Postal has pointed out that identity of sense is a necessary but not a sufficient condition to allow pronominalization. Identity of reference is also required. The sentence, Max’s parents are dead and he deeply misses them, is grammatical because the two noun phrases parents and them have both identity of sense and identity of reference. However, the sentence Max is an orphan, and he deeply misses them, is ungrammatical because the two noun phrases orphan and them have identity of sense in that orphan means a person without parents and them refers to this incorporated term parents, but they do not have identity of reference in that orphan and them do not refer to exactly the same thing. Stated differently, Postal has shown that a feature must reach the surface structure in order to be pronominalized.