| | | Second Thoughts and Improvements |
The GF's (grammatical functions) represented in FrameNet documentation are taken from a limited set, including: Object, for the direct object of a transitive verb; Complement for almost everything else inside a verb phrase, as well as for post-head FE's in adjective phrases and noun phrases. For noun targets, Genitive refers to the possessive modifier (as in [the lad's] decision), Modifier for an adjectival modifier of the noun insofar as it identifies a frame element of the noun's frame (as in [financial] crisis) or the nominal modifier of the head noun of a compound (as in [foot] injury); in the case of an adjectival head, Modifier can be used as the GF of a degree or manner modifier of a predicating adjective ([far] distant), or a nominal modifier of an adjective in an adjectival compound (as in [light] sensitive). And in the case of an adjective used attributively, Head, for the head noun satisfying a frame element of the frame evoked by the adjective (as in missing [child]). There are a few other maximal-phrase-internal GFs (see appropriate section in Chapter 5) covering situations of extraposition and various sorts of discontinuities.
The list just given did not contain a GF Subject. All of those mentioned above are to be found inside the maximal phrase headed by the target word; but there are two situations in which phrases outside of the maximal phrase are functionally linked to a target word. The first involves what we call External, a cover term for anything that satisfies an FE requirement of a target word in any of the following syntactic contexts: (a) as a subject of a finite target verb; (b) as a subject or object of a controlling governor of the target (e.g., as in [the physician] decided to perform the surgery); (c) as the possessive modifier of a governing noun (as in [the physician's] decision to perform the surgery). Notice that in sentence (c) the phrase the physician's is in the Genitive GF relation to the noun decision but bears the Ext relation to the verb perform.
A subcase of a "controlling verb" is that of a verb that can be seen as specifically dedicated to providing external representation for an element of the conceptual structure associated with the meaning of a nominal or adjectival target; these we refer to as support verbs. Subjects and objects of support verbs can be taken as Ext for the dependent word; other complements of the support verb, where relevant, can be treated as complements of the dependent noun or adjective.
This last provision is what allows us to finesse the problem of disputed constituency decisions in the case of support-verb constructions. Thus, for a sentence like He made a statement to the press concerning the bribery case, it is contestable whether the phrase to the press is a complement of the noun statement or the verb make. Ignoring this dispute, FrameNet will tag the phrase as the Addressee complement of the target word. By allowing this phrase under either of the disputed analyses, we don't have to make the constituency decision ourselves.
It would have been theoretically justifiable to omit selecting phrases outside of the standard subcategorization frame of a target word, or we could have limited such excursions to the subjects of finite verbs. But since one of our goals was to provide a database that would include samples of phrases capable of satisfying particular FE requirements of the words we analyzed, this was one way of increasing the size of such a sample.
In the process of noting the function of such words, we have also taken on the obligation of recording the actual support verbs we encountered, thus allowing the database to be a resource for identifying the support verbs that most often accompany particular nouns and adjectives. In doing this, we have taken a relaxed rather than a strict view of identifying support verbs, including alongside of make (as in make an attack), also such semantically richer verbs as launch (as in launch an attack), etc. An extension of FrameNet could be defined which sought to include the full range of Lexical Functions in the sense of I. Mel'cuk.
Since the annotators were tagging the FE's of our target words with FE labels, and automatic processes were adding grammatical function (GF) and phrase type (PT) information, the original plan was to let the full combinatorial possibilities of our target words be determined automatically from the resulting annotations. But it became immediately obvious that this would not allow us to group example annotations by the sets of FEs that defined individual senses, since in many cases FEs that are conceptually a part of the frame are not actually expressed in the sentence. The Alembic Workbench annotator that we use allows us to introduce tags only if they surround pieces of text. Our solution was to introduce into corpus lines, adjacent to each target word, character strings that could bear FE annotations for FEs that were conceptually obligatory but that did not appear as lexical or phrasal material in the sentence []. But then we realized that, since we would need to add such material to corpus lines anyway, we could easily enrich the database by letting these dummy symbols carry certain lexicographically relevant information regarding omissibility conditions.
We posited three kinds of omissibility conditions in our sentences, two of which tend to be relevant to the description of lexical items. These can be divided into constructional, existential, and definite. Constructionally omitted constituents (variously also called structurally omitted) have their omission licensed by the grammatical constructions in which the target word appears and are therefore more or less independent of lexically specific information; these include the omitted subject of imperative sentences, the omitted agent of passive sentences, the omitted subjects of independent gerunds and infinitives ("PRO"), etc. The dummy symbol we use for constructionally null instantiated constituents is CNI.
Bring [CNI] me the head of Alfredo Garcia!Constructionally licensed omissibility is dependent on lexical information only to the extent that not all verbs can be the heads of imperative clauses, only transitive verbs can be made passive, etc. Constructionally licensed omissions are tagged, for the sake of making their FE available, but information about their occurrence is not being catalogued as part of the description of the lexical items in question.John was being followed [CNI].
It was like visiting [CNI] a tartar camp.
The first stage would be to go [CNI] to the schools and interview the teachers.
The indefinite cases (sometimes also referred to as existential) are illustrated by the missing objects of verbs like eat, sew, bake, drink, etc., that is, cases in which these ordinarily transitive verbs can be spoken of as "used intransitively". As is well known, there are often special interpretations of the existentially understood missing objects: for example, with eat the missing entity is likely to be a meal, with bake it is likely to be flour-based foods, with drink it is likely to be alcoholic beverages, etc.; but the essential difference between indefinite/existential and definite/anaphoric omissions is that with existential cases the nature of the missing element can be understood given interpretational conventions, but there is no need for anything about it to have been understood in advance. We put the FE annotations for indefinite null instanitation on the dummy symbol INI.
Have you eaten [INI] already?Joe started to drink [INI] again.
I've been baking [INI] all morning.
The definite (also called anaphoric) cases are those in which the missing element must be something that is already understood in the discourse context. For a sentence like Did anybody find out?, both speaker and addressee are in on what it is that somebody might have discovered; find out permits an anaphoric zero, whereas a verb like ascertain does not. In a sentence like Nobody objected, knowledge about the proposal which evoked no objection is taken for granted in the interlocutors' shared context. And so on. The symbol created to bear the FE annotations for definite null instantiation is DNI.
Did anybody find out [DNI]?There remain various situations in which the notation is unsatisfactory and decision-making about unexpressed FE's is not straightforward. There is what might be thought of as a Generic null instantiation which should probably be kept distinct from the others. With a question like Does your dog bite? one is asking about a general predisposition of a dog that might be a source of danger; an event report that used the verb bite without an object (My dog bit) does not permit either an INI nor a DNI interpretation, nor, in fact, a generic interpretation.Let's ask John. He will know [DNI].
Who came in first? -I think Mark won [DNI].
In the case of words with complex argument structures, the number of ways in which a frame can be elaborated by filling slots may be quite numerous, and we would not want to have to mark, as omitted, FE's standing for each of those elements when they are not realized. Thus, rather than worry about whether the verb write evokes a frame that includes Implement (with a ballpoint pen), Language (in Spanish), Surface (on expensive paper), etc., and marking each of these as missing in a sentence like Do you enjoy writing?, we will simply not mark any of these as INI and will instead describe the ways in which such elaborations are permitted in the entry itself.
There are numerous ways in which a single word incorporates multiple frames, sometimes joined in a time-patterned scenario, sometimes acting simultaneously.
Suppose we wish to put into a single frame words that have to do with someone passing judgment on someone or something, either positively or negatively. Examples might be blame, scorn, admire, etc. The canonical structure is
representing the Judge (A), the Evaluee (B) and the Reason for the judgment (for C-ing).A __ed B for C-ing
The list of such judgment words will include, in addition to those just given, words representing situations in which the Judge expresses the judgment in words, as with criticize and praise. (Criticizing and praising require talk; admiring and scorning do not.) To describe such cases we need to say that the Judge, in the judgment frame, is simultaneously the Speaker in a communication frame. It's possible for these words to understand the recipient of the talk to be identified with the topic of the talk, but they sometimes make room for an Addressee who is distinct from the Evaluee, as in She criticized me before the committee, He praised me to my principal.
But lastly, there are some words, like scold, berate, flatter, and compliment, in which the Evaluee is necessarily the same individual as the Addressee. In My boss scolded me, I am not only the object of his disapproval, but I have to listen to what he says about me.
We would like the database to include all such information, so that the words that participate in more than one frame can show up in response to queries about either of the frames.
There are some words that belong simultaneously to a conversation frame and a dispute frame. Interestingly, these two frames themselves involve a blend, between an abstract Reciprocity frame, operating in a great many domains, and speaking on the one hand and assailing on the other hand.
The reciprocity frame takes as input two (or more) subevents of some transitive relation and creates a complex scenario in which multiple participants operate at both ends of such a relation: the result, in the case of talking, A talks to B and B talks to A (Conversation); if it's assailing, then A assails B and B assails A (Fighting); if it's transferring, then A transfers something to B and B transfers something to A (Exchanging).
The special syntactic pattern alternations that go with reciprocity involve the option (with corresponding semantic difference) of representing the multiple participants either jointly, as a single (semantically plural) constituent, or as two constituents, one primary (the subject in an active sentence) and one secondary (the oblique constituent, typically marked by the preposition with). This gives us such formal differences as
[My brother and I] had a discussion.Conversation words are words (mostly verbs and deverbal nouns) having to do with conversational interaction. As reciprocals, they show up in reciprocal syntax, as with discuss, (have a) discussion, talk, (have a) talk, chat, (have a) chat, etc.[I] had a discussion [with my brother].
[My brother and I] fought.
[I] fought [with my brother].
[My brother and I] exchanged bicycles.
[I] exchanged bicycles [with my brother].
Fight words are words having to do with mutual assailing. As reciprocals they show up in reciprocal syntax: fight, (have a) fight, struggle, (engage in a) struggle, dispute, (have a) dispute, etc.
A semantic/syntactic characteristic of the words in the Fight frame has to do with the Issue over which there is disagreement; this FE can be marked with the preposition over. Notice: We fought over the land, over a woman, etc.
There is a class of words that belong to both the Fight frame and the Conversation frame, being words representing verbal disputes. Examples are argue, argument, dispute (verb and noun), quarrel (verb and noun), etc. While clearly involving conversation, they can be accompanied by an oblique PP indicating the Issue of dispute, and this can be marked by the preposition over. (We argued over the children's education, The family argued over the interpretation of the will.) The preposition over does not on its own communicate the idea of a quarrel or disagreement: it is not possible to import the semantics of quarreling into an ordinary conversation word by the addition of an over-phrase. (*We chatted over the inheritance doesn't work.)
It was suggested above that some frames are complex in that they designate sequences of states of affairs and transitions between them, each of which can itself be separately described as a frame. This is most obviously true in the case of words with resultative meanings (put differently, words that participate in resultative constructions). This can be seen most clearly in the case of verbs that can designate both simple and complex event types, and in the complex event situation they have syntactic objects unmotivated by their basic meanings. In many cases, these are uses that can be explained in terms of the integration of lexical meanings with the meanings of grammatical constructions. Thus, to talk oneself hoarse is to talk until one becomes hoarse; here talk is not a transitive verb, but in the resultative use, it can take a reflexive object; to run one's shoes ragged is to run in a pair of shoes until those shoes become ragged, where run is not by itself a transitive verb, but can take as object the name of the entity that undergoes the change through running; to squeeze the toothpaste out of the tube is to squeeze the toothpaste tube until the toothpaste comes out of it, though the actual activity of squeezing is applied more directly to the tube than its contents. In this case the verb is transitive, but the direct object does not correspond in the most straightforward sense to the object to which squeezing is applied. (Compare shake the apples out of the tree, where the act of shaking is applied directly to the tree.)
In many cases the most natural description of a frame specifies a list of conceptually obligatory FEs, but occasionally single constituents, sometimes because of the complex meanings of single words, contain information that could be spread over two constituents. Thus, the concept of ousting somebody from office requires an understanding of the (former) incumbent of the office and the identity of the office, and these both can be represented separately in a sentence like We ousted Jones as mayor. But in a sentence like We ousted the mayor, the direct object stands for both the office and the incumbent. Curing generally requires an understanding of a disorder or disease and a sufferer of that disease. Sentences which identify both are like We cured the patient of the disease. But there are words that designate sufferers of particular diseases, like leper, permitting both bits of information to be combined in one constituent: we cured the leper.
There are many verb frames which involve a particular kind of entity, in general, but for which there are some words which incorporate information about a particular entity type in their meaning. If we consider verbs of body movement, a typical case is a verb which is expected to co-occur with the name of a bodypart, even when the identity of the body part is clear from the meaning of the verb. A dog wags its tail, people arch their brows, bat their eyes, purse their lips, etc., but in the case of smile, grimace, frown, pout, and scowl, the affected body-part is not separately presented. We say that it is incorporated. Some verbs in this frame can take it or leave it: we can say either (She blinked or (She blinked (her eyes).
| | | Second Thoughts and Improvements |