«Knowledge Ascription by Grammatical Construction Laura A. Michaelis University of Colorado at Boulder 1. Introduction1 While little consensus has emerged ...»
Knowledge Ascription by Grammatical Construction
Laura A. Michaelis
University of Colorado at Boulder
While little consensus has emerged from the debate about the nature of know how, the
parties do appear to agree about two things: first, folk conceptions of knowledge matter
and second, linguistic analysis is a good way to get at those conceptions. Particular
attention has been paid to the syntactic behavior of verbs of knowledge ascription. The
rationale is presumably that a verb’s grammatical frame (i.e., it complement structure) reveals its conceptual structure—in particular, the repertoire of semantic roles that it evokes—and therefore a theory that captures the syntactic behavior of knowledgeascription verbs will also explain what kind of relationship verbs of knowledge ascription express. Thus, for example, Stanley (2011) rejects the Rylean view of know-how in part because it must treat as accidental the fact that both procedural knowledge and propositional knowledge are expressed by clausal complements consisting of a question
word followed by an infinitive, as in (1) versus (2), respectively:
1. She knows how to go.
2. She knows where to go.
The evidence suggests, however, that grammatical constructions do not provide a transparent window onto the meanings of verbs like know, because constructions can, and frequently do, alter the combinatoric potential of verbs with which they combine. As a straightforward illustration of this point, consider the activity verb sweep. It denotes a relation between a person and a surface in (3), but in (4) it denotes a relation between a
person and a substance:
3. She swept the floor.
4. She swept the dirt into a dustpan.
The semantic difference between (3) and (4) is attributable to syntactic context: in (3) sweep occurs in a simple transitive construction while in (4) it occurs in a construction the expresses causation of motion. How do these observations apply to verbs of knowledge ascription? Taking the verb learn as illustrative of the class, I submit that one cannot reasonably infer from the usage in (5), in which learn clearly denotes a relation
between a person and a proposition, that it denotes this same relation in either (6) or (7):
5. I learned that wider tires have better traction.
6. I learned to change a tire.
7. I learned how to change a tire.
Instead, as I will argue, (6) expresses a relation between a person and a procedure and (7) a relation between a person and a method of performing a procedure. In short, the The author gratefully acknowledges help and advice received from Marc Moffett, Knud Lambrecht, Josef Ruppenhofer and Adele Goldberg.
argument roles that a verb assigns differ according to its syntactic context. Does this then mean that the syntactic behaviors of verbs cannot or should not inform our models of knowing how? To the contrary, I will argue, the observed syntactic variability suggests a compromisebetween the Rylean and Intellectualist views: knowledge-ascription verbs assign complements denoting propositions, as per the Intellectualist view, while infinitival constructions assign complements denoting actions, as per the Rylean view.
The analysis that I will offer is based on Construction Grammar. According to Construction Grammar, rules of syntactic combination (like that which describes the noun phrase) are directly associated with interpretive and use conditions, in the form of semantic and pragmatic features that attach to the mother or daughter nodes in these descriptions (Goldberg 1995, Fillmore and Kay 1999, Kay 2002, Michaelis 2004, Sag 2010). This amounts to the claim that syntactic rules mean things. Meaning is generally viewed as the exclusive purview of words, and in the prevailing view of meaning composition, syntactic rules do no more than determine what symbol sequences function as units for syntactic purposes. So while syntactic rules assemble words and their dependent elements into phrases, and the phrases denote complex concepts like predicates and propositions, the rules cannot add conceptual content to that contributed by the words; nor can they alter the combinatoric properties of the words. On this view, which Jackendoff (1997: 48) describes as the “doctrine of syntactically transparent composition”, “[a]ll elements of content in the meaning of a sentence are found in the lexical conceptual structures […] of the lexical items composing the sentence”. A major problem with this view is that, as first observed by Goldberg (1995), syntactic context can in fact alter the combinatoric potential of words, as shown by the following attested
8. Down at the harbor there is a teal-green clubhouse for socializing and parties. Beside it sparkles the community pool. (Vanity Fair 8/01)
9. When a visitor passes through the village, young lamas stop picking up trash to mug for the camera. A gruff ‘police monk’ barks them back to work. (Newsweek 10/13/97) The verbs shown in boldface in (8-9), sparkle and bark, do not usually appear in these particular sentence patterns. By the same token, the sentence patterns exemplified in (8-9) usually contain verbs other than these. The pattern exemplified in (8), in which an intransitive verb precedes its subject and follows a location expression, favors verbs of location like sit and lie. The pattern exemplified in (9), in which a transitive verb is followed by both a direct object and a location expression, favors causative verbs that denote a change of location, e.g., move or push. The verb sparkle is not a verb of location; nor does bark express causation of motion. Counter to the predictions of the syntactically transparent composition, however, such verb-construction conflicts yield not gibberish but new verb meanings: the reader of (8) is inclined to interpret sparkling as the manner of location while the reader of (9) is likely to interpret barking as the (metaphorical) means by which motion is effected.
How are the verb-meaning shifts illustrated in (8-9) effected? Using basic tools of construction-based syntactic analysis, Goldberg (1995) provided a simple and conceptually satisfying answer: verb-construction conflict resolution. Her explanation starts with the foundational premise of Construction Grammar—that grammatical patterns like the ones exemplified in (8-9) have meanings, as indicated by the following
10. Locative inversion construction Form: a locative expression (A) preceding a verb whose subject (B) follows the verb.
Meaning: the location denoted by A has (or comes to have) entity B, in it.
11. Caused motion construction Form: a verb with a subject (A), object (B) and directional expression (C).
Meaning: the entity denoted by A causes the entity denoted by B to go to C.
Given these construction meanings, we can view the novel verb meanings in (8-9) as predictable by-products of verb-construction combination, or more specifically of verbconstruction conflict resolution—an inferential process in which a verb comes to denote the kind of event or state that the construction denotes (Michaelis 2004). In this process, the set of semantic roles associated with the verb is augmented up to that licensed by the construction, as when the verb sparkle acquires a location role. Since we need the locative-inversion and caused-motion constructions anyway, we do not need additional constructions to describe the special meanings in (8-9); nor do we need to create new verb definitions to describe those meanings. In order to reconcile the semantic contribution of verb and construction in such instances, the interpreter must relate the verb meaning to the construction meaning via what Goldberg (1995) calls an integration relation. Integration relations include manner, means and precondition. The manner and means relations are illustrated by (8-9), respectively, and the precondition relation will be illustrated below.
The construction-based model of argument structure resolves certain otherwise paradoxical properties of verbs. For example, while it seems a priori impossible for the verb bark to be transitive and intransitive at the same time, this apparent paradox disappears when we acknowledge the caused-motion construction as the source of the direct object in (9). Crucially for our purposes, the construction-based model of argument structure also suggests an intuitive way to resolve the conflict between the Rylean and Intellectualist positions: verbs of knowledge ascription do indeed, as claimed by the Intellectualists, take propositions as their second arguments, but whether or not that proposition actually surfaces as a complement will depend on the construction with which the verb combines. When, for example, verbs of knowledge-ascription verbs take bare infinitival complements, as in English I learned to drive or French Je sais conduire, they express a relation between a person and a procedure, as per the Rylean view, rather than a relation between a person and a proposition, as per the Intellectualist view. More generally, the appropriate semantic analysis of a verb of knowledge ascription will vary according to the complementation pattern in which the verb is encountered. As shown in (5-7), there are three major complementation patterns for verbs of knowledge ascription;
those examples are repeated here as (12-14):
12. Clausal complement: I learned that wider tires have better traction.
13. Infinitival complement: I learned to change a tire.
14. WH-complement: I learned how to change a tire.
In all but (12), we will assume that the verb’s proposition argument has been suppressed, and that the construction has supplied a distinct second argument. In the case of the infinitival-complement construction exemplified in (13), this second argument is a procedure. In the case of the WH-complement construction exemplified in (14), this second argument is a ‘means’ variable in an open proposition (i.e., ‘One changes a tire in x manner’). The remainder of this paper will be devoted to these two English constructions and the manner in which they interact with verbs of knowledge ascription.
Section 2 will provide an analysis of two major infinitival-complement constructions, Object Control and Subject Control. Section 3 will provide an analysis of the WH Complement construction. Section 4 will contain concluding remarks and a brief consideration of the functional contrast between the two apparently synonymous infinitive constructions exemplified in (13-14).
2. Infinitival complements In arguing against the Rylean view of knowing how, Stanley (2011) points out that it
entails a counterintuitive ambiguity for verbs of knowledge ascription. He states (p. 232):
The Rylean must argue that the English verb “know”, and the French word “savoir”, as well as their cognates in many other languages, are ambiguous between the propositional knowledge verb, and a verb attributing a distinct cognitive state, which is an attitude towards an action-type.
But in fact on a constructionist approach, savoir means the same thing in (15) and (16):
15. Je sais qu’il a raison.
‘I know that he is right.’
16. Je sais nager.
‘I know how to swim.’ Only the construction-integration relations are different in the two cases. Let us concur with Stanley that knowing is a relationship between a person and a proposition. This does not mean, however, that the construction in which a verb of knowledge-ascription appears denotes that relationship. As we saw in section 1 above, constructions can alter the relations that verbs express. The examples discussed in that section were of intransitive verbs (sparkle and bark) to which additional arguments had been added. A more complex case of argument augmentation, and one closer to the case of learn, know and other knowledge-ascription verbs, is that in which an already transitive verb takes a direct-object argument distinct from the one it intrinsically assigns. The verbs win and drink will here be used to illustrate this case. Intuitively speaking, the verb win expresses a relationship between a contestant and a prize, and drink a relationship between a person and a liquid, but these are not the relationships denoted by the constructions in (17-18),
17. He won me a stuffed animal.
18. He drank himself into a stupor.
Sentence (17) illustrates the Ditransitive (or ‘double object’) construction, whose (active voice) form is a verb followed by two noun phrases (in (17), me and a stuffed animal, respectively) and whose direct object (i.e., me) denotes the recipient of a transferred item.
Example (18) illustrates the Resultative construction, whose form is a verb followed by a NP and directional expression (in (18), himself and into a stupor, respectively), and whose direct object (i.e, himself) denotes something or someone who has undergone a change of state. In each of these two examples, there is a mismatch between the semantic roles that the verb calls for and those that the construction supplies: while winning requires only two participants (the victor and the prize), transfer requires three, and while drinking requires a direct object denoting a liquid, in (18) it gets a direct object denoting a human. Assuming the set of verb-construction integration relations described by Goldberg (1995), we can say that in (17) winning is understood to be a precondition for transfer while in (18) drinking is understood to be the means by which one moves (metaphorically) from sobriety to stupefaction. Notice in particular that the integration of the verb drink and the caused-motion construction in (18) requires the removal of the verb’s ordinary second argument (the potable substance) and the replacement of that argument with one licensed by the construction: the affected-party argument. We understand that the drink’s potable-substance argument is present conceptually (since one cannot drink without a liquid), but (18) denotes something that someone did to himself rather than to a beverage. In fact, the suppression of participant roles, and the consequent existential interpretation of those roles, is common in English and other languages (Fillmore 1986).
For example, the verb drink allows null expression of the potablesubstance argument in a variety of frames:
19. She drank from a mug.
20. He drinks.
Significantly, argument omission is also licensed by verbs of knowledge like teach:
21. She teaches first graders.
While the speaker of (21) does not specify the content that the teacher causes her students to know, the relevant content is presumably inferable from context. Were the content
argument to be present, it could be expressed by a noun phrase, as in (22):
22. She teaches first graders reading skills.
Like (18), (22) is an instance of the Ditransitive construction. This construction expresses an act of transfer, in which the direct object denotes a recipient and the second object (e.g., reading skills in (22) above) expresses the theme, or item transferred. (Most ditransitive verbs, including teach, are compatible with an alternate pattern in which the recipient is expressed by a preposition phrase, e.g., She gave a book to me.) While causing someone to know something is not literally an act of transfer, the fact that teach behaves syntactically as a transfer verb indicates that teaching can be construed according to the metaphorical mapping CAUSATION IS TRANSFER, also found in expressions like She gave me a headache and The judge handed him a victory (Lakoff 1993). Thus, the pattern of argument omission seen in (21) is the same as that in sentences like (23-24), in which
the theme argument is likewise missing and recoverable from context:
23. She emailed me [the news].
24. Give me [that book].
Alternatively, the missing complement of (21) might be reconstructed as a finite clause
introduced by the complementizer that, as in (25):
25. She teaches first graders that specific letter sequences correspond to words of English.