2.1 Semantic similarities

Semantically, the three constructions have been said to be (roughly) equivalent (Wilkinson Reference Wilkinson1970: 430, 1976: 164–166; Jackendoff Reference Jackendoff1972: 57, among others). They all involve (i) an individual a (John in (1)), (ii) a property P ₁ that describes a characteristic of an individual (wise), and (iii) another property P ₂ which typically describes an action (leave early) as semantic components. Also, they all force a peculiar ‘relativized’ interpretation of P ₁ (see below).

Some adverbs (adverbial forms) are ambiguous between the ‘subject-oriented’ use and the ‘verb-oriented’ use. Jackendoff (Reference Jackendoff1972: 49) notes, for example, that (2a) is ambiguous between (3a) and (3b); when the adverb is fronted to precede the subject (as in (2)), the subject-oriented interpretation is evoked.Footnote ⁴

1. (2)

   1. (a) John cleverly dropped his cup of coffee.

   2. (b) Cleverly, John dropped his cup of coffee.

1. (3)

   1. (a) It was clever of John to drop his cup of coffee.

   2. (b) The manner in which John dropped his coffee was clever.

Adjectives (adjectival bases) that can fill in the P ₁ slot of the three constructions include those listed below, which all describe a mental/behavioral propensity of a sentient individual.

1. (4) bold, brave, careful, careless, clever, clumsy, (in)considerate, cowardly, crazy, cruel, foolish, impudent, (un)kind, mean, nice, (im)polite, right, (un)wise, wicked, wrong, smart, stupid

There is another, smaller class of adjectives which can participate in the so-adv-cxn and adj-to-inf-cxn, but seemingly not the adj-of-np-cxn.

1. (5) (un)lucky, (un)fortunate

1. (6)

   1. (a) Luckily, John passed the exam.

   2. (b) John was lucky to pass the exam.

   3. (c) ?*It was lucky of John to pass the exam.

      (cf.) It was lucky for John to pass the exam.

Following Wilkinson (Reference Wilkinson1970, Reference Wilkinson1976), I call the former type of adjectives the W(ise) class, and the latter type the L(ucky) class.

Roughly, we can make the following generalizations: (i) the adj-of-np-cxn selects for the W class (and nothing else) for its P ₁ slot, and (ii) the so-adv-cxn and adj-to-inf-cxn select for both the W class and the L class (and nothing else) for their P ₁ slot. There seem to be some exceptions/irregularities, however. For one thing, adjectival bases like silly and cowardly do not have corresponding adverbs derived with -ly, or at least such forms are awkward and marginal (e.g., *~??sillily). This apparently has to do with a phonological (dissimilatory) constraint that disfavors the sequence /ləli/.

Also, as noted by Jackendoff (Reference Jackendoff1972: 57), the base careful is rarely used in the adj-of-np-cxn, although it appears to belong to the W class (the judgements in (7) are Jackendoff's).

1. (7)

   1. (a) Carefully, John spilled the beans.

   2. (b) John was careful to spill the beans.

   3. (c) *It was careful of Jonn to spill the beans.

This is probably because the sequence careful of evokes the ‘transitive’ interpretation (of careful) as in ‘The beaver stayed in the water, because it was careful of predators’, thereby obscuring the other interpretation where the same sequence is part of the adj-of-np-cxn.

There may be other exceptions, and judgements may vary across speakers as to the acceptability of certain combinations of bases and constructions.

2.2 Syntax

Let us next consider some basic syntactic properties of the three constructions.

2.3 The ‘relativized’ interpretation of P1

One intriguing issue about the three constructions is the fact that none of them appears to entail P ₁(a) in a straightforward way. Regarding the adj-to-inf-cxn, Barker (Reference Barker2002) illustrates this point with the following examples.

1. (21)

   1. (a) Feynman is stupid.

   2. (b) Feynman is stupid to dance like that.

The same remark applies to the so-adv-cxn and the adj-of-np-cxn.

One may be tempted to solve this puzzle by resorting to the distinction of essential (permanent) vs. transitory (temporary) properties. Stowell (Reference Stowell and Rothstein1991), in this spirit, proposes that dispositional adjectives like stupid and kind are coerced to receive a stage-level interpretation when combined with an infinitival complement.Footnote ⁸

A simple analysis along these lines, however, cannot cope with the observation made by Wilkinson (Reference Wilkinson1970: 432) that the adj-to-inf-cxn and the adj-of-np-cxn (but not the so-adv-cxn; see below) cannot be embedded under a predicate of intention, desire, or command (see also Barker Reference Barker2002: 3, 26ff.); in the examples to follow, the diacritic # indicates unacceptability that is not necessarily caused by a syntactic reason.Footnote ⁹

1. (22)

   1. (a) #Feynman intended [to be rude to be utterly honest].

   2. (b) #I wanted it to be stupid of Feynman to dance like a fool.

   3. (c) #Bob asked the man to be bold to dash into the cave.

As can be easily confirmed, transitory predicates in general do not resist embedding under intention, command, or desire predicates.

1. (23)

   1. (a) Feynman intended to be available.

   2. (b) I wanted it to be sunny.

   3. (c) Bob asked the man to be in the garden.

To account for the unembeddability of the adj-to-inf-cxn and the adj-of-np-cxn, it therefore does not suffice simply to assign a transitory interpretation to P ₁. The analysis to be developed below, nevertheless, will build in ‘P ₁ on its transitory interpretation’ as an essential component.

In the following I will assume that each W and L adjective allows both essential and transitory interpretations, and for simplicity, that the latter is derived from the former by applying the modifier ‘transitorily’, without being specific about the detailed semantic relation between the two interpretations. The modifier is often implicit, but may be explicitly coded by such means as the ‘progressive’ form of be.

1. (24)

   1. (a) John is nice ↦ nice(john)

   2. (b) John is being nice ↦ (transitorily(nice))(john)

3.1 A hidden conditional

I propose that the meaning shared by the three constructions involves an epistemic conditional statement, and is along the lines of ‘a does P ₂, and from this (and other known facts) it can be inferred that a is transitorily P ₁’. ‘John was wise to leave early’, for example, can be paraphrased as ‘John left early, and given that John left early, he must have been being wise’, or more succinctly, ‘John must have been being wise, {because/considering that} he left early’.

This analysis solves the entailment puzzle and the embeddability puzzle at once. P ₁(a) on its typical essential interpretation is not entailed, because the transitory interpretation is singled out. Also, under this analysis, the issue of unembeddability illustrated in (22) can be treated as a special case of the phenomenon illustrated below.Footnote ¹⁰

1. (25) John is in the office, if the light is on.

   (≈ It can be inferred that John is in the office, if the light is on.)

1. (26) #John intended [to be in the office, if the light is on].

1. (27)

   1. (a) #I want John [to be in the office, if the light is on].

   2. (b) #I want it to be the case [that John is in the office, if the light is on].

1. (28)

   1. (a) #I asked John [to be in the office, if the light is on].

   2. (b) #I asked [that John be in the office, if the light is on].

Quite generally, statements that involve epistemic modal quantification resist embedding under predicates of intention, desire, or command.Footnote ¹¹

1. (29) Look, John left a $20 tip. He must be rich.

1. (30) #John intends to be such that he must be rich.

1. (31)

   1. (a) #I wanted John to be such that he must be rich.

   2. (b) #I wanted it to be the case that John must be rich.

      (cf.) I wanted it to be the case that John is rich.

1. (32)

   1. (a) #I asked John to be such that he must be rich.

   2. (b) #I asked that John must be rich.

      (cf.) I asked that John be rich.

This, intuitively, appears to be because epistemic modal statements cannot be made true by manipulating entities in the actual world, and in this sense are beyond one's intention, desire, or command. We may call this property ‘uncontrollability’.Footnote ¹²

It is interesting to note, in this connection, that the so-adv-cxn, in contrast to the other two constructions, is embeddable.

1. (33)

   1. (a) (?)Ken intended to nicely help out John.

   2. (b) (?)Ken wanted John to wisely make a formal apology.

   3. (c) (?)Ken asked John to boldly dash into the cave.

Sentences like (33a–c) may sound somewhat awkward, but I believe that it is only because of a ‘split infinitive’.

This contrast can be attributed to the fact that the adj-to-inf-cxn and the adj-of-np-cxn on the one hand and the so-adv-cxn on the other assert/presuppose different semantic components. The adj-to-inf-cxn and the adj-of-np-cxn assert that ‘from P ₂(a) it can be inferred that P ₁(a)’, and presupposes that ‘P ₂(a)’; in the so-adv-cxn, the assertion and presupposition are switched (see below). Generally, the presupposed meaning of a complement clause is not interpreted as (part of) the target of intention, desire, etc., as can be seen from the fact that (34a) may well be true when (34b) is false (see Heim Reference Heim1992, Oshima Reference Oshima2006).

1. (34)

   1. (a) Feynman wanted to manage to talk to Seinfeld.

   2. (b) Feynman wanted it to be difficult to talk to Seinfeld.

In the adj-to-inf-cxn and the adj-of-np-cxn, the asserted component, which is the target of intention, desire, etc., is uncontrollable, and this leads to unacceptability. In the so-adv-cxn, the asserted component is controllable, and thus such a problem does not arise.

3.2 Logical formulation

The epistemic conditional inference of the form ‘If p, then (it must be the case that) q’, which is a hidden semantic component of the three constructions, can be formulated as a four-place relation that holds of (i) an epistemic modal base (i.e., a set of worlds that are compatible with the known facts), (ii) an ordering source based on stereotypicality (i.e., a set of propositions that are in accordance with the normal course of events), and (iii) two propositions, one of which serves as a restrictor of the modal base (Kratzer Reference Kratzer1986, Reference Kratzer, von Stechow and Wunderlich1991). Ignoring the ordering source for ease of presentation, the logical form of an epistemic conditional can be approximated as:

1. (35) □_e[p→q]

where □_e is a necessity operator that quantifies over worlds that are epistemically accessible from the world of evaluation. The basic semantic format of the three constructions can then be given as follows.

1. (36) P ₂(a) ∧□_e[P ₂(a)→(transitorily(P ₁))(a)]

If we instantiate a, P ₁, and P ₂ with ‘John’, ‘wise’, and ‘leave early’, we obtain the following, which can be informally paraphrased as ‘John left early, and from this and other known facts, it can be inferred that John was being wise’.

1. (37) leave-early(john)∧□_e[leave-early(john)→(transitorily(wise))(john)]

To refute this statement, one must either convince the utterer that as a matter of fact John did not leave early, or that John was not being wise despite the fact that he left early (i.e., the presented inference does not hold).Footnote ¹³

1. (38)

   1. (A) John was wise to leave the party early.

   2. (B) No, he was not. He had to leave early only because he had a fight and broke his nose.

As mentioned above, the adj-to-inf-cxn and adj-of-np-cxn presuppose, rather than assert, P ₂(a) (Wilkinson Reference Wilkinson1970, Barker Reference Barker2002). This can be confirmed by standard diagnoses involving negation, modal quantification, etc.

1. (39)

   1. (a) John was wise to leave early.

   2. (b) John wasn't wise to leave early.

   3. (c) Perhaps John was wise to leave early.

1. (40)

   1. (a) It was wise of John to leave early.

   2. (b) It wasn't wise of John to leave early.

   3. (c) Perhaps it was wise of John to leave early.

All the sentences in (39) and (40) entail ‘John left early’. The so-adv-cxn, in contrast, asserts P ₂(a) and presupposes ‘From P ₂(a) it can be inferred that P ₁(a)’.

1. (41)

   1. (a) John wisely left early.

   2. (b) Perhaps, John wisely left early.

(41b) does not entail ‘John left early’, but it still conveys a positive evaluation on the act of ‘leaving early’, just as (41a) does. By the same token, if a so-adv-cxn is embedded in the antecedent of a conditional, the entailment ‘From P ₂(a) it can be inferred that P ₁(a)’ survives.

1. (42) If John wisely had left early, he wouldn't have been stabbed.

The assertion and presupposition of a statement can be expressed by a single logical formula, using the connective called preditional (also known as prejunction, transplication; Blau Reference Blau1985, Blamey Reference Blamey, Gabbay and Guenthner1986, Oshima Reference Oshima2006).Footnote ¹⁴

1. (43) preditional

   For any world w and assignment g, [[〈φ; ψ〉]]^w,g is defined iff [[ψ]]^w,g=1. If defined, [[〈φ; ψ〉]]^w,g=[[φ]]^w,g.

In intuitive terms, a formula of the form 〈φ; ψ〉 represents a statement with a presupposition, φ and ψ corresponding to the assertion and the presupposition respectively. When ψ (the presupposition) holds true, 〈φ; ψ〉 is semantically equivalent to φ; when ψ does not hold true, on the other hand, 〈φ; ψ〉 is not evaluated either true or false. By way of illustration, the meaning of John managed to escape, which (roughly) asserts that John escaped and presupposes that it was difficult for John to escape, can be expressed as:

1. (44) 〈escape(john); difficult-for(escape, john)〉

The logical forms of (39a), (40a), and (41a) can now be given as follows:

1. (45)

   1. (a) (39a), (40a) ↦

      〈□_e[leave-early(john)→(transitorily(wise))(john)];

      leave-early(john)〉

   2. (b) (41a) ↦

      〈leave-early(john);

      □_e[leave-early(john)→(transitorily(wise))(john)]〉

3.3 A note on evaluability

A denial of the adj-to-inf-cxn or the adj-of-np-cxn does not appear to mean what it is predicted to mean.

1. (46)

   1. A: {John was wise / it was wise of John} to leave the party early.

   2. B: That's not true. / You're wrong.

Recall that under the proposed analysis, A's utterance can be roughly paraphrased as: ‘John must have been being wise, considering that he left the party early’. B's reply, thus, amounts to ‘It is not the case that John must have been being wise’ (=‘It may be the case that John was not being wise’). This, however, is not the natural interpretation of (46B), as can be confirmed by the observation that it can be naturally followed by (47a), but not by (47b).

1. (47) (as a continuation to (46B))

   1. (a) He was not wise at all. He had to leave early only because he had a fight and broke his nose

   2. (b) ?He (may or) may not have been wise.

This, however, is not a problem specific to (the conditional-based analysis of) the adj-to-inf-cxn and the adj-of-np-cxn. A similar phenomenon is attested with regular, explicit modal statements as well; an evaluation (acceptance, denial, etc.) of a modal statement is often understood as made on the ‘core’ proposition without a modal operator, rather than on the modal statement as a whole.

1. (48)

   1. A: This professor must be smart

   2. B: That's not true. / You're wrong.

   3. (=This professor is not smart.)

   4. (≠ It is not true that this professor must be smart.)

   5. B′: I agree.

   6. (=I agree that this professor is smart.)

   7. (≠ I agree that this professor must be smart.)

   8. (adapted from Papafragou Reference Papafragou2006: 1691)

1. (49)

   1. A: Joe might be in Boston.

   2. B: He can't be in Boston. I just saw him down the hall.

   3. A: Oh, OK.

      1. (i) Then I guess I was wrong.

      2. (ii) #Nonetheless, when I said ‘Joe might be in Boston’, what I said was true.

         (adapted from MacFarlane Reference MacFarlane, Weatherson and Eganforthcoming)

The ‘non-deniability’ of the adj-to-inf-cxn and the adj-of-np-cxn is, thus, fully consistent with the proposed conditional-based analysis. Since these two constructions involve a strong modal operator (necessity operator) in their asserted component, it is only natural that they pattern the same with explicit modal statements like (48A) and (49A) with respect to evaluability.

In the literature on epistemic modality, it has also been pointed out that with appropriate contextual settings, an epistemic modal statement (or at least weak ones with ‘might’, ‘possibly’, etc.) can itself be a target of evaluation.

1. (50) (A and B are meteorologists)

   1. A: It may rain tomorrow.

   2. B: That's not true. / You're wrong.

   3. (=It is not true that it may rain tomorrow.)

   4. B′: I agree.

   5. (=I agree that it may rain tomorrow.)

   6. (adapted from Papafragou Reference Papafragou2006: 1692)

Based on such observations, it has been claimed that two types of epistemic modality need to be distinguished: subjective (not evaluable) and objective (evaluable). Some scholars advocate the view that subjective modality is non-truth-conditional (outside the propositional content) and objective modality is truth-conditional (Lyons Reference Lyons1977, among others). Papafragou (Reference Papafragou2006), to the contrary, argues that the subjective/objective distinction can be reduced to the choice of the relevant conversational background (a set of propositions that determine the epistemic modal base); the conversational background for a subjective epistemic modal statement is the set of propositions known to the speaker at the time of utterance, while for an objective epistemic modal statement it is a set of propositions known to a larger group (e.g., {the speaker, the hearer}, an academic community) and/or known at the contextually specified time that differs from the utterance time. She further demonstrates that the evaluability puzzle illustrated above can be solved without compromising the standard, truth-conditional treatment of epistemic modality à la Kratzer.

An interesting observation regarding the adj-to-inf-cxn and the adj-of-np-cxn is that they appear to single out the subjective interpretation, to the effect that it is impossible to find a context where they (their asserted components) become evaluable. This may simply be because strong modal statements (with ‘must’, etc.) in general favor the subjective interpretation.

1. (51) (A and B are detectives)

2. A: John must have stolen the statue by himself.

3. B: That's not true. / You're wrong.

   1. (i) He had an accomplice.

   2. (ii) ?He may have had an accomplice.

Another possibility is that these constructions are inherently specialized for subjective modality (cf. Papafragou Reference Papafragou2006: 1692). Here I leave the issue open for future discussion.

4.1 Wilkinson's action/event-based analysis

Wilkinson (Reference Wilkinson1976: 164ff.) suggests that a W adjective and an L adjective, when they participate in the three constructions under discussion (as well as some others), are predicated of an action (event) and a proposition, respectively. If this line of analysis is accepted, the sentences in (52) can be paraphrased as ‘John left early, and John's action (act) of leaving early was wise’, and (53) ‘John passed the exam, and this fact (the fact that John passed the exam) was lucky for John’.

1. (52)

   1. (a) Wisely, John left early.

   2. (b) John was wise to leave early.

   3. (c) It was wise of John to leave early.

      (=(1))

1. (53)

   1. (a) Luckily, John passed the exam.

   2. (b) John was lucky to pass the exam.

      (=(6))

W/L adjectives, as a matter of fact, can be predicated of individuals as well as of actions (events)/facts (propositions).

1. (54)

   1. (a) He is wise. / the wise man

   2. (b) John's leaving early was wise. / His act(ion) was wise. / the wise act(ion)

1. (55)

   1. (a) He is lucky. / the lucky man

   2. (b) It was lucky (for me) that it rained. / This fact is lucky. / the lucky fact

Thus we may say that W adjectives are polysemous between a property of individuals and a property of events, and L adjectives between a property of individuals and a property of propositions (sets of worlds).

Although Wilkinson's analysis gives a straightforward solution to the entailment puzzle, it has some weaknesses, too. First, it is not clear how the ambiguity of some W adverbs (mentioned in section 2.1) can be explained. (56), for example, is ambiguous; one reading is roughly equivalent to ‘{It was kind of John/John was kind} to help me out’, and the other ‘John helped me out in a kind manner’.

1. (56) John kindly helped me out.

Suppose John is an employee of an electronic appliance store, whose main job is to help out customers. He is not particulary sympathetic in nature, but he takes pride in doing his job in a professional manner. In this setting, (56) may well be false on the first interpretation, when it is true on the second.

Now, on the second, ‘manner’ interpretation, kindly is clearly predicated of an event. Thus, as long as we adopt the standard ontological assumption that an action is a sort of event (Kamp & Reyle Reference Kamp and Reyle1993: 506, among others), there seems to be no way to represent the difference between the two readings.

Furthermore, the embeddability puzzle remains unexplained under Wilkinson's analysis. That is, given that actions (events), like individuals, are inhabitants of worlds, sentences like (57) and (58) are predicted to be acceptable.

1. (57) #Ken asked (them) that Linda be polite to write letters of appreciation to her supporters.

   (cf.) Ken asked (them) that {Linda's act(ion)/the way Linda acts} be polite.

1. (58) #Ken wanted Bob to be bold to challenge the world record.

   (cf.) Ken wanted {Bob's act(ion)/the way Bob acts} to be bold.

4.2 Barker's metalinguistic analysis

Barker (Reference Barker2002) proposes an innovative analysis of the adj-to-inf-cxn and the adj-of-np-cxn, where the W/L adjectives participating in them are assigned a ‘metalinguistic’ interpretation. First, Barker observes two ‘modes of use’ of vague gradable predicates (which subsume W/L adjectives): descriptive and metalinguistic. When the gradable adjective tall, for example, is used in the normal, descriptive mode, it conveys new information concerning the (actual) world (e.g., the height of a particular individual). When it is used in the metalinguistic mode, on the other hand, it informs the hearer of how to use the word tall appropriately, by providing the contextually relevant standard of tallness. The descriptive use is exemplified by (59), the metalinguistic use by (60).

1. (59)

   1. A: I am going to pick up Dr. Feynman at the airport. What does he look like?

   2. B: Well, Feynman is tall.

1. (60) (Situation: Speakers A and B are at a party; Feynman stands a short distance away.)

   1. A: What do you mean by ‘if you are tall’? What counts as ‘tall’?

   2. B: Well, around here, Feynman is tall.

Building on the dynamic framework in the tradition of Heim (Reference Heim1982, Reference Heim1983) and others, Barker argues that the descriptive use and the metalinguistic use of a vague predicate affect (update) different aspects of the context.

A descriptive use affects the common ground; (59B), for example, adds to the common ground the proposition that Feynman's maximal degree of height exceeds the contextually prevailing standard (say, 180 centimeters). A metalinguistic use, on the other hand, affects the standard of the relevant gradable adjective meaning (e.g., tallness), and thereby resolves or mitigates the vagueness of the adjective in question; (60B), which is uttered in a situation where both the speaker and the hearer can directly observe Feynman's height, does not expand the common ground at all, but only helps the interlocutors determine ‘what counts as tall’ (or, ‘how to use the word tall appropriately’).

To implement the idea that the standards for vague concepts too are a component of the context, the context set (in the Stalnakerian sense) may be defined, among other possibilities,Footnote ¹⁵ as a set of tuples (‘information states’) 〈w, g, d〉 where w is a possible world, g is an assignment function, and d is a delineation – a function that maps gradable predicate meanings to standards (degrees). Suppose, when conversation (59) took place, the prevailing standard of (an adult male's) tallness was 180 centimeters; then, the utterance of speaker B has the effect of excluding from the context set those tuples whose world component w′ is such that [[Feynman is at least 180 cm tall]]^w′=0. In the situation where the conversation (60) took place, in contrast, Feynman's height is already in the common ground – if Feynman was exactly 182 cm tall, then all information states in the context set (before and after the conversation) have a world component w′ such that [[Feynman is 182 cm tall]]^w′=1. What the utterance of speaker B does is to exclude from the context set those tuples whose delineation component is inappropriate. If d′ is such that d′ ([[tall]])=185 cm, then 〈w′, g′, d′〉, 〈w″, g′, d′〉, 〈w′, g″, d′〉, etc. will be excluded (if a man who is 182 cm tall counts as tall, the standard of tallness cannot be as high as 185 cm); if d″ is such that d″ ([[tall]])=175 cm, then 〈w′, g′, d″〉, 〈w″, g′, d″〉, 〈w′, g″, d″〉, etc. will survive (by the same reasoning, the standard of tallness may be as low as 175 cm).

Regarding the adj-to-inf-cxn and the adj-of-np-cxn, Barker claims that in these constructions W/L adjectives are predicated of a proposition, and interpreted only in the metalinguistic mode of use.Footnote ¹⁶ The sentence Feynman is stupid to dance, for example, has no regular, descriptive entailment (putting aside the presupposition that Feynman danced), and merely narrows down the range of possible delineations, so that only those delineations according to which Feynman's dancing (the proposition that Feynman danced) counts as stupid will survive.

Barker's analysis of the adj-to-inf-cxn/adj-of-np-cxn is similar to the conditional-based analysis proposed in the previous section, building on the idea that (some aspect of) their meaning is not about the way things are in the actual world. I find problematic, however, its basic assumption that all W/L adjectives can be semantically predicated of a proposition. While all L adjectives and some W adjectives can take a clausal complement, there are some W adjectives that cannot.

1. (61)

   1. (a) It was lucky (for me) that John left early.

   2. (b) It was {wise/stupid} that John left early.

   3. (c) ?*It was {careful/careless} that John left early.

Furthermore, many W adjectives that can take a clausal complement require that the subject of the clausal complement be sentient, or at least that the event described by the clausal complement be a volitional action. This suggests that at the semantic level, these W adjectives may be predicated of an individual or a (certain kind of) event, but not of a proposition.

1. (62)

   1. (a) ??It was {rude/brave/kind} that the portrait of the secretary-general was removed.

   2. (b) *It was {rude/brave/kind} that it rained.

The observation that W adjectives cannot modify a propositional noun like fact lends further support to this point.

1. (63) *a wise fact, *a careful fact, *a rude fact, *a brave fact, *a kind fact

Another problem with Barker's analysis is the fact that the W or L adjective participating in an adj-to-inf-cxn or adj-of-np-cxn may take the form of a comparative, as in:

1. (64)

   1. (a) John was smarter than Fred to bring a raincoat.

   2. (b) John was luckier than Fred to survive.

As discussed by Barker himself, the truth of a comparative statement can be determined independently from the contextually established standard for the relevant adjective, and thus comparatives cannot be interpreted in the metalinguistic mode. Therefore, the fact that a comparative may fill in the P ₁ slot of the adj-to-inf-cxn clashes with Barker's claim that W/L adjectives in these constructions are used in the metalinguistic mode only.

Under the conditional-based analysis proposed above, adj-of-np-cxns with comparatives would be assigned semantic representations along the lines of (65).

1. (65)

   1. (a) 〈□_e[bring-a-raincoat(john)→(transitorily(smarter))(john, fred)]; bring-a-raincoat(john)〉

      (in prose: John brought a raincoat, and from the fact that John brought a raincoat, it can be inferred that John was being smarter than Fred.)

   2. (b) 〈□_e[survive(john)→(transitorily(luckier))(john, fred)]; survive (john)〉

      (in prose: John survived, and from the fact that John survived, it can be inferred that John was being luckier than Fred.)

Are such semantic representations adequate? One may be skeptical, as (64a) and (64b) appear to entail that Fred did not bring a raincoat and that Fred did not survive, respectively, while neither of these follows from the logical forms in (65). A case can be made, however, that such entailments are only illusory. If the interlocutors are aware that John brought a raincoat and the speaker claims that it is possible to infer from this that John was being smarter than Fred, the hearer would naturally infer that Fred did not bring a raincoat, as otherwise the presented inference has no reasonable ground. To determine whether the messages in question are genuine entailments or mere implicatures, let us suppose the following.

1. (66) John visited Raincity, where it almost always rains. He, smartly, brought a raincoat. On the same day, Fred visited Suncity, where it is almost always hot and sunny. He meant to bring a straw hat, but mistakenly brought a raincoat.

Would (64a) be judged true under this scenario, where both John and Fred brought a raincoat? The judgment is somewhat subtle, but I believe that the answer is yes, which implies that Fred's not bringing a raincoat is not semantically entailed by (64a).

5.1 The subject-oriented adverb construction

In section 3, I proposed that the meaning of a SO-ADV-CXN can be represented by the following scheme, where a, P ₁, and P ₂ are logical translations of the subject, the (subject-oriented) adverb, and the VP, respectively.

1. (73) 〈P ₂(a); □_e[(P ₂(a)→(transitorily(P ₁))(a)]〉

Below is a possible formulation of the so-adv-cxn, a construction that licenses a SO-ADV-CXN.Footnote ¹⁸

1. (74) The subject-oriented adverb construction

   

The type wl-sem-obj is the immediate supertype of the two defined below (here WE_a=well-formed expression of type a).

1. (75)

   1. (a) w-sem-obj:

      

   2. (b) l-sem-obj:

      

5.2 The ‘Adj+to Inf’ construction

The meaning of the ADJ-TO-INF-CXN can be represented in the following form, which is equivalent to (73) except that the assertion and the presupposition are switched.

1. (76) 〈□_e[(P ₂(a)→(transitorily(P ₁))(a)]; P ₂(a)〉

There is more than one conceivable constituent structure for the ADJ-TO-INF-CXN, and the choice largely depends on the general architecture and philosophy of the adopted syntactic framework. Here, I postulate the following (=(14)).

1. (77) [_S [_NP John][_VP was [_AP wise [_VP to leave early]]]]

The most natural place to introduce the ‘construction-specific’ meaning is the AP, where, anomalously, the adjective and the to-infinitive form a constituent despite the fact that neither of them selects for the other.

1. (78) The ‘Adj+to Inf’ construction (AP; e.g., wise to leave early)

   

Note that what is licensed by (78) is a construct whose mother sign is an adjectival phrase, rather than a clause. The+value of the syn|cat|pred attribute indicates that the sign is predicative and thus can appear after be, as a complement of consider, etc. (e.g., I consider John wise to have left early); the+value of the syn|cat|inf attribute indicates that the sign is a to-infinitive, rather than a bare infinitive (Sag et al. Reference Sag, Wasow and Bender2003: 334ff., 361ff.).

The rest of the derivation is straightforward. First, the AP (the mother sign of the adj-to-inf construct) and (an appropriate form of) the auxiliary verb be, which is instantiated from the lexeme presented in (79), form a sign (VP; e.g., was wise to leave early) licensed by the head-complement construction presented in (80). The sem|lf value of the mother sign is determined by sequentially applying the sem|lf value of the head daughter to those of the complements, from left to right. Then, the VP and the subject form a sign (sentence; e.g., John was wise to leave early) licensed by the subject-predicate construction presented in (81).

1. (79)

   

1. (80)

   

1. (81)

   

5.3 The ‘Adj+of NP’ construction

The meaning of the ADJ-OF-NP-CXN is truth-conditionally equivalent to that of the ADJ-TO-INF-CXN. As mentioned earlier, it seems sensible to assume that this construction involves extraposition and is related in some way to the structure given below.

1. (82) [_S [_VP To leave early][_VP was [_AP wise [_PP of John]]]]

To assign (82) the appropriate meaning, the only natural place to introduce the construction-specific meaning is the AP node.

1. (83) The ‘Adj+of NP’ construction (AP; e.g., wise of John)

   

The construction in (83) requires that the sem value of the first daughter be w-sem-obj, rather than wl-sem-obj, to the effect that it does not license a phrase like lucky of John. To derive a sentence like (82) (with the appropriate meaning), it suffices (i) first to combine the AP (the mother sign of (83)) with the auxiliary be by the head-complement construction, and (ii) then to combine the resulting mother sign (VP) with the to-infinitive by the subject-predicate construction.Footnote ¹⁹

The derivation of a sentence like (84), however, is less straightforward.

1. (84) It was wise of John to leave early.

A lexical rule/construction, along the lines of those presented in Pollard & Sag (Reference Pollard and Sag1994: 145ff.) and Sag (Reference Sag2007a: 41), cannot be used, as a phrase (e.g., (was) wise of John) cannot feed into it.

A possible solution is to treat extraposition as a phrasal, rather than a lexical, process, using an analog of the filler-gap construction (Sag et al. Reference Sag, Wasow and Bender2003: 484, Sag Reference Sag2008). Specifically, I suggest postulating the following construction, which may be used as an alternative to the regular subject-predicate construction to form an expression like it was wise of John:

1. (85)

   

The extra attribute plays a role similar to that of the gap attribute, but its value is discharged to the right, rather than to the left, as specified by the following head-extraposition construction (Kim & Sag Reference Kim, Sag and Müller2005: 202).Footnote ²⁰

1. (86)

   

Under this option, the internal structure of the ADJ-OF-NP-CXN will be like (87) (rather than (19)).

1. (87) [[_S [_NP It][_VP was [_AP wise [_PP of John]]]] [_VP to leave early]]

The suggested solution may not be ideal in terms of theoretical parsimony, in that it necessitates separate constructions for subject and object it-extraposition (cf. Kim & Sag Reference Kim, Sag and Müller2005). To deal with expletive it in an object position (e.g., (I) find it difficult to read that book), the suggested analysis of extraposition will need to be supplemented with a separate construction, such as the following.

1. (88)

   

The mother sign of this construction would then feed into (86), thus generating a predicate phrase like find it difficult to read that book.

A case can be made, on the other hand, that a phrasal analysis (e.g., with constructions like (85) and (88)) may lead to a more uniform treatment of varieties of extraposition, considering that cases of adjunct extraposition like (89) appear to call for phrasal analyses anyway.

1. (89) A man came into the room {a. that no one knew / b. with blond hair}. (Kim & Sag Reference Kim, Sag and Müller2005: ex. (42))

In sum, the proposed phrasal constructional treatment of the ADJ-OF-NP-CXN demands a phrasal constructional treatment of extraposition, which contradicts some previous studies but nevertheless appears to be viable. The discussion in the current section contributes to the ongoing debate regarding the strength/weakness of the phrasal constructional approach in comparison to the lexical approach (Müller Reference Müller2006, Reference Müller and Müller2007, among others).