1. INTERROGATIVES AND FREE RELATIVES

We can get an intuitive grip on the distinction between interrogative and free relative wh-phrases by considering the different meanings that wh-phrases can have.Footnote ⁴ Consider the following sentences:

1. (1) I asked what was for dinner.

2. (2) I ate what I was given.

We can give an account of the meaning of sentence (1) by reading the wh-phrase as having an interrogative meaning, denoting a question.Footnote ⁵ Sentence (1) is true if I utter the sentence ‘what's for dinner?’ (or some sentence with a similar meaning) thereby asking the question what is for dinner?. However, it would be wrong to treat the wh-phrase in sentence (2) in the same manner. Sentence (2) doesn't claim that I ate the question what was I given?. Questions aren't the kinds of things we can eat. Hence, in (2) the wh-phrase ‘what I was given’ must denote something else, plausibly some stuff that I was given. Linguists standardly explain this difference in meaning by claiming that in sentence (1) the wh-phrase is functioning as an interrogative phrase, denoting a question, whereas in (2) the wh-phrase is functioning as a free relative noun phrase, denoting an object.

We can often work out how a wh-complement is functioning by considering the meaning of the embedding verb. Some verbs can relate only to questions, requiring an interrogative reading of their wh-complements, whereas other verbs can relate only to non-propositional objects, requiring a free relative reading. For example, ‘ask’, ‘inquire’ and ‘wonder’ can only relate to questions, whereas ‘eat’, ‘take’ and ‘give’ can only relate to objects.

When a verb doesn't make such a restriction in virtue of its meaning we find sentences that are ambiguous between the free relative and embedded question readings, which are helpful for seeing the distinction at issue.Footnote ⁶ Consider the following situation: Tariq and Mona are spies. Their job is to keep track of rival spies who are after their country's nuclear codes, and to inform their superiors whenever the rival spies attempt to communicate with their handlers. Tariq and Mona do not themselves know the nuclear codes. They are tracking two spies: Jorge and Petra. It is common knowledge between them that Petra has a belief about what the codes are, although they don't know the content of that belief, or whether it is true. Tariq utters the following sentence:

1. (3) Jorge knows what Petra believes.

There are two interpretations of this sentence, which have different presuppositions, and make different commitments as to the relation of Jorge and Petra.Footnote ⁷

If we read ‘what Petra believes’ as a free relative, it denotes the thing which Petra believes – that the nuclear code is XYZ. On this reading (3) says that Jorge knows the thing that Petra merely believes: that XYZ is the nuclear code. On this reading, this sentence presupposes that Petra's belief is true, since Petra believes the proposition that Jorge knows, and knowledge is factive. However, this reading says nothing about Jorge's relation to Petra: they might have the same beliefs without ever having heard of one another.

By contrast, if we read ‘what Petra believes’ as an interrogative, it denotes a question, meaning that (3) says that Jorge knows the answer to the question of what Petra believes. This question might either be answered by a proposition like Petra has some belief about what the nuclear code is, or by a proposition like Petra believes that the Nuclear codes are XYZ. On this reading, (3) does not presuppose that Petra has a true belief, because it might be that Jorge only knows that Petra has some belief or other about the codes, without having any knowledge about what the code is. However, this sentence does require that Jorge is aware of Petra, since he cannot know that she believes such-and-such without having some idea about who she is.Footnote ⁸

2. FREE RELATIVE SEMANTICS FOR ‘S KNOWS HOW TO V’

Having got clear on the difference between interrogatives and free relatives, let's connect this distinction to the debate about the nature of knowledge-how. Applying the interrogative and free relative treatments to sentences of the form ‘S knows how to V’ gives us the following general truth conditions (with brackets used to pick out the object of the knowledge relation):

Removing the quotation marks in these semantic treatments gives us two rather different accounts of the nature of knowledge-how.Footnote ⁹ Disquotation on KH-INT yields the claim that knowledge-how is knowledge of a certain kind of proposition about the nature of knowledge-how. By contrast, disquotation on KH-FR yields the claim that knowledge-how is a kind of objectual knowledge of a relevant way of acting. Whereas KH-INT is a natural partner for Propositionalism; KH-FR is a natural partner for an Objectualist account of knowledge-how, as Bach (Reference Bach2012) and Abbott (Reference Abbott2013) observe.

The best-developed version of Objectualism is Bengson and Moffett (Reference Bengson and Moffett2011b) (but see also Michaelson Reference MichaelsonMs). On B&M's view, knowing how to do something is objectual knowledge of a way of engaging in that activity, together with a distinctive kind of action-guiding conception of that way of acting. B&M have an ambivalent relation with the Free Relative semantics. A good deal of the evidence for their view comes from linguistic arguments which can easily be adjusted to support the free relative semantics, as we shall see in the next section. However, B&M are cagey about the connection, claiming that:

In the remainder of this section, I want to address B&M's worry by offering a picture of the relation between linguistic and metaphysical claims. This is controversial territory, so I will proceed carefully. I will first offer an argument for a minimal connection between the semantics for ‘knows how’ ascriptions and the correct account of knowledge-how, which I call the semantic implementability constraint. I will then distinguish this constraint from some of the other ways which Propositionalists have employed linguistic evidence, and show that this constraint doesn't face some of the worries which face these ways of using linguistic evidence. With this methodological background in place, I will answer B&M's worry, pointing out that endorsing the free relative semantics allows the Objectualist to meet the semantic implementability constraint.

I take the semantic implementability constraint to follow from the idea that the subject-matter of the debate about knowledge-how is the folk notion of knowledge-how. The line of thought goes like this. We use sentences of various natural languages to pick out states of knowledge-how. It is reasonable to assume that at least some (although certainly not all) of these sentences are true. Let's assume the English sentence ‘Laura Trott knows how to cycle’ is true.Footnote ¹⁰ If this sentence is true and picks out Trott's state of know-how, then the correct account of the metaphysics of Trott's state of know-how will be compatible with the correct account of the truth-conditions of this sentence. This is just an application of the truth schema – ‘S’ is true iff S – in the right to left direction, pointing out that the correct account of the nature of a particular piece of knowledge-how will be compatible with the correct account of the semantics of the ordinary language sentences used to pick out that state. In general terms, the semantic implementability constraint is just the idea that the correct account of knowledge-how needs to be able to give a compatible semantics for ‘knows how’ ascriptions.

The implementability constraint should be distinguished from some of the more controversial ways in which linguistic evidence has been used in the knowledge-how debate.

Stanley and Williamson (Reference Stanley and Williamson2001), and Stanley (Reference Stanley2011a, Reference Stanley2011b) employ what one might think of as a linguistics-first methodology (see Devitt Reference Devitt2011). Whereas the semantic implementability constraint reads the truth schema from right to left, requiring that a metaphysical account be able to give a compatible semantics, the linguistics-first methodology reads the truth schema from left to right, starting with the best linguistic account of the semantics of ‘knows how’ ascriptions, and then reading off the compatible metaphysics. Although Stanley presents this methodology as requiring nothing more than the commitment to the truth schema (2011b: 143–9), in order to read the truth-schema in the left to right direction one needs to think that linguistic evidence somehow has methodological priority over other kinds of philosophical evidence (Brown Reference Brown2013). To put the point glibly: Stanley might be right that linguists are doing metaphysics in the formal mode (Reference Stanley2011b: 144), but the important question is whether they are doing good metaphysics. The semantic implementability constraint makes no commitment to the priority of linguistic evidence, since it involves reading the truth schema from right to left, meaning that it leaves open the possibility of motivating a novel semantics for ‘knows how’ ascriptions on non-linguistic grounds.

Another way to appeal to linguistic evidence appeals to the linguistic uniformity between ‘knows how’ ascriptions and other ‘knows wh’ ascriptions (see Snowdon Reference Snowdon2004: 6–8; Bengson and Moffett Reference Bengson and Moffett2011b: 178–80; Stanley Reference Stanley2011a: 208). Stanley sums up this argument nicely:

The idea here is that when doing linguistic analysis, we ought to expect constructions which have similar syntactic and semantic behaviour to be generated from a uniform underlying metaphysical structure. Importantly, this kind of argument is at best a ceteris paribus consideration to be fed into our theorising alongside other kinds of considerations. There are many cases in which prima facie syntactically uniform constructions pick out extremely different kinds of entities, as Michaelis observes (Reference Michaelis, Bengson and Moffett2011: 261–2). One reason for especially being suspicious about appeals to linguistic similarity is that there seem to be philosophical dissimilarities between knowledge-how and knowledge-wh. Knowledge-how seems to have rather different epistemic properties to other kinds of knowledge-wh (see the discussion of justification in §3). Notice that non-linguistic dissimilarities do not cause problems for the semantic implementability constraint: if knowledge-how turns out to be a very different kind of state to other kinds of knowledge-wh, we should still want a semantics for ordinary knowledge-how ascriptions.Footnote ¹¹

What are the upshots of the semantic implementability constraint for Objectualism? In order to tell whether an account of the nature of knowledge-how is semantically implementable, a natural approach is to collect the semantics for ‘knows how’ ascriptions proposed by linguists and linguistically informed philosophers, and consider whether any of those semantics is compatible with the metaphysical account. If it turns out that an account of knowledge-how is not compatible with any off-the-shelf semantics, then this is a strike against that account, because this provides evidence that our ordinary ‘knows how’ ascriptions are not picking out the state posited by that account, leaving the proponent of that account open to the charge that they are changing the subject away from our ordinary notion of knowledge-how (Bengson and Moffett Reference Bengson and Moffett2011a: 36–7). However, there are a number of ways in which the supporter of such an account might respond to this challenge. One option would be to offer a novel account of the semantics our ordinary ‘knows how’ ascriptions that is amenable to their metaphysics. An alternative is to opt for an error theory of our ordinary ascriptions. If our ordinary knowledge-how ascriptions are simply false, then the charge of changing the subject loses its sting.

With this background in place, it is clear that the supporter of Objectualism should be in the game of finding a linguistically plausible semantics for ‘knows how’ ascriptions in order to implement their account. How might they do this? Obviously, S&W's preferred interrogative semantics is not compatible with an Objectualist account of knowledge-how. B&M point out that there are a number of alternatives to the interrogative semantics, citing Roberts (Reference Roberts, Hinrichs and Nerbonne2009), Ginzburg (Reference Ginzburg, Bengson and Moffett2011) and Michaelis (Reference Michaelis, Bengson and Moffett2011) as examples.Footnote ¹² However, none of these treatments are friendly to Objectualism. Roberts (Reference Roberts, Hinrichs and Nerbonne2009) builds on work by Dowty and Jacobsen to argue that infinitival how-to phrases are property-denoting verb phrases built on partition semantics, meaning that her account predicts that knowledge-how is a relation to a certain kind of goal-related property. Although Ginzburg (Reference Ginzburg, Bengson and Moffett2011) suggests that some non-English infinitival knowledge ascriptions are non-propositional, he treats ‘S knows how to V’ as involving a relation to a fact. Michaelis (Reference Michaelis, Bengson and Moffett2011) treats wh-complements as presupposing an open proposition (of the form: x is F),Footnote ¹³ and asserting that the agent knows the unbound variable in that open proposition. Michaelis’ view can sound close to Objectualism, but she explicitly claims that her account is truth-conditionally equivalent to Stanley's account (Michaelis Reference Michaelis, Bengson and Moffett2011: 277), meaning that her account should not be construed as an alternative to the interrogative semantics.Footnote ¹⁴

So from the currently available semantics for knowledge-how ascriptions, it seems that the free relative semantics is the Objectualist's best bet for directly meeting the semantic implementability constraint. As we've just noted, if the free relative semantics is not linguistically plausible, this is not a reason to outright reject Objectualism, but it does leave them either taking on the burden of developing a novel semantics, or endorsing an error theory.

3. EVIDENCE THAT ‘HOW TO V’ IS A FREE RELATIVE

B&M appeal to four linguistic phenomena to build an argument for thinking that the ‘how to’ complement in knows how ascriptions denotes an object rather than a proposition (Bengson and Moffett Reference Bengson and Moffett2011b: 178–85):

1. (i) The availability of apparent objectual paraphrases for ‘knows how’ ascriptions,

2. (ii) The fact that ‘knows how’ ascriptions fail to take propositional modifiers,

3. (iii) The oddness of raising the question of justification about knowledge-how,

4. (iv) The gradability of ‘knows how’ ascriptions.

Although B&M present these phenomena as evidence for Objectualism, rather for a free relative semantics, their explanations of these phenomena are also available to the supporter of the free relative semantics. However, I don't think that these phenomena provide a compelling argument for the free relative semantics, since all of them can also be explained by the supporter of an interrogative semantics. This is bad news for both Objectualism and free relative semantics since it removes a central plank in the case for both views.

B&M's first piece of linguistic evidence for thinking that ‘how to V’ denotes an object comes from the paraphrases for ‘knows-how’ ascriptions. They claim that the natural paraphrase for (4) is not S&W's (4a), but rather (4b):

1. (4) Ana knows how to swim.

2. (4a) Ana knows [that w is a way in which she herself can swim].

3. (4b) Ana knows [the way to swim].

They point out that the naturalness of (4b) is good news for the objectualist, since in this paraphrase the object of knowledge seems to be an object-denoting noun phrase – the way to swim – rather than a proposition-denoting that-clause. This paraphrase is also extremely close to the free relative reading of ‘how to swim’, suggesting that the natural paraphrase is free relative friendly.

Although this paraphrase is suggestive, a supporter of the interrogative semantics can easily explain it. ‘Knowledge-the’ ascriptions like (4b) plausibly have a concealed question reading which is equivalent to an interrogative knowledge-wh ascription (White Reference White1982: 31–2; Brogaard Reference Brogaard2008; Bach Reference Bach2012). Consider:

1. (5) Jane knows the capital of Mali.

This sentence has two readings: an objectual reading which says that Jane is acquainted with Bamako, and a concealed question or quasi-interrogative reading which says that Jane knows what the Capital of Mali is but makes no commitments about acquaintance. Given the concealed question reading of ‘knows-the’ ascriptions it is unsurprising that interrogative knowledge-wh can be paraphrased by ‘knowledge-the’ ascriptions, as Pavese points out (2013, footnote 15). Taking a concealed question approach to (4b) the sentence comes out as meaning ‘Ana knows what a way to swim is’, which is not only an interrogative, but also pretty close to S&W's proposed paraphrase (4a).Footnote ^15, Footnote ¹⁶

B&M's second piece of evidence comes from the observation that the complement in ‘knows how’ ascriptions fails to take modifiers that are appropriate for that-clauses. For example:

1. (6) Ana knows that the Prime Minister is a woman – so it must be true!

sounds fine, but:

1. (7) ?Ana knows how to swim – so it must be true!

Seems grammatically odd.Footnote ¹⁷ Following B&M, we might think that sentence (7) is odd because the how-complement denotes an object, and objects cannot be true. The supporter of the free relative semantics can appeal to the same explanation, since on their view ‘how to swim’ is an object-denoting noun-phrase.

B&M also note that ‘knowledge-how’ ascriptions seem not to be ‘bumped up’ to certainty – as knowledge-that ascriptions are – but rather seem to be bumped up to mastery:

1. (8) ? Ana knows how to swim – in fact, she's certain of it!

2. (9) Ana knows how to swim—in fact, she's mastered it!

There is a parallel argument in the offing here. The supporter of a free relative semantics can point out that one cannot be certain of ways, although one can master them.

Although these arguments initially seem appealing, they end up significantly over-generating. It is easy to construct examples in which other kinds of ‘knows-wh’ ascriptions fail to take these modifiers. Consider:

1. (10) ? Ruth knows whether the Prime minister is a woman – so it must be true!

2. (11) ? Lani knows who came to the party — so it must be true.

Sentences (10) and (11) sound just as weird as (7), which suggests that if one wants to take this evidence seriously, it supports the claim that a whole swathe of ‘knows wh’ ascriptions have a free relative reading, which is a contentious position. B&M do float the idea that all knows-wh involves objectual knowledge (see footnote 11), so we might think that they would be happy to endorse a general free relative semantics. However, even they ought to baulk at the suggestion that ‘knows-whether’ ascriptions involve a free relative complement. It's difficult to even make sense of the idea that a ‘whether’ clause denotes an object.Footnote ¹⁸

Furthermore, the supporter of an interrogative semantics can appeal to a fairly simple explanation for the oddness of these sentences. According to the interrogative semantics, a wh-phrase denotes not a proposition, but a question. And on the face of it, questions aren't the kinds of things which can be true or false.Footnote ¹⁹ So, we might think that (7), (10) and (11) are odd because of anaphora failure. The modifying phrase ‘so it must be true’ involves the anaphoric ‘it’, which looks back in the sentence to find something which might be true. Since the interrogative phrase denotes a question, one might think that the anaphoric phrase is odd because question are the only available reference for the anaphoric ‘it’, and questions cannot be true or false. The general lesson is that although the Propositionalist is committed to thinking that ‘knowledge-wh’ ascriptions are made true by states of propositional knowledge, they don't need to think that ‘knows-that’ and ‘knows-wh’ ascriptions have precisely the same semantic or syntactic properties.Footnote ²⁰

The supporter of an interrogative semantics can also explain the appropriateness of (9). Presumably one can master activities as well as ways of acting. And, on the standard interrogative semantics, (9) relates Ana to a question about the activity of swimming. So on the standard interrogative semantics for (9) the first part of the sentence includes a term picking out an activity which can be the subject of anaphoric reference in phrases like ‘in fact, she's mastered it’.

Examples like (8) involving ‘in fact, she's certain of it’ are a little trickier to deal with. Intuitively, one can be certain of questions – consider ‘Paul was certain of who came to the party’ – so the interrogative semantics predicts that these sentences ought to be acceptable. Interestingly, the parallel examples for ‘knows-whether’ are a little strange but seem grammatically acceptable. For example:

1. (12) Ruth knows whether the Prime Minister is a woman – in fact she's certain of it!

sounds fine. This sentence says that Ruth doesn't merely know, but is certain of whether the PM is a woman. Pretty much all of the examples here are controversial, but I think there are some ‘knows how’ sentences involving certainty which seem grammatically acceptable. For example:

1. (13) Ana knows how to pronounce ‘phở’ – in fact she's certain of it!

Sounds grammatically acceptable to me: it says that Ana doesn't merely know, but is certain of how to pronounce ‘phở’.Footnote ²¹ There's certainly room for disagreement about the acceptability of this sentence, and I can easily imagine supporters of the free relative semantic contending sticking to their guns, and claiming that (13) is unacceptable. However, if there is disagreement about our intuitions, this is bad news for the supporter of free relative semantics. If there are conflicting or fuzzy intuitions about the acceptability of a class of sentences, then neither side of the debate ought to be relying on the acceptability (or unacceptability) of these sentences as motivation for their view, which means that examples like (8), (12) and (13) are not admissible evidence for either side.

B&M's third piece of evidence is the fact that ‘knows-how’ ascriptions do not open up the question of justification, unlike ‘knows-that’ ascriptions (Austin Reference Austin1956). They point to the following exchange:

1. (14)

   1. (a) Martin knows how to get to the airport,

   2. (b) Hmmm … is he really justified in believing that?

The response in (14b) is certainly odd, and one might think that this is because knowledge-how is not the kind of knowledge which involves justification.

However, the supporter of an interrogative semantics can explain the oddness of this exchange by appealing to the same kind of anaphora failure which occurs in sentence (7), (8), (10) and (11). The response in (14b) involves an anaphoric ‘that’ which looks for a proposition in (14a) about which one can raise the question of justification. However, if the interrogative semantics is correct, then (14a) ascribes knowledge of a question, meaning that there are no propositions available for the anaphoric ‘that’ to pick up on. Consider a parallel example with ‘knows-whether’:

1. (15)

   1. (a) Martin knows whether to swim

   2. (b) Hmmm … is he really justified in believing that?

The response in (15b) is just as weird as (14b), suggesting that the oddness stems from problems with anaphoric reference to interrogative phrases, rather than any special features of ‘how to V’ phrases.

Putting the oddness of (14b) to one side, one might think that there remains a powerful non-linguistic argument for Objectualism in the offing here. It seems that there are many cases in which agents know how without having a related justified belief (Cath Reference Cath, Bengson and Moffett2011; Glick Reference Glick2011: 408–9; Weatherson Reference Weatherson2016: 12–13). Since most theories of the propositional knowledge relation claim that it requires justification, one might take the observation that we seem to have here a case of knowledge-how without justification as evidence that the relation involved in knowledge-how is something other than propositional knowledge. In particular, one might take this as an argument for thinking that knowledge-how is non-justification entailing objectual knowledge. This is an instance of a wider class of arguments which we might call divergence arguments, which are most familiar from the debate about the Gettierisability of knowledge-how (Poston Reference Poston2009; Cath Reference Cath, Bengson and Moffett2011, Reference Cath2015; Carter and Pritchard Reference Carter and Pritchard2015). These arguments contend that knowledge-how cannot be a kind of knowledge-that because the two kinds of knowledge have different epistemic properties.

A successful divergence argument establishes that knowledge-how involves a relation with different epistemic properties to standard examples of knowledge-that. However this result doesn't establish that the object of knowledge-how is anything other than a proposition. Supposing that knowledge-how does not entail justification, it might be the case that knowledge-how is a special kind of non-justification-entailing propositional knowledge, or that knowledge-how is a non-knowledge constituting relation to a proposition.Footnote ²² Although positing that knowledge-how is a kind of non-justification entailing propositional knowledge seems rather ad hoc, there are various accounts of propositional knowledge on the market which claim that propositional knowledge does not require justification (Goldman Reference Goldman1967; Kornblith Reference Kornblith and Smith2008).Footnote ²³ So, even if the divergence argument for justification goes through, it does not establish that the object of knowledge-how is anything other than a proposition.

B&M's final piece of evidence comes from the apparent gradability of ‘knows-how’ ascriptions. We can say one person knows how to do something better than someone else, or that they know in part how to do something. These qualitative and quantitative modifiers don't apply to ‘knows that’ ascriptions, suggesting that know-how, but not know-that, can come in degrees (Ryle Reference Ryle2009: 46). A free relative semantics is in a nice place to explain these data, since objectual knowledge ascriptions also permit degree modifiers – one can have partial knowledge of Paris, or know Paris better than someone else. These modifiers can be neatly explained by pointing out that one can be more or less acquainted with an object, suggesting that these modifiers grade the knowledge-how relation.

However, a supporter of Interrogative semantics can also provide plausible explanations for the gradability of knowledge-how ascriptions, positing that these modifiers attach to the object of knowledge-how. For example, Stanley treats comparative modifiers like better than as grading the quality of the ways known (Reference Stanley2011b: 31–5) and Pavese argues that degree modifiers like ‘in part’ mark a partial answer to the embedded question (Pavese Reference Pavese2013, Reference Pavese2017) (see also Roberts Reference Roberts, Hinrichs and Nerbonne2009). The upshot is that both Objectualism and Propositionalism can explain the gradability of ‘knows how’ ascriptions, although Objectualism posits a modification of the knowledge relation, and Propositionalism a modification of the objects of knowledge.

To sum up, the four linguistic phenomena that B&M appeal to in support of Objectualism are suggestive of a free relative semantics, although closer attention shows us that a supporter of interrogative semantics can also explain these features, meaning that the ability to explain these linguistic phenomena is not a reason to prefer a free relative semantics over an interrogative semantics.

4. TESTS FOR DISTINGUISHING FREE RELATIVES FROM INTERROGATIVES

There are a number of linguistic tests which can be used to determine whether a ‘knows wh’ ascription has an interrogative or free relative reading (Schaffer Reference Schaffer2009: 486–91). These tests pose a serious problem for the free relative semantics for ‘know-how’, since ‘S knows how to V’ systematically tests positive for an interrogative reading and negative for a free relative reading.

There are five tests for detecting an interrogative reading of a wh-phrase:

1. (i) Wh-the-hell: If the wh-phrase can be extended to an exclamation like wh-the-hell then it has an interrogative reading (Zwicky and Sadock Reference Zwicky, Sadock and Kimball1975).

2. (ii) Co-ordination: If the wh-phrase can be embedded within a verb which only accepts interrogative complements – like ‘wonder’, ‘ask’ or ‘inquire’ – then it has an interrogative reading (Bresnan and Grimshaw Reference Bresnan and Grimshaw1978: 332; Baker Reference Baker1995: 204–7).

3. (iii) Multiple questions: If the wh-phrase can be extended to include multiple question-words, then it has an interrogative reading (Baker Reference Baker1968; Bresnan and Grimshaw Reference Bresnan and Grimshaw1978: 335).

4. (iv) Paraphrase: If the wh-phrase involves, or can be paraphrased by, a question-word that cannot take a free relative reading, then it has an interrogative reading (Baker Reference Baker1995: 217).

5. (v) Infinitive: If the wh-phrase is infinitival, then it is an interrogative, and not a free relative (Baker Reference Baker1995: 216–18; Huddleston and Pullum Reference Huddleston and Pullum2002: 1070–3).

A couple of points about these tests. First, the tests merely detect a reading of a wh-phrase, so testing positive for one reading doesn't establish that the other reading is not available. Some wh-phrases are ambiguous between interrogative and free relative readings (as we saw with sentence (3)). The exception to this is test (v), which does tell against a free relative reading. Second, passing a test is a sufficient condition for a reading, but not a necessary condition. Failing one test is not a guarantee of the absence of a particular reading: for example finite wh-phrases do not pass (v), but will pass all of the other tests for an interrogative reading. However, if a wh-phrase fails all of the tests for a reading, I take this to be good evidence that that reading is not present. Third, these tests detect readings of a given wh-phrase independently of the embedding verb. There may be sentences in which a wh-phrase can have two readings, but the embedding verb forces one of those readings (such as sentence (2)).

A typical know-how ascription – ‘Ruth knows how to get to Larissa’ – gives clearly positive results on tests (i), (ii) and (v). Tests (iii) and (iv) do not give clear positives, but don't cause serious problems for the interrogative semantics.

The wh-the-hell test (i) suggests an interrogative reading, since:

1. (16) I don't know how the hell to get to Larissa

is an acceptable sentence.

The co-ordination test (ii) also suggests an interrogative reading, since ‘how to get to Larissa’ can be moved into verbs that can only accept interrogative complements. Consider:

1. (17) Ruth wondered how to get to Larissa.

2. (18) Ruth asked how to get to Larissa.

3. (19) Ruth inquired how to get to Larissa.

All of these sentences are completely acceptable.

Whereas the patterning of the data on tests (i) and (ii) is robust, with all ‘knows how’ ascriptions passing these tests, the multiple question test (iii) is less decisive. We can find some examples of infinitival how ascriptions with multiple question words. For example:

1. (20) Ruth knows how to get to where.

2. (21) Ruth knows how to get to whom.

However such examples are few and far between, and seem to be of doubtful acceptability. I don't think that the rarity of multiple questions ought to be too much of a worry for an interrogative treatment. For one thing, these tests provide merely sufficient conditions for an interrogative reading. Furthermore, it is just as difficult to find multiple question examples for ‘who to’, ‘where to’ and ‘whether to’, suggesting that the underlying pattern is that it is difficult to construct multiple wh-phrase for sentences involving infinitival wh-phrases.

The paraphrase test (iv) is based on the observation that different question-words seem to be more or less favourable to the free relative reading, with ‘whether’ never taking a free relative reading.Footnote ²⁴ We might try to paraphrase ‘Ruth knows how to get to Larissa’ with a ‘knows-whether’ ascription. Consider:

1. (22) Ruth knows whether taking the road north is the way to get to Larissa.

However, this does not seem like a particularly successful paraphrase for the original sentence. Again, because these tests provide sufficient conditions for an interrogative reading, the doubtfulness of this paraphrase is not a serious worry for the supporter of an interrogative reading.

The infinitive test (v) is crucial, since it both supports an interrogative reading, and speaks against a free relative reading. We have seen (from example (3)) that ‘knows’ can take both interrogatives and free relatives. However, when it comes to embedded infinitival wh-phrases we can only seem to detect interrogative readings. The sentences:

1. (23) Ruth knows when to turn off the road.

2. (24) Ruth knows why to take the high road to Larissa.

3. (25) Ruth knows whether to start off early or late.

seem only to mean that Ruth knows the propositions which answer the various indirect questions, and do not have objectual readings.Footnote ²⁵ Additionally verbs that can only take free relative wh-phrases – like ‘take’, ‘ate’ and ‘gave’ – cannot be combined with an infinitival wh-phrase. Consider:

1. (26) ? Jane took what to use for cleaning the board.

2. (27) ? John ate what to eat.

3. (28) ? Amy gave what to use to write on the board.

Since there are no good examples of infinitival wh-phrases with a free relative reading, test (v) both strongly suggests an interrogative reading of ‘S knows how to V’, and speaks against a free relative reading.

Let's take stock. The results of tests (i), (ii) and (v) strongly suggest the availability of an interrogative reading for sentences of the form ‘S knows how to V’, and test (v) also speaks against a free relative reading. Tests (iii) and (iv) are not so decisive, but in the context of the positive results on the other tests these failures do not speak against the interrogative reading.

This means that the simple view according to which ‘how to V’ always functions as a free relative in ‘knows how’ ascriptions is implausible. There are two ways in which the supporter of the free relative semantics might respond to these results: to argue that the free relative reading holds true for some sub-set of ‘knows how’ ascriptions, or to argue that ‘knows how’ ascriptions are ambiguous between the two readings.

The first move restricts attention to a sub-set of ‘knows how’, and suggest that some ‘knows-how’ ascriptions are interrogatives and others free relatives. Although it is true that the philosophically interesting class of knowledge-how is not just all knowledge picked out with a how-complement, the most plausible way to linguistically delimit the philosophically interesting kind of know-how is to focus on ‘S knows how to V’ (Hornsby Reference Hornsby1980: 84; Glick Reference Glick2011: 427; Wiggins Reference Wiggins2012), which test (v) suggests is an especially implausible candidate for a free relative reading.Footnote ²⁶ A more plausible move would be to appeal to ambiguity, claiming that ‘knows how’ has both interrogative and free relative readings, like sentence (3). This kind of view could explain the positive results on tests (i) and (ii) since if ‘know-how’ were ambiguous, the addition of interrogative modifiers or embedding within a verb like ‘wonder’ would force the interrogative reading.

However, the ambiguity view runs into serious problems. For one thing, test (v) speaks both for the interrogative reading, and against the free relative reading. It is also worth noting that B&M have themselves pointed out that ‘knows how’ fails standard tests for ambiguity (Bengson and Moffett Reference Bengson and Moffett2007: 38–40; Bengson et al. Reference Bengson, Moffett and Wright2009: 393–4). Furthermore, the standard tests for a free relative reading come up negative, suggesting that there is no available free relative reading. There are two tests for a free relative reading:

1. (vi) Wh-ever: If the wh-word can be extended to wh-ever, then the wh-phrase is a free relative (Bresnan and Grimshaw Reference Bresnan and Grimshaw1978: 334–5; Huddleston and Pullum Reference Huddleston and Pullum2002: 986–7).

2. (vii) Co-ordination: if the wh-phrase can be combined with a verb which cannot take an interrogative complement – like ‘took’, ‘believes’ or ’ate’ – then it has a free relative reading (Bresnan and Grimshaw Reference Bresnan and Grimshaw1978: 332; Baker Reference Baker1995: 204–7).

Test (vi) gives a negative result for a free relative reading. Sentences of the form:

1. (29) ? Ruth knows however to get to Larissa.

are not acceptable.

The co-ordination test with ‘believes’ (vii) also brings bad news. If the ambiguity story were correct, we would expect that ‘how to V’ could combine with ‘believes’, and other predicates which cannot embedded interrogatives. Consider:

1. (30) ? Ruth believes how to get to Larissa.

The weirdness of this result seems to reconfirm Ryle's observation that we don't believe-how (Ryle Reference Ryle2009:17).Footnote ²⁷

Kent Bach (2012) suggests an extension of this test, claiming that ‘practice’, ‘demonstrate’ and ‘perfect’ cannot relate to propositions, meaning that the fact that these verbs can embed ‘how to V’ provides evidence for a free relative reading. It is certainly true that ‘practice’, ‘demonstrate’ and ‘perfect’ cannot take that-clauses as their objects.Footnote ²⁸ However, if we look a little closer at the wh-phrases that can embed in these verbs, it becomes clear that they can embed interrogatives (see Pavese Reference Pavese2013: footnote 15). These verbs can take infinitival wh-phrases, ‘whether’ questions and multiple questions, which we have seen above can only be interrogatives:

1. (31) Mark practiced what to say.

2. (32) Jane demonstrated whether to use ‘less’ or ‘fewer’ in the example sentences.Footnote ²⁹

3. (33) Ana perfected what to do when.

The acceptability of (31), (32) and (33) suggests that these verbs can take interrogatives, meaning that the fact that ‘how to V’ can embed in them does not provide evidence for a free relative reading.

As I noted above, the failure to pass all of the tests for a given reading provides strong evidence that that reading is not present, meaning that the failure of ‘S knows how to V’ to pass (vi) and (vii) gives us strong evidence that it has no free relative reading. Together with test (v) this establishes a strong case against the suggestion that ‘S knows how to V’ has a free relative reading. I conclude that the how-complement in ‘S knows how to V’ can take an interrogative reading, but not a free relative reading.