Dever on What-the-hell and Grice

From Dever at

https://webspace.utexas.edu/deverj/personal/papers/compositionality.pdf

The section in Dever's paper is entitled,

"Some Problem Cases for Compositionality"

"While the results of §2.1 show that when semantic theories are sufficiently unconstrained, compositionality can be cheaply obtained, the question remains whether a satisfactory semantic theory for a natural language, one properly responsive to natural constraints on semantic and syntactic facts, can be given compositional form."

"In this section we consider two problem cases for the construction of compositional semantics, examining the data which resist a compositional treatment and then considering ways of overcoming that resistance. The goal is not to settle the question of whether natural languages have compositional semantics, or even the smaller questions of whether the particular phenomena discussed here have a compositional semantics, but rather to see how questions of compositionality influence semantic theorizing."

3.1 What the Hell

The Problem:

Compare the following two sentences:

(13) Who bought that book?

(14) Who the hell bought that book?36

Dever noets:

"The literature abounds in problem cases for compositionality and treatments thereof. In addition to the issues addressed below, see among many others discussion of compositionality and:

-- independence-friendly logics in [28], [29], [30], [31], [32]; prototype theory in [16] (ch. 5), [19], [18], [39], [53], [62];

-- idioms
in [42], [41], [52], [71];

-- ‘unless’ in [27], [54], [66];

-- propositional attitude cases in innumerable places, but
especially [60] (ch. 4), [61] (ch. 8), [46], [8];

-- ‘any’ and other negative polarity items in [48], [47], [4], [50],
[45];

-- anaphora in [38], [24], and [34]. [36]

"also contains an overview of several compositionally-problematic
semantic phenomena."

36

"These two examples are drawn from [10]".

"The behaviour of ‘the hell’ phrases was first noted in [58]."

"The two are roughly synonymous. While the addition of ‘the hell’ alters the rhetorical impact of (14) --

perhaps encouraging the

conversational implicature

that 'It is surprising that the book was bought'),

the core semantic value of each is a request for information about the identity of a book buyer."

"Whatever the semantic contribution of ‘the hell’, it must be compatible with the close semantic proximity of (13) and (14)."

"However, ‘who’ and ‘who the hell’, or, more generally, ‘wh-’ and ‘wh- the hell’
expressions, diverge in meaning in other contexts. Thus:37

"The minimal variant of adding a modal auxilliary causes a difference to emerge:

(15) Who would buy that book?

(16) Who the hell would buy that book?

"The first of these is most naturally read as a request for information, but the
second is most naturally, and perhaps obligatorily, read as an indirect assertion
that nobody would buy that book.38"

"When the original examples are embedded in an indirect question, a difference
in grammaticality emerges:

– I know who bought that book.

– I know who the hell bought that book.

"‘Who the hell’ phrases are grammatical only in negative contexts, whether overt:
37

"The first three of these examples are drawn from [10]; the last draws from [10] and [58].

38

"[10] claims that only the indirect assertion reading of (16) is available, but I find the data less univocal.

"The pressure toward the indirect assertion reading, in my judgement, increases with the strength of the attached vulgarity."

"In order to keep this volume suitable for a family audience, I have used ‘the hell’
throughout, but the reader is encouraged to substitute as his imagination allows."

– I don’t know who the hell bought that book.

in the antecedent of a conditional:

– If anyone knows who the hell bought that book, please tell me.

or in the scope of so-called adversative attitude verbs:39

– John refused to tell me who the hell bought that book.

"‘The hell’ blocks certain scope readings of sentences with multiple quantifiers.:"

Thus:

– What did everyone buy for Max?

"is ambiguous between a reading on which ‘everyone’ takes wide scope, and people
make separate purchases for Max, and a reading on which ‘everyone’ takes
narrow scope, and there is some one thing bought by everyone for Max."

"However:

– What the hell did everyone buy for Max?

"allows only the second of these two readings."

"‘Wh- the hell’ phrases, unlike normal ‘wh-’ phrases, cannot enter into anaphoric
attachments."

Thus:

– Someonei walked in the park, but I don’t know whoi.

is acceptable, but:

– Someonei walked in the park, but I don’t know who the helli.

is not.

"Similarly, ‘which’ phrases, which require an anaphoric link to a contextually
provided range of salient objects, do not allow ‘the hell’ modification."

39See [47].
27

– Which the hell book did you read that in?

"The puzzle for compositional semantics is to show

how ‘the hell’ can systematically

contribute to the meanings of larger expressions in a way that allows

its impact to be minimal, if anything at all, in (14), but much greater in the other cases set out above.40"

A Solution:41 A simple ‘who’ question can have its interpretation influenced by
linking the range of admissible answers to a contextually-provided domain. Thus consider the following dialogue:
(17)

A:

Various friends of mine voted for each of the different presidential
candidates in the 2000 election.

B:

Really? Who voted for David McReynolds?

"B’s question is not answered by specifying an arbitrary McReynolds voter (and does
not require listing all such voters); rather, it calls for a (or all) McReynolds voters among A’s friends."

"In another context, however, ‘Who voted for David McReynolds?’
can receive an unlinked reading, in which it calls for the total list of McReynolds voters."

"Suppose the semantic function of ‘the hell’ is to require that the range of admissible answers to a wh-question include novel answers – ones not already provided as possible by contextual linkages of the sort just discussed."

"When a wh-question is an unlinked one, as on one natural reading of (13), adding ‘the hell’ has no effect, because when
unlinked, all answers are novel."

"But when the wh-question is a linked one, adding ‘the
hell’ has a semantic impact. Thus consider:
40

"A compositional semantics need account for the failures of grammaticality such as

I know who the hell bought that book

and

Which the hell book did you read that in.

only if the syntax is not thoroughly autonomous.

"It is tempting to think that grammatical failures due to failures of anaphoric linkage, at least, have a semantic explanation."

"The following solution is a simplified and modified version of the proposal of [
10].

Any shortcomings of it are due to the present alterations.
28

(18)

A:

Various friends of mine voted for each of the different presidential
candidates in the 2000 election.

B:

Really? Who the hell voted for David McReynolds?

"This dialogue, unlike the first, creates the implicature that B expects all of A’s friends not to have voted for McReynolds."

"If the effect of adding ‘the hell’ is to insist on the admissibility of novel answers (here, people other than those B counts as A’s friends),
this new implicature is to be expected."

"The various effects of ‘the hell’ noted above now fall out."

"‘Wh- the hell’ phrases refuse anaphoric linkage because that linkage dictates the
range over which the wh- phrase ranges, which contradicts the novelty requirement
imposed by ‘the hell’.42 ‘Which’ phrases, which always require anaphoric/contextual
linkage, can thus never combine with ‘the hell’."

"A question of the form ‘Who would buy that book?’ takes as answer pairs of people
and possible situations.43 Given the broad total range of possible situations,
such a question is typically linked to a contextually-provided range of admissible
situations.

44

"Adding ‘the hell’ to form ‘Who the hell would buy that book?’
requires the admissibility of novel answers, and thus defeats any contextuallyprovided restriction on admissible situations. But once all possible situations are provided, the question becomes trivialized: anyone would, in some situation, buy the book."

"The asking of trivial questions, though, is pragmatically proscribed,

"Although note the acceptability of:

– Someone walked in the park, but I don’t know who the hell it was.


"Thus: ‘Albert, if it has a chapter on direct reference’, ‘Louisa, if autographed copies are available’, etc.

"Thus ruling out answers such as ‘Brian, if we threaten to kidnap his dog if he doesn’t’."

"and an alternative communicative explanation is favoured, such as the explanation
that the speaker is emphasizing the remoteness of any situation in which the
book is bought."

"The requirement of novelty imposed by ‘the hell’ is impossible to fulfill when
the ‘wh- the hell’ phrase is simply imbedded in an operator of positive epistemic
commitment. To say that I know who the hell bought the book is to undermine,
by my knowledge, the requisite novelty of the admissible book buyers. Similarly
an epistemically positive operator in the antecedent of a conditional, such as the
earlier."

If anyone knows who the hell bought the book, please tell me.

"creates no conflict with the novelty requirement, since the function of the antecedent is to entertain hypothetical situations. Operators of negative epistemic
commitment, such as ‘refused to tell’, for similar reasons allow ‘the hell’ modification."

"The novelty requirement thus explains the distributional facts noted
above."

"The novelty requirement makes ‘wh- the hell’ phrases negative polarity items,
where various sorts of negation license the introduction of novelties. Suppose
that negative polarity items are subject to:

(Immediate Scope Constraint)

"A negative polarity item can appear only in the
immediate scope of its licensing negative item.45

Consider again:

– What the hell did everyone buy for Max?

See [50] for formulation and defense of the Immediate Scope Constraint.
and assume that the licensing item is the marker of interrogative force.

"If ‘everyone’ is raised to give it scope over ‘what the hell’, it intervenes between ‘what the hell’ and its licenser, violating the Immediate Scope Constraint. The unavailability of a reading wide-scoped for ’everyone’ is thus explained.

------

SECOND PROBLEM-CASE:


3.2 Many Scandinavians

The Problem:

"Fourteen Scandinavians have won the Nobel prize in literature.46 Since
there have been only 99 Nobel laureates in literature, and since Scandinavians are only about 0.5% of the world’s population, the following claim looks acceptable:

(19) Many Scandinavians have won the Nobel prize in Literature.

On reflection, however, the acceptability of (19) is puzzling.

"Fourteen, after all, is not by most natural standards many."

"Consider the oddity of:

(20)

Many Scandinavians have emigrated to the United States. Fourteen, in fact.

----

"The acceptability of (19) seems to derive from having fourteen qualify not as many
Scandinavians, but rather as many winners of the Nobel prize in literature, and thus
from reading (19) as:

(21)

Many winners of the Nobel prize in literature have been Scandinavian.

"A similar apparent ‘swapping of positions’ can be found in sentences with adverbs of
quantification47:

46Bj¨ornstjerne Bj¨ornson, Selma Lagerl¨of, Verner von Heidenstam, Karl Gjellerup, Henrik Pontoppidan, Knut Hamsun, Sigrid Undset, Erik Karlfeldt, Frans Eemil Sillanp¨a¨a, Johannes Jensen, P¨ar Lagerkvist, Nelly Sachs, Eyvind Johnson, and Harry Martinson.

"This observation and the subsequent problematic sentence

(19) are both due to [69] in 1985.

"The intervening years have not been kind to the recognition of Scandinavian
literature, and the number remains 14."

47See [9] and [7]
31

(22)

Scandinavians often win the Nobel prize in literature. (The Nobel prize in
literature is often won by Scandinavians.)

and in sentences with generics48:

(23) Scandinavians win the Nobel prize in literature, but Americans win the Nobel
prize in economics.

(The Nobel prize in literature is won (generically) by
Scandanavians, but the Nobel prize in economics is won (generically) by
Americans.)

"Some other determiners, such as ‘few’ and ‘several’, exhibit similar behaviour. These cases look like violations of the semantic locality component of compositionality."

"‘many’ in imposing its cardinality constraint requires semantic interaction with the
syntactically distant verb phrase.49

A Solution:

"We concentrate on addressing the problem in its ‘many’ form, deferring
integration of these suggestions with theories of adverbs of quantification and
generics. In some cases, the semantic role of ‘many’ in a sentence of the form ‘Many
X’s Y’ is merely to require that the number of X’s that Y is above some minimum
threshold cardinality .50 This sort of ‘many’ can be given a straightforward compositional semantics."

48See [6] and [7]

"The apparent position swapping can be taken as suggesting that ‘many’ is symmetric, in the sense that

‘Many X’s Y’ and ‘Many Y’s X’ are equivalent."

"From the perspective of compositionality, however, the observation
that ‘many’ is symmetric (if correct) merely restates the problem – how can ‘many’ be symmetric, given that symmetry requires giving equal semantic footing to the immediate complement of ‘many’ and a syntactically distant verb phrase?"

"The required number of Y-ing X’s will surely be vague, but the idealization to a specific is harmless here.

~many = f : }(D) 7! }(}(D)), f (X) = {Y : |X
T
Y| }
• ~[S [DP many X] [VP Y]] =true iff ~Y 2 ~many(~X)

"‘Many’ applied to a noun phrase thus yields a collection of sets, each of which contains many of the satisfiers of the noun phrase. If any of those sets is the extension of the verb phrase, the sentence is true. ‘Many’ is symmetric on this ‘cardinality’ semantics, in the sense that ‘Many X’s Y’ is equivalent to ‘Many Y’s X’, which would allow for position
swapping without alteration of truth value, but no plausible value for accounts for the truth of (19)."

"In other cases, though, the impact of ‘many’ seems more subtle than a simple cardinality constraint. Compare the following."

(24)

Many philosophers of mathematics have read Russell and Whitehead’s
Principia Mathematica.

(25)

Many Brazilians have read Russell and Whitehead’s Principia Mathematica.

"Suppose that among the 2000 philosophers of mathematics, 800 have read the Principia, and that among the 180 million Brazilians, again 800 have read the Principia."

"Then (24) looks true and (25) false, which is impossible if ‘many’ simply imposes a
cardinality constraint. This suggests the following alternative reading of ‘many’:
• ~many = f : }(D) 7! }(}(D)), f (X) = {Y : |X
T
Y| |X|}
• ~[S [DP many X] [VP Y]] =true iff ~Y 2 ~many(~X)
where sets the threshold percentage for manyness. Symmetry is now lost, since the
complement noun phrase to ‘many’ has the privileged role of providing the number of
objects a percentage of which must satisfy the verb phrase. This second, ‘proportionate’
reading thus also gives the wrong analysis of (19).

"On both the cardinality and the proportionate analyses, if the number of X’s who Y
is the same as the number of Z’s who W, then many X’s Y if and only if many Z’s W.
However, some examples fail to fit this pattern."

(26) Many penguins live in Antarctica.

(27) Many penguins live in my bedroom.

"Twelve penguins in a bedroom suffices for many, but will hardly do for a whole continent."

"Here ‘many’ means something like more than one would expect, where expectations
are set (in part) by the verb phrase. This ‘expectation’ semantics can be roughly
characterized by requiring that X’s Y at a rate greater than the general rate of Y-ing."

• ~many = f : }(D) 7! }(}(D)), f (X) = {Y : |X
T
Y| |X|·Y|
|D|
}

~[S [DP many X] [VP Y]] =true iff ~Y 2 ~many(~X)

"This is a crude measure of expectation, but it suffices for a start. The resulting semanticsis symmetric, like the cardinality semantics, and gives the desired result for (19), since Nobel laureates in literature are represented among Scandinavians at a higher rate than that at which they appear in the general population.
The ‘expectation’ semantics uses ‘value loading’ to get the right truth conditions for (19) in a compositional manner, by granting ‘many’ a parameterized sensitivity to the verb phrase interpretation which is ‘passed up’ until semantic composition meets that part of the sentence."

"However, it fails to explain why the ‘position swapped’ reading
of (19) seems preferable to the straight reading, given that it makes the two equivalent."

"It also yields undesirable results in closely related cases. Nobel laureates in literature appear in the general population at a rate of about one in every 60 million."

"Thus St. Lucia, with its population of some 200,000 and a single Nobel laureate in literature52, dramatically exceeds the expectation threshold. But the claim:

(28) Many St. Lucians have won the Nobel prize in literature.

"I idealize here by assuming all Nobel laureates in literature are currently alive.

52Derek Walcott

"seems false. One laureate out of the 99 is too few, no matter how few the St. Lucians are."

"This result is predicted if the ‘proportionate’ semantics is applied to the position
swapped."

(29) Many winners of the Nobel prize in literature have been from St. Lucia.

"But the position swapping is then again a problem for compositionality."

"Two possible moves at this point."

"Combine the ‘proportionate’ semantics with a syntactic story swapping the argument
positions of ‘many’ sentences prior to semantic analysis. Thus (19), at
the level of semantic analysis, would be ‘Many winners of the Nobel prize in
literature have been Scandinavian’, which would then combine with the ‘proportionate’
semantics to yield to desired result."

"However, note that quantification
over empty classes is typically pragmatically disfavoured; hence the peculiarity
of:

(30) Many Freedonians have won the Nobel prize in literature.

But as [7] observes, a sentence like:

(31) Many Scandinavians have won the Nobel prize in silly walks.

... seems simply false, rather than pragmatically disfavoured, despite the fact that
the syntactic swapping story would make the class quantified over the empty
class of Nobel laureates in silly walks."

"Give a ‘reverse proportionate’ semantics, by altering the class a minimal percentage
of which needs to behave as required:

– ~many = f : }(D) 7! }(}(D)), f (X) = {Y : |X
T
Y| |Y|}

(19), under this approach, requires a certain minimal percentage of the Nobel
laureates in literature to be Scandinavian.53

"However, we must now explain why
53Note that this approach allows (28) to be false.
35

"...‘many’ allows both proportionate and reverse proportionate semantics, while
‘most’ allows only the proportionate semantics:
– ~most = f : }(D) 7! }(}(D)), f (X) = {Y : |X
T
Y| |X|
2 }
– ~most = f : }(D) 7! }(}(D)), f (X) = {Y : |X
T
Y| |Y|
2 }

-- and so on.