In order to better understand the various meaning components of all-clefts, it is useful to compare them to the larger class of constructions called exclusives in Coppock & Beaver (2014). These include lexical items such as only, >mere(ly), >sole(ly), etc. Although all items discussed in Coppock & Beaver (2014) are lexical items, and an all-cleft is a complex construction, the meaning similarities between them make a direct comparison possible. The basic reading of VP only (and several other exclusives) is the complement exclusion reading, paraphrasable by nothing other than:

This is the “negative” or “at most” meaning component of only, which is at-issue content for only. Additionally, exclusives have a “positive” or “at least” meaning component: in (11) this is the prejacent, i.e. that Mary ate a salad. With VP only, this positive component is non-at-issue (presupposed).4 Coppock & Beaver (2014) provide a unifying analysis of exclusives in which the positive and negative meaning components are represented by the operators MIN(π) and MAX(π), respectively. These express that there is some answer to the current question under discussion that is at least as strong (MIN) as the prejacent π, or that there is no stronger answer (MAX) than the prejacent π (see Velleman et al. 2012; Coppock & Beaver 2014; DeVeaugh-Geiss et al. 2018 for more discussion and details on the MIN-MAX-analysis of exclusives).

All-clefts also have the complement exclusion reading, for example (5b) conveys that I ate nothing other than a salad. In order to see how this meaning component of all-clefts relates to the division of positive and negative meaning components of exclusives, it is worth making a brief excursion to it-clefts (e.g. It was John that called) and their semantic analysis.

Velleman et al. (2012) propose that it-clefts and only are symmetric in the sense that what only presupposes, an it-cleft makes at-issue, and vice-versa. The following test, going back to Horn (1981) and used by Velleman et al. (2012: 445), is based on how at-issue and non-at-issue meaning components behave in context.

Both the it-cleft (12a) and the only-sentence (13a) contain a meaning component that Mary ate a pizza. In (12a), this leads to infelicity because of information redundancy, whereas (13a) is fine: this supports the idea that Mary ate a pizza is at-issue meaning in an it-cleft (so that (12a) asserts the same information again), but non-at-issue meaning with only ((13a) asserts something new, namely that Mary ate nothing other than a pizza).

The negation sentences in (12b) and (13b) make a similar point: assuming that negation targets at-issue meaning, the tests show that Mary ate a pizza is at-issue in (12) (so that negating the cleft contradicts the preceding sentence). For only, we can explain that (13b) does not lead to contradiction because negation targets the at-issue component that Mary ate nothing other than a pizza. This does not contradict the preceding sentence.

We can apply the same tests to wh-pseudoclefts and all-clefts (see also Homer 2019 for discussion of similar tests):

We find that wh-pseudoclefts behave like it-clefts in making the proposition that Mary ate a pizza at-issue, which leads to redundancy in (14a), and contradiction in (14b). All-clefts, however, despite their syntactic similarity to wh-pseudoclefts, pattern with only: in (15a), that Mary ate a pizza is non-at-issue, and what is asserted is the novel, at-issue information that Mary ate nothing other than a pizza. In (15b), what gets negated is the claim that Mary ate nothing other than a pizza (these judgments are a little easier with a pitch accent on all).

As mentioned in the introduction, the characteristic property of all-clefts is the associated inference that what is represented by the relative clause isn’t much, the smallness effect.5 What is crucial about the smallness effect, is its very limited distribution: no smallness effect arises when a relative clause headed by all appears in a non-copular sentence. Below are some examples drawn from the British National Corpus (BNC),6 for which we don’t find a smallness effect:

Likewise, as mentioned in relation to (2b), repeated below, no smallness effect is found with everything-RCs in copular constructions.

Hence, the smallness effect is neither specific to the lexical item all as opposed to everything (witness (16)), nor to the copular construction as opposed to the non-copular construction (witness (17)), but to the combination of both.

Moreover, the smallness effect is not only found in all-clefts, but also in constructions with only, as well as some other constructions, such as no more (see Homer 2019: 17). The remainder of this subsection studies the smallness reading in more detail, and compares the smallness effects in various constructions.

In addition to the complement exclusion reading, only in (11) carries an inference that a salad isn’t much to eat (for Mary). This is similar to the smallness effect of all-clefts, but in the literature on exclusives this is usually known as the mirative effect of only (e.g. Beaver & Clark 2008: 250; Zeevat 2009). For example, Beaver & Clark (2008) talk about the “discourse function” of only as “weakening an expectation”:

Likewise, Zeevat (2009) argues that “the semantic contribution of only is only low quantity mirativity: less than expected” (p. 124).

That this mirative effect is a separate meaning component from the complement exclusion reading can be shown by using a complement that denotes something that is contextually understood as “large”:

This still conveys that Mary ate nothing other than a steak (the complement exclusion reading), but has a “funny” interpretation because it suggests that a huge steak is a small meal (for Mary). This is due to the mirativity effect (Homer 2019 has similar discussion of “funny” smallness readings).

A similar conclusion is reached by Homer (2019), who points out that the smallness effect is defeasible, as it can be detached from the complement exclusion reading. He makes the point for all-clefts, but the same observation holds for only (replace A’s response by No, he only ate a roasted pig for lunch):

In this context, only the complement exclusion reading remains in A’s response, and there is no smallness effect.

Thus, we see that both only and all-clefts carry a reading that I refer to as the smallness effect, but which is also known as mirativity. It is independent of the complement exclusion reading, and is weak in the sense that it can be overridden by context. To further investigate the status of the smallness meaning component, it is useful to see what happens when the triggering construction (only or an all-cleft) is embedded in environments such as questions or negation (the discussion and data below follow a similar discussion in Homer 2019). In questions, the smallness effect is preserved (again this effect is stronger when a “large” complement is used instead, such as 1 kg steak, recall (19)):

At-issue meaning is what gets asked in a question, so (21) targets the at-issue component of all-clefts, i.e. whether you ate nothing other than a salad for dinner. The presupposed part (that you ate a salad for dinner) projects out of the question, and the relevant judgment is what happens with the smallness effect. The reported judgment is that the inference that a salad isn’t much to eat for dinner is preserved in (21), although “maybe not as strongly” according to Homer (2019: 17).

For negation, facts are more complicated:

As for (22a), it has been observed before that specificational copular sentences are incompatible with predicate negation, unless a specific sort of correction or contrast reading is intended (den Dikken 2017: 40). In (22b), Homer (2019) claims there is no smallness effect, although the judgment seems to more subtle here.

The only-counterparts of (21) and (22b) have a similar status: mirativity is preserved in (23), but less clearly so in (24).

The data discussed above, which show that the smallness effect is defeasible, and does not project in all environments, lead Homer (2019) to the conclusion that the smallness effect is not a presupposition, but possibly akin to a conversational implicature (his p. 18), although different views exist in earlier literature regarding the mirativity of only (Zeevat 2009, and see Homer 2019: fn. 17). I will remain neutral with regard to the precise status of the smallness effect in all-clefts and other constructions, but what is important for my proposed analysis is the close parallelism between all-clefts and only.

In addition to the complement exclusion reading (section 2.2.1 above), some exclusives have what is called a rank-order reading by Coppock & Beaver (2014). This is illustrated for the exclusive merely in (25):

This sentence does not mean that this is nothing other than a down payment, but rather that it is nothing more than a down payment. Complement-exclusion and rank-order readings are analyzed in the same MIN-MAX framework in Coppock & Beaver (2014), but the nature of the scale is different (see there for details). Note that in addition to a complement exclusion reading, VP only can also have a rank-order reading, but noun-modifying only does not.

This raises the question whether all-clefts allow rank-order readings as well. A simple case, based on a hierarchy scale of employment, works with only as well as with all-clefts:

For a more contextual example, the rank-order reading of all-clefts can be illustrated in the following context, borrowed from Beaver & Clark (2008: 258). Imagine that John is at a conference, which is also attended by many philosophers. John is collecting signatures, and tries to get signatures from the most famous philosophers. With this context in mind, the following sentences refer to the fame-scale, and not the quantity scale:

Both sentences mean that John didn’t get an autograph from a more famous philosopher than Soames. They do not mean that the only signature John got, was one from Soames.

Another case, provided to me by Jessica Rett (p.c.), is modeled after a scenario from Zanuttini & Portner (2003: 50):

Again, the relevant reading here is not that I put in nothing other than A peppers, but that I didn’t put in anything more spicy than A peppers. In other words, the scale here is one of spiciness, and not of exclusion.

The presence of rank-order readings with all-clefts will turn out to be important in developing a formal analysis (in section 4), because here paraphrasability with the only thing breaks down:

The reason is that the only thing contains adjectival only which does not have a rank-order reading. Hence, an analysis that is based on quantity scales alone, or on adjectival only, will not be able to account for rank-order readings of all-clefts.

In summary, in this section I have described parallels between the mirativity of only and the smallness effect in all-clefts in terms of their meaning components and information-structural properties, their behavior under embedding in questions and under negation, and the availability of rank-order readings. This suggests strongly that the smallness effect and the mirativity of only have a similar source. I will come back to this in section 3.3. I will finish this section with a side note on focus properties of all-clefts as compared to only.

Overt only, when used adverbially, is focus-sensitive in that its semantic scope is marked by a pitch accent. See Coppock & Beaver (2014) for more discussion on the focus-sensitivity of exclusives and its source. In the case of all-clefts, it is a rather obvious observation that the semantic scope of only in the paraphrase cannot be bigger than the counterweight clause in the all-cleft:

Can the scope of only in the paraphrase be smaller than the syntactic counterweight? With overt only, we can use a pitch accent to scope-mark constituents within a DP:

For all-clefts, judgments are a little tricky. Clearly, the DP cannot be broken up in the precopular clause, (32a,b):

Putting a pitch accent on French or books as in (32c,d) does not seem to have the same effect as in (31). The clearest reading available for both (32c) and (32d) is I only like [French books]_F. The pitch accenting can be used for other purposes, for example when correcting an earlier statement You only like GERMAN books, right?. If these judgments are right, the scope of only in the paraphrase coincides precisely (is neither bigger nor smaller) than the counterweight constituent in the all-cleft.

The few mentions of the smallness effect in previous literature have all been illustrated with cases in which the relative clause headed by all is syntactically a single clause. In (43) the accompanying clause has an embedded clause introduced by the verb say. In this case, we find an ambiguity with respect to the smallness effect: a “high construal” reading in which smallness is about John’s saying, and a “low construal” reading in which smallness is about Mary’s wanting.

When the smallness effect associates with the higher clause, (43) has reading (i), paraphrasable as John only said that Mary wanted a pencil for her birthday. When the smallness effect associates with the lower clause, we get reading (ii), paraphrasable as John said that Mary only wanted a pencil for her birthday.

This ambiguity, which to my knowledge has not been reported before, is important because it suggests that the smallness effect has a syntactic origin. To see this, it is useful to compare (43) with two sets of similar scope ambiguities with reportative verbs discussed in earlier literature. The first arises with adjectival superlatives or only, as discussed in Bhatt (2002). The superlative adjective first/only in the DP in (44) can have a high construal reading (paraphrased in (i)), or a low construal reading (paraphrased in (ii)):

The second ambiguity that I want to compare (43) to is one associated with temporal adverbial clauses (Larson 1987; Haegeman 2009).13

Both the superlative adjective ambiguity in (44), and the temporal adverbial ambiguity in (45), have been accounted for by a syntactic mechanism where the low and high construal readings correspond to syntactic movement from the lower or higher clause (movement of the superlative operator -est in (44), and movement of the temporal adverbial clause in (45)).

A reviewer suggests that the reported ambiguities with all-clefts may instead be due to different ways of association with focus. However, as was discussed in the note on page 13, all-clefts do not associate with a focus-marked constituent in the way that exclusives such as only do. Hence, it is not clear how the ambiguity in (43) can be explained as an ambiguity of focus association.

Note that all examples in (43)/(44)/(45) contain the verb say. Indeed, it has been noted that the ambiguity in (45) does not arise with factive verbs (Haegeman 2009: §6). When say is replaced with factive regret, only the high construal reading remains:

This may be related to factive verbs being a barrier for movement from the embedded clause (sentences with factive embedding verbs also do not show syntactic Main Clause Phenomena, see Haegeman 2009 for discussion), or to the evidential reading that reporting verbs such as say and claim have (see Hunter et al. 2006).

Likewise, although not discussed by Bhatt (2002), we may observe that when say is replaced by regret in (44), the low construal reading is no longer available:

Now going back to all-clefts, we find that the “embedded smallness effect” observed with say in (43) is not found with factive verbs such as regret:14

In summary, there are systematic parallels between all-clefts and the other two constructions illustrated in (46) and (47): they all have an ambiguity of high and low construal readings with reporting verbs, but not with other verbs such as factive ones. These parallels provide support for the view that, just as has been proposed for superlative DPs (Bhatt 2002, recall discussion above) and temporal adverbial clauses (Haegeman 2009, recall discussion above), the smallness effect in all-clefts is due to syntactic movement in the accompanying clause.

The second set of data relevant for the movement claim goes back to the restriction on matrix negation in all-clefts, recall the discussion around (39) above. We have already seen that all-clefts do not allow matrix negation in their accompanying clause, (49). However, when the negation is in an embedded clause, the all-cleft is grammatical, as in example (50):

Here only the high construal reading is available. This suggests that negation has a blocking effect: movement out of the embedded clause past negation (in order to obtain a low construal reading) is not possible. On the other hand, the shorter movement step out of the matrix clause, which does not pass negation, is possible, and results in the high construal reading.

The following review is a summary of Traugott (2008), the only work that I know of that investigates the historical development of all-clefts in English.

The oldest attested examples of all-clefts in English have a reading where all means everything, and the postcopular clause is ascriptive or purposive (all examples in (52) and (53) are cited from Traugott 2008):

These were distributive universally quantified statements, which would be expressed with everything-RCs in present-day English.

After 1600, all-clefts started to be used in a reading with a smallness effect. Traugott argues that these constructions appeared mainly in adversative contexts, that is, ones in which “everything that one person says or does may not be enough for some other person or may be interpreted as mistaken/inadequate” (p. 160). This is illustrated with the negative case in (53a), and the modal case in (53b) in which the to-phrase no longer has a purposive reading.

In these examples, the smallness was literally expressed in the negative contexts all-clefts appeared in. The assertion that one’s actions or words were inadequate/not sufficient eventually became grammaticalized, and is now an inference associated with the word all in all-clefts. This is an example of change occurring in “dialogic” discourse contexts, which is to say involving two different viewpoints in turn-taking discourse. This is a situation that has been claimed to enable pragmatic change from neutral expressions to expressions encoding multiple perspectives (see Traugott 2008: 145 for more details).

The crucial historic change was thus that a distributive, universally quantified statement (“every action I took is x”) got interpreted as something measuring out an amount (“the sum of actions I took is x”). This amount may now be compared to some level of expectation.

This historic perspective gives us a number of insights that are relevant for developing a formal synchronic account of all-clefts. Traugott shows that the development of all-clefts took place as a separate “track” from other types of clefts such as wh-pseudoclefts. Thus, the highly restricted distribution of the smallness effect can be understood because it is a grammaticalized inference connected to a specific construction. Hence, to pursue a strategy that posits a general semantics for the universal quantifier all that comes out as having a smallness effect in certain specific environments, but not others (Approach 1 above), although perhaps attractive from a general principle of parsimony, is not compatible with the historic facts that all in this construction simply got reinterpreted because of the contexts it appeared in, and shifted away from its original universal quantifier meaning.

The opposite option, to posit a full lexical ambiguity (Approach 2), is too extreme: all₂ is not a separate lexical item, but specifically associated with the construction it appears in. Hence, I will argue for Approach 3. All in all-clefts is no longer a universal quantifier, but got reinterpreted as part of the derivational process of building all-clefts. Before proceeding to the analysis, I will review two existing accounts. The first is a recent theoretical account of all-clefts, Homer (2019), which is an instantiation of Approach 1. The second is a prominent analysis of specificational pseudoclefts, den Dikken, Meinunger & Wilder (2000), but the authors also offer some comments on all-clefts.

Homer (2019) assumes that there is a single morpheme all, which is not a universal quantifier, but rather a quantity superlative, similar to words such as most, following Hallman (2016). The accompanying clause is taken to be a free relative, which semantically serves as the comparison set for the superlative.

Homer explains the narrow distribution of the smallness effect as reflecting a semantic type distinction: all-clauses with a smallness reading denote properties (type ⟨e,t⟩). This makes them suitable to fit in copular sentences, but not in argument positions (which require an expression of type e, or a generalized quantifier). The explanation for their status as properties follows Sharvit (2015), who makes the same claim for DPs of the form the only NP. They can get a non-definite (i.e. property) reading in copular positions.

The regular distributive (i.e. universal quantifier) reading of all-RCs in argument position is obtained by type-lifting the superlative all (∃-closure). The smallness effect, which Homer assumes is associated with the semantics of the superlative operator -est, then disappears with this type-lifting operation, although the details of the process are not made entirely clear.

I have given arguments against the free relative analysis of the accompanying clause in section 3.1. In addition, there are a number of reasons to doubt Homer’s claim that the accompanying clause is an argument of all. First, English morphology does not distinguish between nominal all and adjectival all, but other languages, such as Dutch, do.

The nominal form, alles, is the one that is also used in all-clefts. This makes it less likely that it takes the accompanying clause as an argument, as one would expect this for adjectival all only.

A second problem comes from partitive constructions. Quantity expressions such as most and (according to the hypothesis) all can take partitive constructions. In this case, it seems clear that the partitive phrase it combines with indeed contains a free relative:

Crucially, though, (56b) does not have a smallness reading, even though the sentence has precisely the structure (quantity superlative + free relative) that Homer ascribes to all-clefts.

The type-based analysis for the distribution of the smallness effect runs into a problem described by Bošković (1997: 238n). He draws attention to specificational pseudoclefts in which the counterweight is morphologically an adjective:

Bošković relates his discussion to an analysis of Partee (1986), but the same argument goes for Homer’s story. In order for unusual to function as the postcopular material, it has to be type-lifted into something of type e. This incorrectly predicts, as Bošković points out, that a sentence like What John likes is tall should have a meaning like What John likes is tallness. The corresponding all-clefts (All John is, is tall/unusual) make the same point: in Homer’s story tall/unusual has to be of type e. So *All John likes, is tall is predicted to mean All John likes, is tallness. These data argue instead for an account in which all-clefts are related to their simple counterparts John {is/*likes} tall (cf. my claim in section 3.3).

In Homer’s (2019) account there is no movement in the precopular clause: it assumes that all modifies a free relative, but there is no movement of all from the accompanying clause into its final position. So his account does not meet desideratum 2, and it is not clear how it would account for the embedded smallness data, and the blocking effect of negation.

Finally, Homer’s (2019) account has a problem with rank-order readings of all-clefts (discussion in section 2.2.3 above). Because the semantics of all as a quantity superlative is based on the part-whole relation on quantity (mereology), it would analyze All John is, is a simple employee as having a complement-exclusion reading, i.e. The only thing John is, is a simple employee (see (26)/(29a) above). It is precisely the rank-order readings in which this type of only-paraphrasability breaks down: rank-order readings are possible for adverbial only, but not for adjectival only (as in the only thing), which is the basis of Homer’s analysis.

There is a great variety of accounts of the structure of pseudoclefts available (see den Dikken 2017 for a very extensive overview), and a review of even a selection of them is outside the scope of this paper. However, I will discuss one prominent analysis of specificational pseudoclefts, den Dikken, Meinunger & Wilder (2000), which is also one of the few that makes some remarks on all-clefts.

den Dikken, Meinunger & Wilder (2000) make a distinction between two types of specificational pseudocleft (SPC): Type A and Type B. The main differences have to do with the syntactic category of counterweight (full IP for Type A, other possibilities for Type B), as well as the reversibility of the pseudocleft (Type A non-reversible, Type B reversible; I refer to the original paper for further details). In their brief discussion of all-clefts, the authors state that all-clefts “display all the restrictions on Type A SPCs” (p. 81, their italics), such as connectivity effects (as discussed in my section 2.1). On the other hand, a basic property of all-clefts is that they are reversible. This would classify all-clefts as Type B SPCs, not Type A. When we look at the syntactic category of counterweights, we find a variety of options, such as IPs, VPs, DPs, and APs:

Such a variety of possible options make them more like Type B SPCs. However, PP counterweights are not possible in all-clefts unless the preposition appears (also) in the precopular position (58e), in contrast with other Type B SPCs (example from den Dikken, Meinunger & Wilder 2000: 84):15

Putnam & van Koppen (2011) consider similar data, but conclude that all-clefts are ambiguous between Type A and Type B SPCs (while their alls-construction is always of Type A, see their p. 102).16 Yet, it seems difficult to associate this putative ambiguity with a difference in meaning for all-clefts. More importantly, earlier we already concluded that den Dikken, Meinunger & Wilder’s question-answer account of (Type A) SPCs does not extend to all-clefts because they clearly lack the morphology of question-answer pairs (see discussion around my (34)). I conclude that all-clefts do not fit well with den Dikken, Meinunger & Wilder’s Type A/B distinction. I will thus reject an analysis of all-clefts along the lines of den Dikken, Meinunger & Wilder’s (2000) analysis of specificational pseudoclefts.