1 Introduction

This paper has two complementary goals. The first is to report on the expression of free choice in Tibetan, based on original fieldwork.Footnote 1 Universal free choice items in Tibetan are a combination of a wh-word and the particle’ang, optionally preceded by a nominal domain.Footnote 2

figure a

Example (1) describes someone who is not picky about their food. Dumplings? Norbu eats them. Frog? He eats that too. Whatever the food, Norbu eats it.

The second goal is to motivate a new compositional semantics for universal free choice based on the overt morphosyntax of these Tibetan FCIs.’ang is quite transparently the combination of the copular verb yin,Footnote 3 conditional suffix na, and scalar focus particle yang ‘even’ . The combination may indeed appear transparently as, but is commonly contracted to’ang in both writing and speech, and may further reduce to’i in casual speech.Footnote 4 Goldstein 2001 lists all three forms (p. 1000), but identifies’ang as the canonical form. I follow this convention here and report all examples with’ang.

figure c

In addition to forming wh-FCIs,’ang has two other uses, as a counterexpectational discourse particle—i.e. the translation equivalent for English ‘but’ and ‘however’—and as a concessive scalar particle. I discuss these uses and their compositional semantics in Erlewine 2020.

Here I pursue the null hypothesis, that’ang in the expression of free choice indeed decomposes into the ingredients in (2). The structure in (1) is thus literally as in (3).’ang is a concessive conditional (i.e. even if; see e.g. König 1986) containing a copular description with a wh-word.

figure d

My core analytic contribution in this paper will be to show how these ingredients in (3) together give rise to the expression’s behavior as a universal FCI, without stipulating universal quantificational force. Previous work has discussed both empirical and analytic connections between universal free choice and (concessive) conditionals, as well as to ever free relatives and so-called unconditionals (see e.g. Gawron 2001; Rawlins 2008a, b, 2013; Szabolcsi 2019; Balusu 2019, 2020). Existing analyses which take the connection between these constructions seriously either stipulate a covert universal quantifier in these constructions (Menéndez-Benito 2005, 2010; Rawlins 2008a, b, 2013) or propose to derive universal force from a strengthening process (Chierchia 2013; Szabolcsi 2019). I argue that universal quantificational force is instead simply a necessary consequence of the semantics of conditionals, the scalar particle ‘even,’ and the wh-phrase interpreted as a kind of indefinite, in combination.

2 An Approach to wh-quantification

I begin by introducing my assumptions regarding the compositional semantics of wh-phrases and their interaction with focus particles such as even. Studies of the semantics of wh-questions and focus association have both motivated the idea that natural language meanings may make reference to sets of alternative denotations that vary in a systematic way. In a larger project in progress (see e.g. Erlewine 2019, in prep.), I pursue the hypothesis that these two forms of “alternatives” in grammar can be productively integrated, with the result being a compositional semantics for a wide range of non-interrogative uses of wh-words, i.e. wh-quantification. I present the core of this approach here, illustrating through its application to wh-even NPIs in Tibetan (Erlewine and Kotek 2016).

I begin with a brief sketch of the compositional semantics of focus association in the framework of Alternative Semantics (Rooth 1985, 1992). Consider the interpretation of the English example with the focus particle even.

figure f

Following Karttunen and Peters 1979, the addition of even here introduces a requirement that the possibility of Tashi coming to the party is somehow particularly unlikely, compared to the possibility of other people coming to the party.Footnote 5

Let us see how this meaning can be computed compositionally. We annotate the position of focus in a sentence with ...\(_\text {F}\) (Jackendoff 1972). As Jackendoff discusses, with even in pre-subject position in English, focus must be on the subject or a subpart thereof. We therefore take the LF structure for to be as in . In Alternative Semantics, each syntactic object \(\alpha \) has two different corresponding meanings: its ordinary semantic value, , and a set of alternative denotations of equal semantic type, . The alternative set is a set of propositions that includes the prejacent proposition , as well as other contextually restricted alternative propositions that vary in the focused position. Even introduces the inference in , requiring that the prejacent proposition (that Tashi came) be the least likely among the alternatives .

figure p

The general recipe for this scalar inference of even is given in . Even simply passes up the ordinary value of its complement ; thus in , the at-issue content is the prejacent proposition, ‘that Tashi came to the party.’

figure t

Finally, even also has the function of “resetting” the alternative set to be the singleton set of the ordinary value (6c).

Let’s now step back and discuss the computation of ordinary and alternative set denotations. Just as ordinary denotations of complex expressions are determined by the denotations of their subparts , where \(\circ \) is the appropriate mode of composition (e.g. functional application), Alternative Semantics provides a procedure for calculating the alternative set denotation for a complex expression, in .

figure w

In words, for \(\alpha \) with two daughters \(\beta \) and \(\gamma \), each alternative denotation for \(\beta \) is composed with each alternative denotation for \(\gamma \); the collection of such results is the alternative set denotation for \(\alpha \).

This method for the computation of sets of alternatives in is also useful for the interpretation of in-situ wh-phrases, as was proposed earlier in Hamblin 1973. Wh-phrases have the denotation of a set of alternatives, which then compose pointwise with other material to yield the denotation of a question as a set of alternative propositions, corresponding to possible answers. I follow Ramchand 1996, 1997, Beck 2006, and Kotek 2014, 2019, in casting this Hamblinian system of wh-alternatives within the Roothian two-dimensional semantic system just presented. Wh-phrases have an alternative set denotation corresponding to its Hamblin alternatives, but no defined ordinary semantic value. See for example the denotation of who in ; its alternative set is the set of contextually-determined animate individuals which may count as short answers to who.

figure aa

Consider now the interpretation of the Tibetan wh-containing clause in below. This example must be interpreted as a wh-question, even without the final question marker gas. Tibetan is a wh-in-situ language and does not have bare wh-indefinites.

figure ac

Composing ‘who’ (8) with the rest of the clause, we yield :

figure ae

To grammatically interpret as a question, the alternatives that have been calculated as an alternative set ( ) must be made the ordinary semantic value, which is the denotation that is ultimately interpreted. This is accomplished by the interrogative complementizer (Beck 2006) or by a dedicated adjoined operator, AltShift (Kotek 2019). See especially Kotek 2019 for more on the use of this framework for the interpretation of wh-questions.

Our interest, however, is in the non-interrogative use of wh-phrases, especially in concert with focus particles. In addition to’ang FCIs, Tibetan forms NPIs through the combination of a wh-phrase and the scalar particle yang ‘even’ (Erlewine and Kotek 2016), as in :

figure ai

Let’s consider the interpretation of the grammatical (with negation) and ungrammatical (negation-less) variants of in turn. Following Erlewine and Kotek 2016, I take the focus particle yang to correspond to a unary even operator taking propositional scope at LF, as schematized in .Footnote 6 When we attempt to compute the even in this structure, however, we run into a problem. The semantics for the scalar inference of even (6a) requires that its sister have a defined ordinary value, but the sister of even in is a wh-containing clause, as in (10), and therefore does not have a defined ordinary value.

figure am

To avoid this issue, I propose the adjunction of a covert operator \(\exists \) (13) that defines an ordinary value that is the disjunction of its sister’s alternative set, and simply passes up its sister’s alternative set as its own.Footnote 7\(^,\)Footnote 8

figure an

The full LF for (11) is thus as follows in . The denotation for is as in (10), which has no defined ordinary value. The application of \(\exists \) in results in . Negation applies pointwise in . Now notice that asymmetrically entails every alternative in . This ensures that the scalar inference of even will always be true. The end result will be equivalent to the proposition ‘that no one came to the party,’ as desired.

figure ax

Now consider the variant of this structure without negation. gives the scalar inference predicted by even applying directly to in :

figure bb

Because the prejacent ‘that someone came to the party’ is asymmetrically entailed by each alternative in , this requirement in is a contradiction. This scalar inference of even can never be satisfied. Following Lahiri 1998, this fatal requirement of even in leads to the ungrammaticality of the wh-even expression without a licensing negation.Footnote 9

In this way, the Hamblin semantics of wh-phrases can be productively combined with the Roothian semantics of focus, for example giving us a compositional semantics for wh-even NPIs in Tibetan. With this background on the compositional semantics of wh-phrases and their interaction with focus particles in place, we are now in a position to turn to the compositional semantics of’ang FCIs.

3 On the Syntax of’ang

Next I address the syntax of’ang FCIs. I first address its external syntax—how the’ang expression relates to its containing clause—and then its internal syntax—i.e. the nature of the copular relation.

Taking its morphology at face value,’ang is a wh-containing conditional clause, to which the scalar focus particle yang has adjoined, and I propose that it is interpreted as such. However, there is evidence that this whole FCI structure may actually occupy a nominal argument position. Consider example (16). Here the’ang FCI hosts the dative case marker -la:

figure bg

The’ang FCI is a clause in an argument position which describes that argument, and thus in broad strokes resembles a head-internal relative clause or a so-called amalgam structure (Lakoff 1974; also Kluck 2011), as in :

figure bi

Here I propose to follow an intuition developed by Shimoyama (1999) for the interpretation of Japanese head-internal relatives, and independently by Hirsch (2016) for English ever free relatives. This idea is that the embedded clause is interpreted higher at LF, as adjoined to the embedding clause, and that the argument position is then interpreted as a pronoun anaphoric to an individual described in the clause.Footnote 10 As a concrete example, then, assuming a surface structure for (16) roughly isomorphic to below, the corresponding LF for its interpretation will resemble .

figure bl

I model the scalar particle yang as a unary even operator at LF (Erlewine and Kotek 2016; see footnote 6), taking the entire conditional structure, with its consequent clause, as its complement. This is reflected in above.

Next, we turn to the internal syntax of’ang. Again, following the overt morphology, I take the antecedent of the conditional to be a copular description involving a wh-phrase. I will suggest here that, within the Higgins 1973 classification of copular clauses, this is (in many cases) a specificational copular clause. Specificational copular clauses are distinguished through their information structure and use as well as in their syntax; for instance, pronominal reference to specificational subjects involve the neuter pronoun, as in (19a):

figure bn

Mikkelsen and subsequent authors have taken such facts to reflect that the subject of a specificational copular clause is not a referential expression of type e. In particular, Romero 2005 proposes that (definite) specificational subjects are individual concepts (functions from worlds to individuals); see also Arregi, Francez, and Martinovic to appear for recent support. As individual concepts, (definite) specificational subjects will not impose a uniqueness requirement for the nominal restriction on the evaluation world, although they will impose a uniqueness requirement on the referent given a particular evaluation world or situation. We will return to this detail, as well as discussion of indefinite specificational subjects, in Sect. 5.3.

In cases such as (16) with explicit nominal domain ‘child’ or (1) above with kha.lag ‘food,’ I take these nominals to be the first argument, or the “subject,” of the specificational copula. In the absence of such a nominal, I posit a corresponding null nominal (pro) as the first argument. The second argument of the copula is the wh-word whose alternative set ranges over individuals of type e, de re.Footnote 11 This discussion thus motivates the informal, literal translation of the specificational copular clauses using the English ‘{it/the child} is who’ in (18b) or ‘{it/the food} is what’ for (1).

An alternative analysis would be for these nominals to form a constituent with the wh-word to form a complex wh-phrase. However, complex wh-phrases in Tibetan are headed by postnominal ‘which’ and’ang FCIs cannot be built from such which-phrases:

figure bo

Therefore, I argue that the copular verb takes the noun phrase—or if absent, a corresponding null nominal—and the wh-word as two separate arguments.

4 Interpreting’ang

With these preliminaries in place, we now turn to the compositional semantics of’ang. As discussed above,’ang is a transparent combination of the copular verb yin, conditional suffix na, and scalar particle yang ‘even,’ in an amalgam-like argument position. In this section, I will show how these ingredients (even without considering ‘even’) together in the examples presented above yield a universal free choice expression. In particular, my approach does not need to stipulate the universal force for these expressions as in Menéndez- Benito 2005, 2010 or Rawlins 2008a, b, 2013, nor derive universal force from a secondary strengthening process as in Chierchia 2013 and Szabolcsi 2019.

Once we have established how universal force comes about in these grammatical examples, in Sect. 5, I show how this construction enforces universal force. There, yang ‘even’ will play a star role. Just as association with ‘even’ can build NPIs from indefinites (Lee and Horn 1995; Lahiri 1998), as we also saw in Tibetan in Sect. 2, the logical properties of ‘even’ will serve to ensure that’ang be interpreted as a universal FCI.

Recall that Tibetan’ang FCIs may be in argument positions.Footnote 12 I proposed in Sect. 3 above that a FCI in argument position is interpreted at LF as a conditional clause adjoined to the containing clause, with unary even taking the entire conditional structure as its sister.

figure bp

The \(\exists \) operator in is the covert operator discussed in Sect. 2 above. Note that \(\phi \) is a wh-containing clause, and thus without the insertion of \(\exists \), the sister of even would have no defined ordinary value (prejacent) and thus the result would be uninterpretable at LF.

As discussed in Sect. 3 above, the antecedent of the conditional \(\phi \) is a specificational copular clause. I adopt the view that the subjects of specificational copular clauses are individual concepts (Romero 2005). Individual concepts are functions of type \(\langle s,e\rangle \) from worlds or situations to individuals. Situations are subparts of possible worlds, which may be thought of as limited to particular times or places (see e.g. Kratzer 1989; Heim 1990). The type s is used for all situations, including worlds, which are simply maximal situations.

Concretely, I assume that these specificational subjects as in (21b) involve a definite determiner as in (22), taken from Elbourne’s work on definite descriptions in situation semantics. As Tibetan is an article-less language, I assume that the is unpronounced. Composing the with a nominal property such as ‘child’ in (23) yields the individual concept denotation in (24) of type \(\langle s,e\rangle \).

figure br

Individual concepts of this form will be undefined for world/situations where the property’s extension is not unique.

In cases with no nominal restrictor, I assume a corresponding null nominal (indicated as pro in (21b) above) which refers to a contextually salient property P, and which we can informally describe as “the P.” Below, I will refer to this salient property as P in the general case, whether pronounced or not.

figure bs

As proposed in Sect. 3, in LFs for’ang FCIs, there is a pronoun in FCI’s surface argument position which is related to the subject of the conditional clause in some way. I used co-indexation above as in “pro\(_\text {i}\) ... pro\(_\text {i}\)” as a notational device to highlight the link between these two nominals, but we are now in a position to specify this relationship. Specifically, I propose that these two positions refer to the same individual concept: “the P.” In the antecedent clause \(\phi \), “the P” is the specificational subject. In the consequent clause \(\psi \), “the P” is evaluated with respect to \(\psi \)’s situation or world of evaluation. We can restate the structure in (21b) in these terms as follows:

figure bt

I now turn to the compositional semantics of this LF, beginning with the antecedent of the conditional, \(\phi \). Given the semantics for ‘who’ (8) and \(\exists \) (13) above, we yield the following two-dimensional denotation for \(\phi \) in (26):

figure bu

The ordinary value of \(\phi \) is a proposition—a predicate of situations—which presupposes that there is a unique P-individual in its argument situation s and will return true if that individual is Tashi or Sonam or Migmar, etc.; e.g. in the domain of ‘who.’ The individual alternatives in each similarly presuppose a unique P-individual in the situation, but then return true when it is a particular individual in the domain.

We now turn to the interpretation of the conditional and its consequent \(\psi \). I adopt the now standard approach to conditionals as restricting the domain of a modal or temporal operator in the consequent clause (Lewis 1975; Kratzer 1979, 1986; von Fintel 1994). The modal/temporal operator in the consequent \(\psi \) (the overt main clause) in both examples that we have seen so far (in (1) and (16)) is the imperfective aspect with generic/habitual interpretation. Following Arregui, Rivero, and Salanova 2014 and citations there, I model the imperfective as a type of universal modal that quantifies over a particular set of situations. In particular, for generic or habitual imperfectives, in turn following Cipria and Roberts 2000, the relevant set of situations will be “normal or usual” sub-situations of the topic situation, formally described as “characteristic” (Cipria and Roberts 2000: 325). I write \(s' \le \!_{ch}\;s\) to indicate that \(s'\) is a characteristic sub-situation of s.

I spell out the interpretation of \(\psi \) with its imperfective quantification in . As \(\psi \) does not contain any alternative-generating (e.g. focused or wh) expression, .

figure bz

Note that, in the third line in , I have unpacked the definedness requirement of “the P” and allowed this condition to restrict the set of relevant sub-situations \(s'\). For example, if P is ‘child,’ we are allowing ourselves to look at only those characteristic sub-situations where there is a unique child to refer to.Footnote 13 In all such situations, Pema talks to that child.

We now can calculate our full conditional clause, “if \(\phi \), \(\psi \).” Recall that the conditional clause \(\phi \) acts as a restrictor on the modal base of the \(\psi \)’s modal quantification. The two-dimensional denotation for “if \(\phi \), \(\psi \)” is thus as in . The effects of this conditional restriction are boxed here for presentation:

figure cc

The final ingredient in the’ang LF in (26) is even. As even does not change the at-issue (asserted) content, our work in interpreting example (16) is now done, in . (I discuss the contribution of even in the following section.) What does this result in express? It claims that, in all characteristic sub-situations \(s'\) of the topic situation s where (a) there is a unique P (e.g. ‘child’) in \(s'\) and (b) that unique P is Tashi or Sonam or Migmar, etc.—e.g., an individual in the domain of ‘who’—Pema talks to that unique P.

Let’s restate this again in slightly more informal terms, to build an intuition for the claim. Concretely, let our salient property P be ‘child,’ and assume that all individuals that satisfy ‘child’ are in the domain of ‘who.’ Then, conveys the following:

figure cg

Note that does not require Pema to have actually spoken with any or all of these children. Instead, it uses the modal semantics of the imperfective to allow ourselves to consider different “characteristic” situations with different children present. What about a situation with Tashi? Pema talks to him. How about Sonam? Pema talks to her too. Pema talks to any child. We have successfully derived the expression of universal free choice.

How did we do this? In particular, where did the universal force of the FCI come from? The universal quantificational force of’ang in this example is that of the imperfective modal/temporal operator, whose modal base was restricted by the conditional. The imperfective introduces universal quantification over situations (see e.g. Arregui et al. 2014), with a shared individual concept evaluated in both the conditional and its prejacent, allowing us to indirectly universally quantify over different individuals in different situations.Footnote 14 On this approach, this universal force need not be stipulated as in Menéndez-Benito 2005, 2010 or Rawlins 2008a, b, 2013, nor does it need to be derived using a strengthening procedure as in Chierchia 2013 and Szabolcsi 2019. Instead, it is simply a reflection of an ingredient that is already there: the modal/temporal operator restricted by the conditional.

5 Restricting the Distribution of’ang

In the previous section, we saw how the’ang FCI derives the effect of universal quantification over a set of individuals, parasitic on a universal modal/temporal quantifier in the sentence. In this section, I discuss two principled ways in which the use and interpretation of’ang is restricted. First, I discuss the role of the scalar particle yang ‘even’ in ensuring the FCI’s universal quantificational force. Second, I discuss the incompatibility of’ang in necessity statements and episodic descriptions, and offer a new intuition for the nature of so-called subtrigging effects (LeGrand 1975).

5.1 Enforcing Universal Force

I begin by discussing the role of yang ‘even’ in enforcing the universal quantificational force of’ang. First, we consider the effect of even in example (16), which applies last in its LF (26). I repeat the two-dimensional denotation of even’s sister, “if \(\phi \), \(\psi \),” here blurring out the material that is common to all propositions, so we can more easily see their interrelationships.

figure ci

We observe that the ordinary value asymmetrically entails each of the alternatives in If “in every situation where the unique P is Tashi or Sonam or ..., blah is true,” then it follows that “in every situation where the unique P is Tashi, blah,” and “in every situation where the unique P is Sonam, blah,” etc., but not vice versa. The prejacent proposition of even is necessarily less likely than all of its alternatives, so the scalar inference of [even [if \(\phi \), \(\psi \)]] will always be true. The addition of even is felicitous here.Footnote 15

What happens if the conditional instead restricts an existential modal/temporal quantifier, e.g. a possibility modal, instead of the universal imperfective operator of the examples above? Schematically again, we can expect to yield denotations for “if \(\phi \), \(\psi \)” of the form in . The salient change from is boxed.

figure cn

Here, with existential quantification over situations, the entailment relationships between the prejacent and its alternatives have reversed. Each alternative in now asymmetrically entails the prejacent If any proposition of the form “there is a situation where the unique P is Tashi, and blah is true” or “there is a situation where the unique P is Sonam, and blah is true” etc. is true, it follows that “there is a situation where the unique P is Tashi or Sonam or... and blah is true” will necessarily be true. In this case, the prejacent is logically weaker than its alternatives. even applied to “if \(\phi \), \(\psi \)” with a possibility modal will thus lead to a systematically unsatisfiable scalar inference, resulting in ungrammaticality.

The scalar particle yang ‘even’ in Tibetan’ang FCIs thus plays a crucial role in ensuring that’ang always expresses universal free choice, just as it may serve a crucial role in explaining the distribution of NPIs (see e.g. Lee and Horn 1995; Lahiri 1998; Erlewine and Kotek 2016). The logical requirements of even—quantifying over the prejacent and its alternatives using the independently motivated semantics of wh-alternatives and their disjunction by \(\exists \), introduced in Sect. 2—ensures that the conditional clause of’ang restricts a universal modal/temporal operator, and therefore that’ang itself will always have universal force.

Practically,’ang does cooccur with possibility modals, as in example (33) below. The verb form in this example differs from (1) in the addition of the deontic possibility modal chog, and is also grammatical. The interpretation of’ang here is again a universal FCI.

figure cq

In such examples, there is in principle a choice as to which modal/temporal operator the conditional clause restricts. If the conditional of’ang restricts the ability modal, we yield prejacent and alternative set denotations of the form in (32), leading to ungrammaticality due to an unsatisfiable scalar inference of even. Instead, the conditional clause must be construed as restricting the modal base of the higher imperfective operator, leading to the attested meaning where universal free choice takes scope over the possibility modal.

5.2 On the Granularity of Modal Quantification

The approach to universal free choice presented here may at first glance lead us to predict the availability of’ang FCIs in sentences with any universal modal/temporal operator, whereas in reality its distribution is further restricted. For example, the use of’ang with the deontic necessity modal dgos is judged as highly marked, just as its intended translation in English is as well.

figure cr

Following the presentation in Sect. 4 above, we predict (34) to have an LF representation as in (35) below. In every deontically best accessible world, where the unique medicine is x, you take x.

figure cs

The problem with (34/35), I suggest, is a conflict between the granularity of the modal quantification and the uniqueness requirement of the definite individual concept “the medicine.” Specifically, I take the modal must here to quantify over possible worlds that are best according to an ordering source. The ordering source introduces considerations of what ought to be done in particular cases, but it does not change facts of the world, such as the uniqueness of medicine. In each world of evaluation, the uniqueness requirement is not satisfied, and thus the sentence cannot be evaluated.Footnote 16 In contrast, in the grammatical examples above, the conditional in’ang restricted the domain of quantification over a set of situations which could be granular enough to be restricted to situations with unique P-individuals.

A similar analysis applies to episodic descriptions, which is another context where’ang FCIs are unavailable. See example (36) and a grammatical, FCI-less baseline in .

figure cu
figure cv

Episodic descriptions simply claim the existence of a particular type of event: here, (37) asserts that there was a completion of an eating event, in the past,Footnote 17 in the halo of the speech time ‘now.’ There is no overt modal/temporal operator. Let us assume, following Kratzer 1986, that the conditional in’ang in such a case will restrict the modal base of a high, covert epistemic necessity modal.Footnote 18 Assuming that such a covert epistemic necessity modal quantifies over doxastically accessible worlds, we will again run into problems satisfying the uniqueness requirement of the specificational subject.

5.3 Subtrigging

As with FCIs in other languages, though, the restrictions on the distribution of’ang may not be absolute bans. Specifically, the restriction due to issues with the granularity of modal quantification just introduced above only holds in so far as the subject of the specificational clause is definite; see (35). Instead, if the content of the conditional clause in’ang takes an indefinite specificational subject, as schematized in (38), this problem could be avoided.Footnote 19

figure cw

In particular, the structure of the form in (38) will not require the worlds (or situations) that are quantified over to have a unique individual that satisfies the property ‘medicine,’ which I claim led to the unavailability of the’ang FCI in example (34). In reality, example (34) is judged as unacceptable, so this alternate parse in (38) with an indefinite specificational subject must not be available in example (34), if it is indeed ever available.

I propose that parses for’ang with indefinite specificational subjects, as sketched in (38) above, are in principle available, and that this option holds the key to understanding another aspect of the distribution of FCIs. Specifically, I predict that the availability of the indefinite subject parse as in (38)—which predicts the availability of the FCI without quantification over granular situations, and thus in a wider range of contexts—should only be as good as the general availability of specificational copular clauses with indefinite subjects.

It has been independently observed that subjects of specificational copular clauses are generally definite, but tolerate certain exceptions:

figure cx

In particular, modification—especially by relative clauses—seems to lead to acceptability. See e.g. See e.g. Mikkelsen 2005: ch. 8, Heycock 2012, Comorovski 2007, and more recently Milway 2020 for discussion.

I suggest that this restricted acceptability of indefinite specificational subjects and its amelioration as in (39) is in turn responsible for the similar amelioration of FCIs in some environments when modified, dubbed “subtrigging” by LeGrand (1975). Tibetan exhibits this subtrigging effect as well: Example (40) differs from the unacceptable (34) in the addition of a relative clause on the nominal domain and is judged as perfectly acceptable.

figure cy

Again, taking the morphology of the FCI seriously—in this case, that’ang involves a copular description—led to this novel connection between the behavior of FCIs and specificational copular clauses. I will leave a further understanding of the nature of this effect itself for future work.

Finally, we should also wonder whether the explanation for FCI subtrigging effects that I suggest here can or should be extended to account for apparently parallel subtrigging contrasts in languages such as English , where FCIs do not obviously reflect the involvement of a specificational copula. I will also leave the exploration of this question for future work.

figure da

6 Summary and Outlook

This paper develops a new compositional semantics for universal free choice, from the predictable interactions of a number of ingredients. A specificational copular conditional clause describes an individual concept which the consequent then makes reference to. The conditional restricts the modal base of a modal/temporal operator in the sentence. And finally, the scalar particle even associating with a wh-indefinite, enforces that the modified modal/temporal operator be a universal quantifier over situations. This leads indirectly to a kind of universal quantification over individuals in the domain of the FCI. The end result is a new approach to the universal force of universal FCIs, without directly prescribing or deriving this force. Furthermore, we have seen that this approach offers a new analytic possibility for reducing so-called “subtrigging” effects to an independently observed constraint on indefinite subjects in specificational copular clauses.

Most importantly, this proposal is not a hypothetical proof-of-concept. The expression of universal free choice in Tibetan, documented here, transparently involves these ingredients: a concessive conditional clause (even if) with a copular description, ranging over different possible individual referents in the domain of a wh-word.

If this analytic approach for universal free choice is truly successful, we might imagine that other languages would also express universal free choice claims in this way. There is recent work suggesting exactly this. For example, Rahul Balusu (2019; 2020) has investigated different uses of concessive conditional expressions in a range of Dravidian languages; one such use is the formation of FCIs with exactly the same surface morphological makeup as in Tibetan: a wh-word with a copula and concessive conditional ending. Balusu also describes these copular descriptions as specificational.

figure db

There are, however, some subtle differences in these constructions amongst different Dravidian languages, and between them and Tibetan. For example, Telugu wh-ai-naa FCIs allow both universal ‘anything’ as well as existential ‘something or other’ readings, although Kannada and Tibetan do not have such existential readings. See Balusu 2019, 2020 for discussion.

Additional evidence comes from the form of universal FCIs in Japanese, which appear to be formed of a wh-phrase with a particle demo, as in . On the identity of this particle demo, Nakanishi (2006: 141) states, “-Demo can be morphologically decomposed into the copular verb -de followed by -mo [even]. However, it is not clear whether this decomposition is necessary.” In recent work, however, Hiraiwa and Nakanishi (to appear) push the decompositional hypothesis a step further, specifically proposing that Japanese wh-demo FCIs have the underlying structure in , with a type of ellipsis obscuring the conditional morphology.Footnote 20

figure de

Whether expressions with demo indeed always reflect the structure in in the synchronic grammar of Japanese—or if the hypothesized structure in is better thought of as the diachronic source for what is now a single grammaticalized particle, demo—in my opinion warrants further debate. Still, the parallel as in is additional fodder for the broad cross-linguistic viability of the decompositional approach to universal free choice developed here. See also Haspelmath 1997: 135–140 for discussion of indefinite expressions in many other languages which also exhibit morphological traces of copulas and concessive conditional morphology, some of which are still clearly FCIs, whereas others have extended to other indefinite types (pp. 149–150).

Furthermore, each of these concessive copular conditional expressions in both Dravidian languages and Japanese have a number of additional uses, which in fact largely overlap with the range of uses for Tibetan’ang (Erlewine 2020). The clear parallels in both the morphosyntactic composition and interpretational range of these expressions, across these genetically unrelated languages, further strengthens the motivation to take the decompositional approach to these expressions seriously, as well as to better document and understand the microvariation observed in their fine-grained behavior.