Not every pronoun is always a pronoun

Abstract

A homonymy analysis is proposed to explain the so-called “demonstrative use” of personal pronouns. This analysis explains why some pronouns (it) do not allow a demonstrative use, as demonstrated in Nunberg (1993). The absence of a demonstrative feature in it can also account for the fact that it does not allow deferred reference. It is argued on the basis of the structure of the nominal demonstrative paradigm that the homonymy analysis is more parsimonious than a single-item analysis.

1 The “demonstrative use” of pronouns

Personal pronouns such as he and she can typically be used in three different ways: as bound variables, as unbound anaphors, and in the manner of demonstratives, as illustrated in (1):

(1)	[Pointing at a woman who has not previously been referred to:]
	She's up early!

While some authors have entertained the possibility that a lexical ambiguity may underlie these various options (see, e.g., Kaplan, 1989a, b; Braun, 2008), the standard assumption today (e.g. Elbourne, 2008; Heim and Kratzer, 1998) is that there is one lexical item she, with a single lexical semantics that gives rise to these various options in different structures. A straightforward and widely adopted implementation is to assume that pronouns are variables, which can either be bound, or remain unbound and receive their reference via a contextually given assignment function that somehow reflects the referential intentions of the speaker (Heim and Kratzer, 1998; Partee, 1978, a.o.). The latter option then gives rise to the “demonstrative use” of pronouns.

In this paper I will argue for the homonymy approach. There exist both a personal pronoun she, and a pronominal demonstrative she. Personal pronouns cannot be “used demonstratively,” hence what appears in (1) must be the demonstrative. This approach may appear less theoretically parsimonious than the alternative, but appearances are deceiving, as I will argue in Sect. 5. The immediate advantage is that it will allow us to describe the distinction between pronouns that can be used demonstratively, and pronouns that can not: I will argue that the former have demonstrative homonyms, which allow demonstrative use, while the latter are always personal pronouns, which do not. I heavily rely on Nunberg (1993) for the basic facts: unlike she, English it cannot be used demonstratively, nor does it allow “deferred reference,” as I discuss in Sects. 2 and 3. What I have to add to Nunberg's seminal work is a discussion of additional observations, and a more technical analysis of the distinction drawn there between deictic (“strongly indexical”) and merely indexical pronouns, which will support the homonymy approach.

2 Referent selection

As Nunberg (1993: 34) observes (see also Elbourne, 2005; Wolter, 2006), it cannot be used demonstratively:

(2)		[Pointing at a vase which has not previously been referred to:]
	*	I bought it yesterday.	(cf. this)

Nunberg concludes from this observation that while it can be “contextual” (in examples such as (6) below), it is not deictic (does not involve “explicit indication of a feature of the context of utterance”), so that it cannot be used demonstratively. He does not, however, provide an explanation for the distinction between it and she or he. Nunberg appears reluctant to accept a homonymy analysis of he, and holds instead that while he lacks an “explicit deictic component,” it can nonetheless be used deictically, with deixis “supplied by a demonstration.” This does not help to explain however why it, which equally lacks an explicit deictic component, cannot have deixis supplied by a demonstration.¹

The homonymy analysis will help solve this problem. I propose that unlike the personal pronoun he, the pronominal demonstrative he has a deictic component to its meaning that is supplied by a component of its featural makeup; let's call the relevant feature dem. It, on the other hand, is only a personal pronoun: it never spells out dem (for reasons discussed in Sect. 5), hence lacks this semantic component.

The obvious alternative—non-homonymy—approach to the contrast between (1) and (2) would be that personal pronouns, such as he, though lacking dem, can be used demonstratively in general, but that there is some exceptional deficiency in it that blocks this usage in (2). In this section I will review several possible alternatives along these lines, and discuss some additional data showing a contrast between (s)he and it that will serve to disprove them.

One obvious explanation for (2) might be that it, being a weak pronoun (see, e.g., Cardinaletti and Starke, 1996, 1999), cannot be stressed (as suggested e.g. by Wallenberg, 2008: 490), whereas demonstratively used pronouns must be (plausibly because their referents are not Given, in the sense of Schwarzschild (1999: 151, (25a)), so that they require Focus). However, Cardinaletti and Starke (1999: 161ff.) argue against the view that weak pronouns cannot receive stress in general. For it in particular this was refuted by Wolter (2006: 206), who provides the examples in (3), with stressed it in subject position and to the right of a verbal particle. A stressed anaphoric it also occurs in (4a) and (4b), and a stressed donkey pronoun it in (4c) (from Sauerland, 2008: (68b)). Examples (9b) and (15b) below illustrate the same fact. I use bold face to indicate the presence of a pitch accent.

(3)	a.	The dog chased the cat and then it chased the dog.
	b.	At first we were afraid that the avian flu would wipe out the population, but fortunately the vaccine wiped out it.
(4)	a.	Life may be giving up on me, but I'm not giving up on it. (ER, S. 1, ep. 2)
	b.	I have walked in that land […] I do not fear death, but I fear it.
				(Le Guin, 2001: 195)
	c.	Every farmer who owns a donkey beats it, but every farmer who owns a horse beats it.

In addition, the data in (5) demonstrate that operators that associate with focus can associate with it:

(5)	a.	It, too, is worth saving.
	b.	But even it is flawed.

Another potential explanation would employ the concept of strong familiarity from Roberts (2003) and later work. A discourse referent that has not been mentioned in the current discourse, but whose existence is, e.g., contextually entailed, or generally accepted in the culture, is weakly familiar. A discourse referent is strongly familiar iff it has been explicitly introduced, i.e. via the (previous) utterance of a DP. Now one might suppose that it for some reason requires a strongly familiar discourse referent; this will block (2).

This alternative fails for two reasons. A weak counterargument starts from the example in (6a), again adapted from Nunberg (1993: 33).

(6)	a.	[Speaker and hearer are walking through the Taj Mahal:]
		Gee, he certainly spared no expense.
	b.	[Speaker and hearer stare at a shrine for a deceased pet for a while. Then:]
		They must've really loved it!

While the builder of the Taj Mahal is not being demonstrated in (6a), he seems salient enough “in the context or in the consciousness of participants” to be the referent of he. (6b) shows that such contextual use is possible for it as well, as noted by Nunberg. This appears to refute the strong familiarity thesis for it. However, I hesitate to rely fully on this refutation, because I suspect, without strong evidence, that the discourse participants may be treating the discourse referents accessed by he and it in (6) as if they were strongly familiar, by pretending that the referents had been introduced by DPs in a silent or “pretend” preceding discourse discussing the builder/pet in question. This speculation would at least simplify an explanation for the absence of focus on the pronouns, since otherwise there is no salient antecedent available to render them Given, in the sense of Schwarzschild (1999: 151, (25a)) (see also Spathas, 2010 for extensive discussion of focus on pronominals). Also, the examples only appear to be felicitous in a situation that would actually allow such a preceding discourse to have taken place: I feel (6a) would be odd if it were spoken just as the participants were meeting in front of the Taj Mahal.

Fortunately, we can derive direct counterevidence to the strong familiarity hypothesis from the fact that discourse referents are not accessible for a demonstrative use of it even when they are strongly familiar. Compare (7) to (8):

(7)		[Pointing at one vase after another in a shop window:]
		Here is a vase, and here is a vase, and here is a vase, and
	*	it is really expensive!
(8)		[Pointing at one man after another in a lineup:]
		Here is a suspect, and here is a suspect, and here is a suspect, and
		he is the perp!

(8) shows that he, when accompanied by a demonstration, can be used to select one out of several previously mentioned referents. This is not possible for it in (7). We draw two conclusions from this. First, it shows that postulating a strong familiarity requirement for it will not suffice to explain its exceptional behavior, since the vase pointed to in (7) is strongly familiar. This is enough to reject the strong familiarity thesis.²

But (7) also supports a further conclusion. In other contexts, it can be used anaphorically to pick up an existing discourse referent, provided of course that the intended referent is sufficiently salient.³ In view of this, (7) suggests that a mere act of demonstration does not suffice to render the demonstrated object salient—or more salient than other possible referents. This will explain the infelicity of (7), as the hearer will then not be able to resolve the reference of it. But if this is so, then how does the act of demonstration help to pick out the referent of he in (8)? The acceptability of (8) shows that he is sensitive to the act of demonstration in a way that goes beyond the act of demonstration merely singling out one referent as the most salient in the context. We can explain this by assuming, first, that the semantics of demonstratives indeed makes reference to the act of demonstration, however defined, in the utterance context (contra, e.g., Wolter, 2006; Elbourne, 2008; Nowak, 2016, but following, e.g., Kaplan, 1989b; Roberts, 2002; Hinterwimmer, 2019), and, second, that he in (8) is a demonstrative.

One final way of trying to pin the contrast between he and it on a deficiency in the latter might be to require that the referent of it must be not just strongly familiar, but uniquely so. However, this is defeated by the data in (9), with a context (after the first conjunct) containing two discourse referents that are both strongly familiar, where both are available as referents for it, with Focus being used in (9b) to select the less topical one among them (doing the work that the act of demonstration fails to do in (7)).

(9)	a.	First the dogi chased the catj, and then iti chased me.
	b.	First the dogi chased the catj, and then itj chased me.

The uniqueness requirement in itself (Wolter, 2006) must also fail for one instance of it in the second sentence of (10), whatever the salience relations established by the first:

(10)	First the cat chased the mouse. Then it caught it.

So far, we have seen that it and (s)he differ in that the latter may be used demonstratively to refer to a previously unmentioned referent in (1), and to select among several possible referents in (8), whereas the former may not. A third distinction is seen in (11)–(14). (11) illustrates the well-known fact (see e.g. Kaplan, 1989b: 586ff.; Braun, 1996) that subsequent uses of the same demonstrative or pure indexical, when accompanied by a change in the utterance context, may select different referents:

(11)	a.	It hurts here, but not here.
	b.	That is bigger than that.	(Braun, 1996: (1))
	c.	It wasn't this big, it was this big.
	d.	I can only do it like so, not like so.
	e.	I like you, but I don't like you.
	f.	Now I can hear it, but now I can’t.

This may be treated, for instance, by assuming that an expression is interpreted not relative to a context but to a sequence of contexts, or that a context contains not a single demonstratum, and a single addressee, etc, but sequences of them. My concern here is not to argue for a particular analysis of these facts but to point out that he and she again pattern with demonstratives here, while it does not (see also Wolter, 2006):

(12)		I like her, but I don't like her. [pointing first at one, then at another person]
(13)	*	I like it, but I don’t like it. [pointing first at one, then at another vase]

Again, rendering the referents strongly familiar does not help, even though we have seen in (9) that it, when Focused, can shift the Topic:

(14)		This is my dog, and this is my cat, and
	*	I like it, but I don’t like it.

The data in (13) and (14) confirm the conclusion from (7) that he and she, in selecting their referents, are sensitive to the act of demonstration in a way that it is not.

A final piece of evidence against a deficiency in it comes from (15):

(15)			[Pointing at a vase]
	a.	*	That's owner is quite rich.
	b.		Its owner is quite rich.

(15b) shows an exceptional syntactic context where it can be used demonstratively. It appears that its is the best spell-out here because the demonstrative pronouns this and that do not have possessive forms, as illustrated in (15a), a trait also found in some other languages, and also shared by inanimate wh-words, which are closely related.⁴ Perhaps, dem is subject to impoverishment in the context of poss. But whatever the explanation, if it had some inherent incompatibility with demonstration, the improvement in (15b) would be quite unexpected. I will return to the blocking relation between it and that in Sects. 5 and 6.

To summarize, we have seen that it differs from (s)he in that: (a) a referent for it cannot be introduced by demonstration (2); (b) it cannot select among multiple possible referents by demonstration (7); and (c) it is not affected by shifts in utterance context (13). I have also argued that these observations cannot be plausibly attributed to a deficiency in it. They can however be explained, as I will discuss in Section 4 below, by assuming that (s)he optionally spells out a demonstrative feature, dem, whereas it does not, and that it is the interpretation of this feature that allows demonstrating gestures to help determine the referent of a demonstrative.

3 Deferred reference

An additional distinction between it and other pronominals observed by Nunberg (1993) is that it does not allow “deferred reference,” a generalization of the deferred ostension of Quine (1968) which Nunberg explicates as follows. Consider (16):

(16)	[Marie, a Dutch linguist, speaking:]
	That's not how we do things.

The referent of indexicals like we depends on the context of utterance due to a “deictic component” to its meaning, which picks out an “index,” in this case the speaker, Marie. However, sometimes the referent of a deictic expression is not identical to its index; this is when the reference is “deferred.” On the other hand, the relation between index and referent is not unrestricted; e.g., the referent of we must include the index. Such restrictions result from a “relational component” to the semantic make-up of the pronoun. As a result, we in (16) may refer to the set of linguists, or Dutch people, or women, etc. Finally, there may be a “classificatory component” which further restricts the choice of referent (not index); this includes such features as plural on we.

As Nunberg observes, deferred reference is also found with demonstratives, including demonstratively used personal pronouns:⁵

(17)	[Pointing at a stable, to refer to the horse usually kept there:]
	That horse didn't do too well in the race.
(18)	[Pointing at a location in an architectural mock-up:]
	We can build the bomb-shelter here.
(19)	[Pointing at an exam, referring to the student who wrote it:]
	He didn't do too well.

In each of these cases it is from the remove between the demonstrating gesture and the understood referent that we can tell that there is deferred reference. Deferred reference is especially clear in cases where the gender or number features of the indexical, which apply to the referent, would not fit the index. Thus, (20) confirms that she allows deferred reference:

(20)	[Pointing at a male patient, referring to the female doctor who successfully treated him:]
	She did a fantastic job!

Turning to the issue at hand, demonstrating that it does not allow deferred reference is complicated. Given that it cannot be used demonstratively, we cannot check for deferred reference by observing a possible remove between the demonstration and the referent. So (21) is ruled out independently.

(21)		[Pointing at a cabinet, to refer to the rare earth usually kept there:]
	??	It was stolen during the night.		(cf. that (rare earth))

Furthermore, Nunberg claims that anaphorically used expressions do not allow deferred reference. If so, this would only leave “contextual use” as in (6) as a test case. Consider Nunberg’s examples in (22) (his (66)/(67)):

(22)			[Driving past an accident site:]		Nunberg (1993)
	a.		Gosh, it must have happened just a second ago.
	b.	?	It has happened a number of times on this stretch of highway.			(cf. that)

(22a) allows contextual use for it. But we cannot easily use the contextually salient accident as an index and then refer to accidents of that type in (22b); although this is fine with that in place of it.

Another example of this sort is given in (23).

(23)			[We’ve been looking at a gaudy display of the home team’s sports trophies:]
	a.		The fans love it.		[“contextual use”—compare (6b)]
	b.		The fans love this.
	c.	??	The fans usually love it.
	d.		The fans usually love this.

While this in (23d) can generalize over different sports teams’ trophy displays, such a reading is quite difficult to obtain in (23c). So examples (22) and (23) confirm that it does not allow deferred reference. However, if one supposes, as Nunberg does, that anaphora somehow blocks deferred reference, and also that the contextual use of it is a hidden case of anaphora (an option we considered in connection with (6b)), then these examples, like (21), might still already follow from other properties of it.

Fortunately, there are data suggesting that anaphoricity does not always block deferred reference.⁶ Consider (24). We’re at a (women’s) painting workshop where every participant makes exactly one painting, and vice versa.

(24)	a.		[Pointing at a canvas dripping with many layers of paint (the painter is not present):]
			Look at that! I wouldn’t want her working in my living room!
	b.		[Pointing at a painter covered in paint (the painting is not in sight):]
		??	Look at her! I wouldn’t want it hanging in my living room! (cf. that (painting))

Her in (24a) shows deferred reference. The demonstrating gesture and the referent coincide for that, but the subsequent occurrence of her apparently takes this referent as its index and, forced by the animacy requirement in its classificatory component, selects its own referent, the painter, in function of that index.⁷ We can now observe a clear contrast with (24b), confirming Nunberg’s assessment that it disallows deferred reference, while her allows it.⁸

Assuming on the basis of these admittedly sparse data that it does not allow deferred reference, how can this be explained? Nunberg (1993) does not provide an explanation. While he observes that expressions lacking a deictic component do not allow deferred reference, he does not explain why this is so, nor does he explain how it, in contrast to (s)he, comes to lack a deictic component in the first place, as discussed above. Postulating that it, for whatever reason, lacks a “relational component” will not work, since Nunberg’s relational component is not the source of the referential deferment but merely restricts its range. To solve this issue we will borrow a device from Elbourne (2008), where—technicalities aside—the function that maps the index to the referent (which he calls the relational component but which despite this name does not do the work of the relational component in Nunberg (1993), but that of the Referential Function RF of Nunberg (1979)—see also Sag (1993)) is the value of a free function variable, R, present in the internal syntax of pronouns and demonstratives. This will allow us to describe the difference between it and other pronominals in terms of the absence or presence of R, as shown in the next section.

To summarize this section, Nunberg’s (1992, 1993) deferred reference is another phenomenon where he, she pattern with other demonstratives, while it does not. This can be explained by assuming that he and she optionally spell out a demonstrative feature, the presence of which is a prerequisite for deferred reference. An additional result, as already observed by Nunberg (1993: 33), is that recognizing he_[dem] and she_[dem] as dedicated indexicals removes the sole counterexamples that stood in the way of identifying his “strong indexicals” (those expressions that are indicative and subject to deferred reference) with the category of dedicated indexicals.

4 The composition of demonstratives

In the previous sections we have seen that referent selection for it, unlike (s)he, is insensitive to demonstration, and that it probably does not allow deferred reference. In this section we show how this distinction can be described in terms of differences in the featural makeup of these pronominals if we assume that it is never a demonstrative, but (s)he can be.

We start by adopting Elbourne’s (2008) assumption that pronominals may contain a free function variable R that is responsible for deferred reference. His syntax is illustrated in (25) (Elbourne’s (48)).

(25)	[he [R1 i2 ]]

Here, i is a free variable whose value functions as Nunberg’s index; the function R applies to the index and yields (the set containing) the deferred referent(s); he functions as a regular definite determiner.

We cannot adopt Elbourne’s proposal as is, for two reasons. First, his treatment of the semantics of pronouns and demonstratives makes no reference to a demonstration or gesture in the context of utterance (i is just a free variable), so that it provides no descriptive means to distinguish expressions that are sensitive to gestures ((s)he) from ones that are not (it). In addition, R plays a crucial role in Elbourne’s treatment of bound anaphora. Since, like all pronominals, it can be bound, this would mean that it also contains R, hence should allow deferred reference. If we want to explain the data in the previous section we therefore cannot adopt Elbourne’s analysis of bound anaphora. On the other hand, Elbourne’s basic approach will allow us to describe the fact that only demonstratives allow deferred reference, namely by postulating that R only occurs with demonstratives. Naturally, in view of examples like (19), (20) and (24a) this will only work if we accept that demonstratively used he, she are also demonstratives.

Let’s adopt Elbourne’s syntax for demonstrative pronouns, but with the demonstrative feature, instead of a free variable i, functioning as the index:

(26)

[DP Det [RP R dem[distal] ]]

that

Assume that the abstract definite determiner Det can optionally select as its complement a phrase with head R, which in turn subcategorizes for a complement with the feature dem. The feature/value pair dem[distal] is spelled out as that, as discussed in the next section; possibly it surfaces in D⁰. The structure is interpreted as follows (I use the colon notation from Heim and Kratzer (1998) to flag a presupposition; I use δ_c to stand for the demonstratum (not the demonstration) in context c; and following Elbourne’s (2008: 413) notational convention I use the ι-operator for the presuppositional definite determiner meaning):

(27)	a.	〚dem[distal]e〛g,c	=	(λx:far_from_the_speaker(x).x)(δc)
	b.	〚R dem[distal] 〛g,c	=	g(R)((λx:far_from_the_speaker(x).x)(δc))
	c.	〚Det<et,e> 〛g,c	=	λP.ιxP(x)
	d.	〚Det [ R dem[distal]] 〛g,c	=	ιx.(g(R)(δc))(x) [presupposing δc is distal]

The assumptions here are that the feature dem simply refers to the demonstratum δ_c in the context of utterance, while the feature value distal adds the presupposition that δ_c is “far from the speaker” in some sense. The basic function of R is to lift δ_c, which is of type e, to type <e,t>, so that the abstract definite determiner Det can apply to it. As a result, that in (27d) refers to the unique object that has the contextually given R-relation with the demonstratum δ_c in the utterance context, under the presuppositions that δ_c is distal and that R yields a singleton. In principle, R can denote any contextually salient function (e.g., from a stable to the property of being a horse housed in that stable), yielding deferred reference.⁹ In the absence of an appropriate contextually salient function, R can default to Partee’s (1987) ident operator that lifts x to λy.y = x (cf. Elbourne, 2008: 423). In this case, the referent is the demonstratum, so reference is not deferred.

Any further features, such as number or gender, will be subfeatures of the determiner, so that the presuppositions they trigger apply to the referent of the DP, not to the index (cf. footnote 5). As for adnominal demonstratives, Elbourne proposes a slightly different syntax so as to accommodate the explicit NP component; I suspect it might be preferable to treat the NP as an appositive adjunct, but I will defer discussion of this option to another occasion.

While the feature complex dominated by DP in (26) is spelled out by that, a demonstrative with additional features animate (and fem), presumably in D⁰, will be spelled out as (s)he:

(28)

[DP Det[animate (fem)] [RP R dem[distal] ]]

(s)he

The interpretation will be the same as in (27), except for the added presuppositions of animacy (and female gender) applied to the output of the definite determiner. We can assume that any other value than distal will yield the same spell-out (as in Lander and Haegeman, 2016) or that a value neutral is available, an issue that arises more generally for demonstratives without a proximal/medial/distal distinction.

As for non-demonstrative pronominals, their precise treatment is not crucial to our argument. We may adopt the common assumption going back to Postal (1966) that the complement of Det is, not RP, but a zero NP_k, which denotes λx.x=k, where k is a free variable.

(29)	a.	[DP Det [NP ∅k ]]		it
	b.	[DP Det[animate (fem)] [NP ∅k ]]		(s)he

The variable k can be bound, or derive its value from the assignment function to yield an unbound anaphoric reading in the usual way. Depending on how we want to treat examples such as (6), we can assume in addition that the discourse referent k must be present in the preceding discourse. The important observation is that, absent the R variable, deferred reference cannot obtain, and absent the dem feature, the reference of the DP must be obtained without recourse to the demonstratum δ_c in the utterance context, in accordance with the data in Sects. 2 and 3. The relation between the two observations comes from the stipulated subcategorization of R for dem.

The reader will note that my account here is limited to those interpretive options that are relevant to the distinction between it and (s)he, namely the “exophoric” demonstrative use of demonstratives. Other uses of course exist as well, notably bound uses and uses as (donkey) anaphors. For the logic of the argument to go through it is enough to postulate that dem is a necessary condition for demonstrative use, and that (s)he can bear this feature whereas it can not. The fact that dem does not constitute a sufficient condition for such use, suggesting it may allow other interpretations than (27a), does not affect the argument directly. Obviously, however, a unified account of all uses of demonstratives would be preferred; I suspect that the bound uses and the (donkey) anaphoric uses may be obtained by allowing R in (27b) to yield a function from situations to individuals, along the lines of Elbourne’s (2005) treatment of donkey anaphora. The demonstratum in such cases then might be equated with the (linguistic) antecedent, as suggested by the analysis in Hinterwimmer (2019). I must leave these issues for further research.

Another difficult question I cannot resolve here is the following. We now understand why in selecting the referent for it, language cannot employ the mechanism stated in (27), which takes the formal demonstratum in the utterance context and calculates the referent from there. This mechanism is not available because it lacks the feature dem that triggers it. Nonetheless, in situations like (7) where several possible referents for a pronoun are available, we expect a cooperative hearer to take all available information into account in trying to resolve its reference, just as in resolving a case of lexical ambiguity the hearer can be expected to apply all their knowledge of the world and the context in making a best guess at the intended reading. Why then would the speaker gesturing at an object while using the pronoun it not help the hearer decide that the intended referent might very well be the object the speaker is gesturing at?¹⁰ While this is obviously not the place to develop a theory of the psychology of ambiguity resolution that will answer this question, it is important to observe that the pronoun it is not alone in not being sensitive to gestures in this way. As discussed in Sect. 6 below, the same holds for the contextual expression local and the indexical now. One interesting avenue to explore is suggested by Terenghi’s (2022) take on the Fregean incompleteness of demonstratives (Frege, 1918; see Salmon, 2018 for recent discussion). Terenghi proposes that demonstrating gestures (however abstractly realized) are parts of the spell-out of demonstratives, hence part of the utterance itself, rather than parts of the context of utterance. If so, demonstrating gestures that accompany an expression which does not have gestures in its PF representation will come out as simply uninterpretable. Again, I must leave an elaboration of this solution for further research.

A final question raised by the proposal in this section is why the demonstrative structure in (28) and the non-demonstrative structure in (29b) can both surface as (s)␣he. Is this an irreducible lexical ambiguity? The next section shows that it is not.

5 But isn’t homonymy an inelegant solution?

At first glance, the homonymy approach to the demonstrative use of pronominals appears less parsimonious than a single-item alternative. In this section, I will argue that the opposite is true. First of all, when we adopt the standpoint of Distributive Morphology (DM; Halle and Marantz, 1993) that terminal nodes at the syntax-PF interface are abstract feature complexes (“morphemes”) that are then spelled out by the insertion of vocabulary items, the homonymy approach does not require multiple lexical entries for (s)he. We can assume instead that the relevant vocabulary items are specified as shown in the vocabulary insertion rules in (30), the ordering of which I will leave as a stipulation here (φ, which I left out of the discussion in Sect. 4, stands for the feature that distinguishes pronominals from adnominals in Cowper and Hall (2009); I leave out additional features necessary for a complete treatment of the pronominal system but not relevant here):

(30)	she	↔	[ φ, animate, fem ]
	he	↔	[ φ, animate ]
	that	↔	[ dem ]
	it	↔	[ φ ]
	the	↔	[ d ]

In the DM framework, a vocabulary item may be used to spell out an abstract feature set iff that feature set is a superset of the features that the vocabulary item is specified for (Halle's, 1997 Subset Principle). Hence, since we specify the vocabulary item she as [φ, animate, fem], it may be used to spell out the morpheme [φ, animate, fem] (the personal pronoun in (29b)) as well as the morpheme [φ, animate, fem, dem] (the demonstrative pronoun in (28))—and likewise for he. Thus, what appeared to be a case of irreducible homonymy can be treated as a simple case of syncretism, explained by the underspecification of (s)he for the demonstrative feature. As for the other items in (30), since (s)he require animate, inanimate demonstratives are spelled out that, as do dem morphemes lacking the φ (pronominal) feature, leading to that man; but dem morphemes specified for animate may not be spelled out as that, as this is blocked by he and she (see below), Finally, an inanimate [dem] pronominal comes out as that rather than it, by the ordering in (30)—whether this ordering can be derived from other considerations is a question I will leave open here (one technical solution would be to rely on the fact that that is more specific, since it spells out distal in addition to dem).

It turns out that this approach is in fact not less, but more elegant than the theory that results if we insist that he and she are always personal pronouns. The argument comes from the structure of the pronominal and demonstrative paradigms in English. We start by observing that this and that, in their adnominal use, are compatible with animates and inanimates alike (see (31)), whereas in their pronominal use they can only refer to inanimates, as (33) shows.¹¹

(31)	a.		[Pointing at a vase:]
			I bought that vase yesterday.
	b.		[Pointing at a man:]
			I met that man yesterday.
(32)			[Pointing at a vase:]
			I bought that yesterday.
(33)			[Pointing at a man:]
	a.		I met him yesterday.
	b.	*	I met that yesterday.

This means that, if we were to suppose that he is not a demonstrative pronoun, the English demonstrative pronominal paradigm would contain a curious gap in not providing a [+animate] item. I submit that this apparent gap is actually filled by he and she. This also explains the otherwise puzzling fact that that does not spread into the supposedly empty [+animate] slot, despite being compatible with animates.

While this argument is independent of the DM approach, we can place it on a more technical footing by considering the possible feature complexes that one might suppose make up the pronominal and adnominal paradigms in this framework (see, e.g., Harley and Ritter, 2002; Cowper and Hall, 2009).

Suppose that English indeed allows a morpheme that combines the feature dem with whatever feature setting signals +animacy (either the presence of a feature animate, or the absence of a privative feature inanimate). If so, it is clear that this morpheme is spelled out by (s)he, yielding (33a). Otherwise, (33b) (or some other alternative spell-out) would be acceptable. We must accept, therefore, that (s)he can be demonstratives, unless we assume that the feature combination dem + animate cannot occur.

Note, incidentally, that this argument is separate from the question of how to ensure that the animate demonstrative pronoun, if it exists as I propose, is spelled out as (s)he, not that. In view of (31), we do not want to suppose that that is specified for [inanimate], but neither do we want to postulate two vocabulary items that, a pronominal inanimate one and an adnominal one unspecified for animacy. The obvious solution is that in (33) he, which expresses [animate], blocks that, which does not. But this is not automatic: why doesn’t that, which expresses [dem], block he, which does not? Several solutions are possible. We could employ an extrinsic ordering among vocabulary insertion rules, as in (30) (Halle and Marantz, 1993), or a preference for spelling out the feature [animate] over spelling out [dem], as considered in Wolter (2006). Or we can postulate an impoverishment rule (see Bonet, 1991; Halle and Marantz, 1993) that deletes the feature [dem] in the context of [animate], in the pronominal context. Then (33) contains the impoverished bundle [animate], which can only be spelled out by he. But whatever the solution, accepting the existence of animate demonstrative pronouns implies that (s)he spell it out.

Returning to the issue at hand, could it be supposed that the dem feature is incompatible with animacy, rendering the animate demonstrative slot in the paradigm uninhabitable? It turns out that this would lead to a significantly less elegant grammar. First of all, the incompatibility of the dem feature with animacy would need to be limited to the dem feature as it occurs in pronominals. Otherwise, we could not account for (31b). So we would need to postulate two demonstrativity features: a pronominal [dem₁] in (32)/(33), which is not compatible with animacy, and an adnominal [dem₂] in (31), which is. We then face the accidental fact that both [dem₁] and [dem₂] are spelled out by that. So we need a lexical ambiguity for that, with one that specified for [dem₁], and the other for [dem₂], and likewise for their variants this, these, those. So our attempt to avoid a homonymy analysis of (s)he has actually made things worse, as the required ambiguity of that etc. is irreducible and cannot be treated with underspecification.¹²

In addition, technical issues would arise in implementing the supposed animacy gap in the pronominal demonstrative. If we were to assume there is a (privative) feature [inanimate], we can simply stipulate that [dem₁] entails [inanimate] in the (English) feature geometry. However, it is usually assumed (e.g., Cowper and Hall, 2009; Harley and Ritter, 2002) that [animate] is the marked feature—in which case we need to block the combination [animate, dem₁, …], allowing only [animate, …] and [dem₁, …]. We cannot achieve this via entailment between features, and it is not obvious how we can. Furthermore, a puzzle now arises as to why [dem₁, …], spelled out that, cannot have an animate referent, even though [animate, …], spelled out he, can have a demonstrated referent. If absence of [animate] in [dem₁, …] entails inanimacy, why doesn’t absence of [dem₁] in [animate, …] entail absence of demonstrativity? Put differently, why is it that in the situation (33), where an animate entity is being demonstrated, [animate] is a better fit than [dem₁]? Ordering of VI rules, or impoverishment, will not help here, since we are dealing with a competition among feature complexes given the intended referent, not a competition among vocabulary items given a feature complex.

To complicate matters further, whichever stipulations we adopt so as to preclude the animate demonstrative pronoun must be specific to English and cannot be allowed to affect the universal pronominal feature geometry, since, as we expect, languages exist that do allow animate demonstrative pronominals, distinguishable from personal pronouns (for instance, German der, Dutch deze, Spanish aquel, Rovigo Italian custú/colú (Ledgeway, 2015)).

Considering, finally, that—as far as I am aware—there is no positive motivation for maintaining that English lacks animate demonstrative pronouns, I conclude that the assumption that this is so is unwarranted, and leads to serious complications in the grammar. Supposing instead that animate demonstrative pronouns do exist, and are spelled out by (s)he, on the other hand, only leads to a minor complication (the possible need for an impoverishment rule mentioned earlier); it does not force us to manufacture an empty slot in the demonstrative paradigm, and it explains why pronominal that does not spread to animate contexts, while adnominal that does.

6 Why blocking is not enough by itself

I have argued in Sect. 2 that he and she can spell out pronominal demonstratives. This explains why they differ in usage from it, which is always a personal pronoun. Presumably, it cannot spell out a demonstrative pronoun because this spell-out is blocked by that, as stated in (30). Now the reader may wonder if we cannot explain the difference between it and (s)he directly on the basis of blocking. The reasoning might go as follows. He and she are always personal pronouns, which may however be “used demonstratively.” The reason why it may not be used demonstratively is that this usage is blocked by that, because that explicitly signals demonstrativity. The purpose of this final section is to explain why this simplified approach will not work.

It is important to realize that we are no longer talking here about blocking at the level of spell-out, where a given morphosyntactic representation needs to be spelled out phonologically, and one spell-out blocks another, as in our proposal, where that blocks it as the preferred spell-out given an underlying morphosyntactic representation containing an inanimate pronominal with a dem feature.¹³

We are looking, rather, at blocking at the level of underlying representations, where given the context and the speaker’s intentions, one morphosyntactic representation blocks another. An example is blocking of the indefinite article by the definite article in contexts that satisfy the latter’s presuppositions, under Heim's (1991) Maximize Presupposition (MP). We would need to suppose that pronominal structures without the [dem] feature can in principle be used demonstratively, but that given a context where the speaker intends to refer to the demonstratum, a pronominal without [dem] is sometimes blocked by a pronominal with [dem], because of a preference for using a pronominal with [dem] when possible. Then dem (spelled out that) would block absence of dem (spelled out it) in “demonstrative contexts.” One possible implementation would be to postulate that the feature [dem] triggers the presupposition that the referent is being demonstrated in the context of utterance,¹⁴ in combination with MP.

The first problem with this approach is that it only targets the referent selection data from Sect. 2. It does not explain why it does not allow deferred reference (while he and she do), as discussed in Sect. 3. It also does not deal with the problems discussed in Sect. 4: maintaining that he and she are not demonstratives still creates a difficult to explain gap in the paradigm. Furthermore, to explain why that does not block (s)he, in demonstrative animate contexts, we still need to postulate either that that (but only pronominal that) is marked as [inanimate], or that the language prefers signaling or marking animacy over signaling demonstrativity, which becomes even harder to understand if both are regulated by Maximize Presupposition, applying equally to [animate] and to [dem]. In fact, considering that we are now postulating a mechanism that favors the use of one piece of syntactic structure over another given a particular referential intention, we fail to understand why he, in an animate masculine demonstrative context, is not blocked by that man, which both marks demonstrativity and triggers the presupposition that the referent is masculine and animate.

Another difficulty attending the blocking approach is that it and that are not always in complementary distribution, as is well known. Consider first (34):

(34)	[Pointing at a mound of white powder:]
	Look at that stash of coke! That/it must be worth a fortune!

To explain such cases, one could rely on the fact that there is some indeterminacy in how speakers and hearers construe the context. However, too much indeterminacy in, e.g., whether a demonstrating gesture has already made an object salient before a pronoun is used to refer to it, will upset the explanation for cases like (7), where it is completely blocked.

Consider next (35):

(35)	[Pointing at a vase...:]
	I bought this in Hong Kong. It was very nice there, by the way. Great food!
	[...again pointing at the vase:]
	So, do you like it?

(35) shows clearly that there is no requirement to explicitly signal a demonstrating gesture by employing a pronominal with [dem], so the blocking approach fails.

What seems to be the essential difference between (35), which allows both this and it, and examples like (7), where it is blocked, is that, while in both cases the referent is being demonstrated, it is only in (7) that the demonstration plays a crucial role in resolving the reference. If so, the obligatory use of the demonstrative pronoun in (7) does not reflect a property of the referent that must be expressed because of Maximize Presupposition, such as the property of being demonstrated. It reflects a choice of referential strategy: even though the vase in (35) is being demonstrated, hence may well be the demonstratum δ_c, the speaker chooses not to refer to it as δ_c, but via whatever mechanism is available for personal pronouns (plausibly, identification with an established discourse referent). In (7), this mechanism will not serve due to the multitude of available discourse referents, so that the demonstrative strategy must be chosen. If the referential strategies are so fundamentally different, that is another reason why for items that allow both, such as (s)he, the homonymy analysis proposed here would be more plausible than different uses of the same lexical item.¹⁵

Finally, the blocking approach would fail to account for those indexicals which—like it—are not sensitive to demonstration, but for which no blocking alternative exists. I will discuss two very different cases, namely local and now. For the former case I again profit from Nunberg (1993: 35), who observes that while elements such as local (cf. Partee, 1989) are “contextual,” they disallow deferred reference (because they are not “strong indexicals”). He adds, albeit without providing the relevant examples, that local is not sensitive to demonstration. Consider (36):

(36)	a.		[Walking through a crowded neighborhood:]
			The local shopkeepers must be doing really well.
	b.		[Pointing first at one, then at another neighborhood (say, from an airplane):]
		*	The local shopkeepers must be doing really well, but the local ones must be doing poorly.
	c.		Here, the local shopkeepers must be doing really well, but here, the local ones must be doing poorly.

(36a) shows that local is indexical. But it patterns with it in (13) rather than that and (s)he in (11) and (12), in contexts like (36b) where demonstration is required for referent resolution; it requires anchoring to a separate demonstrative as in (36c). Local must miss some feature (the [dem] feature) that would render it sensitive to demonstration, as we cannot attribute (36b) to blocking by some item local' that overtly marks the demonstration.

Now, on the other hand, is a paradigm case of a pure indexical.¹⁶ Nunberg provides examples demonstrating that now allows deferred reference, as expected from a strong indexical (see Nunberg, 1992: (34)), but he does not discuss the fact that, like local, now, too is insensitive to demonstration.¹⁷ It is, of course, a defining property of pure indexicals—as opposed to true demonstratives—that their content in a context of utterance does not depend either on demonstration or on the speaker’s intentions. However, actually demonstrating that now is a pure indexical in this sense and is positively insensitive to demonstration is another matter.

As Kaplan (1989a: 491) observes, the locative pure indexical here has a true demonstrative use, or counterpart, which occurs e.g. in In two weeks, I will be here [while pointing at a city on a map]. Temporal then in (37b) allows a similar usage. But now in (37a) does not:

(37)			[Pointing first at one, then at another spot on the calendar:]
	a.	*	Now I was happy, but now I wasn’t.
	b.		Then I was happy, but then I wasn’t.

Nonetheless, one might maintain that now is special, not in being positively insensitive to demonstration, but in having such a restrictive Kaplanian character that on its indexical use its content is already fully determined as identical to the time of utterance, simply leaving no room for the demonstration to further affect its reference, in the same way that adding demonstrating gestures to an utterance of I is simply “irrelevant” (Kaplan, 1989a).

However, as observed by e.g. Kaplan (1989a: 491, fn. 12), Nunberg (1993: 29) and Perry (1997, 2017), while now must pick out a temporal interval that includes the speech time, the size of this interval is not fixed by the conventional meaning (for Perry, this is grounds for not (fully) considering now a pure (or ‘automatic’) indexical). In principle, we might therefore expect that a demonstration could contribute to narrowing down the understood time interval, by conveying the speaker’s intentions. But we can observe that this is not the case. The gesturing in (38) does not convey what the modification with right manages to convey in (39).

(38)	#	Now [pointing at the year on the calendar] I’m working at Gooperman & Fitch, but now [pointing at the week on the calendar] I’m on leave.
(39)		I’m working at Gooperman & Fitch now, but right now I’m on leave.

On the other hand, a similar specification of the size of the intended space or temporal interval can be achieved for then and here in (40):

(40)	a.	Then [pointing at a year on the calendar] I was working at Gooperman & Fitch, but then [pointing at a week in that year on the calendar] I was on leave.
	b.	Here [pointing at the US on the map] I usually feel safe, except here [pointing at Alabama] I don’t.

The conclusion, again, is that some indexical expressions are insensitive to demonstration, even though they are not blocked by an overtly demonstrative counterpart. A blocking approach to it, besides having the drawbacks mentioned earlier, would therefore never be more than a partial solution. On a more general note, observe that it, local and now do not form a natural class. In particular, local is a merely contextual expression, in Nunberg’s (1993) classification, and is not directly referential (‘indicative’); now is its opposite as a dedicated and pure indexical; it, depending on the treatment of (6b), may not be an indexical at all. It is therefore less likely that we will succeed in deriving their insensitivity to demonstration from a shared property that distinguishes them from other indexicals (aside from the fact that in Sect. 2 we failed to nail down what that property might be in the case of it). The more plausible approach is to attribute a shared property, namely the feature [dem] and its semantics, to those expressions that, like he and she and that, are sensitive to demonstration.

7 Conclusion

Starting from the observations in Nunberg (1993), I have demonstrated that the pronoun it differs both from the pronouns he and she and from demonstratives like that, in that referent selection for it is not sensitive to demonstrating gestures in the context of utterance, and in that it disallows deferred reference. Having dismissed a range of possible alternative explanations, I have proposed that he and she sometimes spell out the feature that also characterizes demonstratives; only it never spells out this feature. I have sketched an account of the internal structure of demonstrative pronominals, based on Elbourne (2008), in which this demonstrative feature semantically links their interpretation to the demonstratum in the context of utterance, and is in turn selected by the functional head responsible for deferred reference. I have also shown how a Distributed Morphology based underspecification account allows us to describe that he and she are sometimes, but not always demonstratives, without the need for a duplication of lexical entries. In addition, I have argued that the structure of the pronominal paradigm favors the hypothesis that he and she can be demonstratives. Finally, I have argued that a possible alternative account of the central data in terms of blocking is unsuccessful, partly by identifying other indexicals that are insensitive to demonstrating gestures despite not having a blocking alternative. How the proposed semantics relates to the semantics of bound and anaphoric pronominals is a question left for further research.