The impersonal gets personal

Abstract

Multicultural London English (MLE) has developed a new pronoun, man, with a number of interesting properties. It can be interpreted as any person and number combination (1sg, 1pl, 2sg, 2pl, 3sg, 3pl), and at first blush appears the allow a generic impersonal reading. One central strand of analyses of the ‘generic’ or ‘quasi-universal’ interpretation of impersonal pronouns involves treating them as featurally impoverished pronouns which act as variables bound by a generic operator either high in the clause (Moltmann 2006; Sigurðsson and Egerland 2009, a.o.) or at the top of the DP (Ackema and Neeleman 2018). What is interesting about MLE man is that it seems to allow a generic reading while being generally resistant to binding: the pronoun never behaves as a bound variable, and so it could not reasonably be assumed that a generic operator can bind it. I show, however, that a closer look at the facts reveals that the quasi-generic interpretation, and all other possible interpretations of the pronoun, can be explained instead by assuming that it has a featureless person head, which introduces the full lattice of possible referents including speaker and addressee (following Harbour 2016), and then allowing contextually determined subsets of that full lattice to be picked out by a choice function, modelled on the epsilon operator of von Heusinger (2004). This novel data from MLE suggests that generic-like interpretations can arise even where generic binding is not possible, but that the traditional strategy of generating these interpretations are still needed to capture the full typology of impersonal pronouns.

1 Introduction

Innovations in languages are not only interesting from a historical and a sociolinguistic perspective, where questions of social meaning, networks, indexicality, contact and acquisition are all central and important, but also from a formal syntactic and semantic perspective. New grammatical forms often provide a gold mine of data which can reveal the limits of variation in feature inventories, constraints on the computational system of syntax, and on the mapping from syntax to semantics and phonology. Analyses of innovations can also occasionally help to tighten up existing theories, or adjudicate between competing theories.

In this paper I consider one innovation, the pronominal form man in Multicultural London English (MLE), which I believe is important for all of the above reasons. I show how its distribution and range of interpretations gives us an insight into the nature of pronouns in general, and more specifically how at first glance it suggests a rethink to existing approaches to impersonal pronouns, but ultimately fits quite naturally into current theories of grammar.

There is a rich literature on a variety of impersonal pronominal forms across languages, the most widely studied of these being Germanic and Romance impersonal forms. Many existing analyses of the generic interpretation of impersonal pronouns have in common that they treat the pronouns as featurally deficient or underspecified in the syntax (Sigurðsson and Egerland 2009; Hoekstra 2010; Fenger 2015; Holmberg and Phimsawat 2016; Ackema and Neeleman 2016), and that this deficiency means that their interpretation is not constrained by phi-features, but instead is imposed externally, be it through agreement with some head in the clause (e.g., D’Alessandro and Alexiadou 2002), or binding by a quantificational operator (e.g., Moltmann 2006; Malamud 2012). I focus in this paper mainly on the latter kind of analysis involving binding, which I briefly introduce here.

An early proposal along those lines was put forward in Chierchia (1995).¹ Putting aside the specifics of the analysis, Chierchia argues that generic readings of Italian si are the result of a variable being bound by a generic operator at the edge of VP, in a way analogous to that of indefinite generics. This means that si canta, ‘one sings’, in a generic context, has the (very simplified) logical form of (1).

Here Gn is “a null universal adverb with a special modal quality (which enables it to tolerate exceptions)” (1995:111). More recent treatments of impersonal pronouns across languages have followed in this tradition, and suggested that impersonal pronouns are themselves underspecified pronouns, which act simply as variables bound by a generic operator. Moltmann (2006), Sigurðsson and Egerland (2009), Malamud (2012), Roberts (2015), Fenger (2015), Holmberg and Phimsawat (2016), and Ackema and Neeleman (2018) all feature a version of this story, although Ackema and Neeleman have the generic operator merged inside the DP, while the others all feature a generic operator high in the clause.²

In this paper I show that a “feature deficient” approach to a newly innovated pronominal man in MLE is appropriate, but argue that a generic binding analysis for the impersonal reading associated with the pronoun is not. Although MLE man exhibits some of the characteristics of an impersonal pronoun, in that it can have what looks like a generic reading, I show that it cannot act as a bound variable in a number of environments. Thus I conclude that it cannot get its generic interpretation through binding, although ultimately I argue that this “generic” interpretation is not quite the same as that associated with other impersonal pronouns. I also show that MLE man has the peculiar property of being able to contextually pick out referents in an almost unconstrained way: it can act as a personal pronoun of any person and feature combination, given the appropriate context. I propose that these two characteristics of MLE man, its resistance to binding and its ability to either get a pseudo-generic interpretation, or a definite personal interpretation of any person and number, can be captured if we assume (following most of the literature) that the pronoun lacks person features, but that its interpretation is determined through the obligatory projection of a definiteness head, which picks out a salient individual from the set of potential referents, in the spirit of von Heusinger’s (2004) treatment of definiteness.³

The paper is structured as follows: in Sect. 2, I introduce the properties of MLE man, and elaborate on the different interpretations that the pronoun can have; in Sect. 3, I use agreement facts across two different groups of speakers of MLE man to show that the pronoun carries no person features, although it may have number features; in Sect. 4, I show that MLE man cannot be bound; in Sect. 5, I propose an analysis of the properties of MLE man which eschews a generic operator approach, and instead treats MLE man as a pronominal form with no ϕ-features that obligatorily projects a D head.

2 A new (im)personal pronoun in MLE

Cheshire (2013) documents the grammaticalization of a new pronominal form, man, in what is termed ‘Multicultural London English’: that is, the English spoken by young, typically working-class Londoners who have grown up in and live in a multicultural and multilingual environment.^4,5 Cheshire suggests that man is a general pronominal form, which developed from plural use of a bare noun, and that it is currently in the process of becoming grammaticalized as a first person singular personal pronoun. The suggestion is that an original bare (number neutral) form man ultimately developed into the pronominal form, and that this bare form was likely adopted from Jamaican Creole man, which can also take on something like an impersonal use in certain contexts, at least insofar as it has a kind of generic force (examples from Cheshire 2013):

The bare noun form man which can be used simply to mean ‘people’/‘men’ also still exists in MLE. It can be modified by quantificational elements such as many, bare (‘a lot (of)’), or couple, by numerals, and by adjectives, and cannot give rise to 3sg agreement as a subject. Some examples are given in (4).⁶

In its pronominal use, no modification is allowed, 3sg agreement is possible,⁷ and a wide range of interpretations are available according to context, including speaker reference.⁸

The pronominal form is particularly interesting because of the wide range of interpretations that it can have. In the next subsection, I lay out some examples of those different interpretations, and clarify some syntactic properties of the pronoun.⁹

2.1 Interpreting man

Cheshire provides a summary of some of the properties of man, derived from 11 tokens of the pronominal form extracted from a set of recordings from two projects on working-class London English, the Linguistic Innovators Project,¹⁰ and the Multicultural London English Project,¹¹ alongside another set of 83 tokens (total) from two documentaries (One Mile Away and Giggs Biography) and a film (Anuvahood). The focus is on the grammatical role of man (subject, object, possessive), and the interpretation of the pronoun in terms of person and number (1sg, 2sg, 3sg, 1pl, impersonal/generic). Grammatical role is summarised in the following table (adapted from Cheshire 2013):¹²

Man can appear in subject, object, and possessive position (as man’s), but appears to be favoured as a subject in each data-set.

Interpretation is summarised in the following table (also adapted from Cheshire 2013; see fn. 12):

Here we see that the pronoun can have a 1st, 2nd and 3rd person interpretation, as well as an impersonal interpretation, but 1st person singular interpretation appears to be favoured (although this effect is mainly due to the massive preference for man being used in the first person in Anuvahood). Cheshire suggests that the high frequency of use with a first person singular meaning is an indication that the form is in the process of being grammaticalized as a 1sg pronoun.¹³

What is not gleaned from the corpus, but is clear from further examples, is that man can have an even broader range of interpretations associated with it: in fact, all person and number combinations, along with an impersonal reading, are possible, given the right context. The fact that 2pl and 3pl uses do not show up in the data set from Cheshire (2013) does not mean that they are categorically excluded. In the following sections, I give the details of the different readings available, and expand on the syntactico-semantic properties of each. I begin with a discussion of impersonal readings, and then move on to the definite personal readings. I follow up with a discussion of verb agreement, and then expand on the final property of man which is central to my argument, the fact that it cannot act as a bound variable.

2.2 Impersonal readings

Before describing the nature of the impersonal use of MLE man, it is necessary to clarify some facts about impersonal pronouns in general. First, I want to clarify that by “impersonal pronoun” I mean the set of “dedicated impersonal pronouns” (Fenger 2015), otherwise called “man-constructions” (Siewierska 2011) that have been described in the literature, which are dedicated pronouns etymologically related to a noun meaning ‘man’ or ‘person’ or ‘one.’ I also include the kind of interpretation available with impersonal uses of otherwise personal pronouns (such as the impersonal use of 2sgyou in English). I will throughout refer to dedicated impersonal pronouns simply as “impersonals.” There is quite a rich literature on the topic, and unfortunately the terminology employed therein is not uniform, and neither is the focus of inquiry. What is clear in the literature, however, is that there are a number of different interpretations that impersonal pronouns can take on, and that different pronouns’ interpretative properties are constrained by what look like crosslinguistically quite stable tendencies which are good targets for generalisations. Putting aside terminological differences, the literature has identified the following different interpretations, each of which are associated with at least some subset of impersonal pronouns:

• Obligatorily inclusive generic reading: refers “quasi-universally”¹⁴ to a group that must include the speaker (and potentially the addressee).

• Optionally inclusive generic reading: refers “quasi-universally” to a group that need not include the speaker, but can.

• Definite personal reading:¹⁵ refers to a specific (atomic or plural) individual, in the way that a personal pronoun normally does.

• Arbitrary reading: refers “quasi-existentially” to some group/individual (which typically excludes the speaker).

Impersonal pronouns across languages allow for a generic interpretation, where the pronoun refers to a “quasi-universal” set of relevant humans. Take the following Swedish example (Egerland 2003):

It holds generally of all people that they have to work if they want to earn a living,¹⁶ although the generic use of the impersonal here of course allows for certain exceptions in the way that generic DPs normally do.¹⁷ It is this generic reading of impersonal pronouns that has previously been analysed as resulting from the binding of a variable by a generic operator, and thus it is this reading that is of interest to us here.

Research on impersonal pronouns, especially in Germanic and Romance languages, has revealed that different impersonals in different languages can be categorized into two main groups: those pronouns that allow ‘optionally inclusive’ generic interpretations, and those pronouns that take an ‘obligatorily inclusive’ generic interpretation (Hoekstra 2010). To illustrate the two, I take examples of impersonal pronouns from Frisian and German (from Hoekstra 2010, slightly modified).

The examples in (11) show that both Frisian and German impersonal pronouns can have a reading where the speaker is potentially included in the set of referents: the fact that one needs a car to live in the countryside extends to the speaker too. The examples in (12), on the other hand, show that in Frisian, the impersonal pronoun obligatorily includes the speaker: reference to some generic group of others is impossible, unlike German man. Frisian impersonal men is therefore said to be obligatorily inclusive, while German man is said to be optionally inclusive.¹⁸ Impersonals across Germanic and Romance can generally be placed in one of these two categories:¹⁹

Beyond the quasi-universal generic reading, Germanic impersonals also exhibit an existential arbitrary reading. Take German man, for example. Malamud (2012) gives the contrasting examples in (14) to show the various interpretations available.

The kind of episodic existential interpretation exhibited in (14b) is not available for all impersonals, however, as similar examples involving English one show (my judgment, see also discussion in Malamud 2012:5–6):

The argument in Moltmann (2006) and Malamud (2012) is that one must be quantified over by a sentential operator of some kind, and this is because of a generic feature on the pronoun, which needs to be checked. The lack of such a checking relation in the existential case rules out the kind of reading exhibited in (14b) with one.

MLE man exhibits a similar variety of impersonal properties to one, although there are crucial differences which will feed into my ultimate analysis of MLE man. First, it seems that it can have a generic impersonal interpretation (16a,b,c), seems to be able to have both a speaker inclusive and exclusive interpretation (16d), and resists an episodic arbitrary interpretation (16e).

Examples (16a,b,c) describe general restrictions that apply to any individual, including the speaker if relevant. (16a) means any person who wants to do well has to work hard, and (16b) means that any person who wants to go to that restaurant has to wear a tie, and (16c) mean that any person who wants to hit the tall friend has to jump. (16d) means that all (relevant) Korean people eat insects.²⁰ However, (16e) cannot mean that some individual burned down a house: instead it can only be interpreted as meaning that I, you, or some other specific individual burned down a house, and thus gets a definite personal interpretation.²¹

Although something like a generic interpretation appears possible, such an interpretation is not possible with Q-adverbs in out-of-the-blue contexts. This is unusual, since quantificational binding effects with Q-adverbs is one standard test for genericity (Krifka et al. 1995).

For the time being I put aside this fact, but come back to discuss it in detail in Sect. 4.6.

2.3 Definite personal readings

Alongside the impersonal generic reading, MLE man can also have a definite personal interpretation, where a specific referent is picked out, identified through contextual cues. The interpretation can be 1st, 2nd or 3rd person, and either singular or plural. Examples of different types of definite use are given in (18), including some examples from outside of the MLE corpus:^22,23

This characteristic of MLE man is particularly striking. The intended referent is understood from context alone, and no morphological or phonological change takes place when man refers to different persons. No c-commanding linguistic antecedent is required, as shown by (18c,d,e). So long as there is a readily identifiable salient individual, interpretation is possible. (18c) is particularly interesting in this regard, as there isn’t a discourse antecedent at all: it is a reporting of a telephone conversation, where this is the first utterance of the speaker. Man in this case picks out the addressee because it could only reasonably be referring to the addressee, given that there is no other salient individual apart from the speaker, and asking the question of oneself (the speaker) would be pragmatically odd in the situation.

Some other impersonal pronouns across Germanic and Romance also have definite readings, but this is usually restricted to the first person, and occasionally the 3rd person. French on can have 1pl interpretation (Cinque 1988), Icelandic , can have a 1sg interpretation, and less clearly a 1pl interpretation (Sigurðsson and Egerland 2009), Swedish man can have a 1sg, 1pl and in some cases a 3sg interpretation, and Frisian men can have a 1sg interpretation (Hoekstra 2010). The ability of MLE man to take on any person and number combination will be central to my analysis, and I return to an in depth discussion of the pronoun’s flexibility in interpretation in Sect. 5.

2.4 Switching reference

An interesting effect of the context dependence of MLE man’s interpretation is that the referent of each use of man can be different within a single sentence, as long as context allows. (19a,b) are two such examples of reference shifting dependent on context. It is also possible for man to pick out a referent which is then later referred back to with a normal personal pronoun: (19c) is an example of this.

This shows how central contextual salience is to the interpretation of MLE man. There are of course some constraints on when (and in which configurations) it can pick out a particular referent. As we will see in more detail in Sect. 4.2, man cannot act as a local domain form (in the sense of Déchaine and Wiltschko 2010), and thus cannot be coreferential with an antecedent in its binding domain. Take for example (19a): the first instance of man cannot be coreferential with he, and the second occurrence cannot be coreferential with I, which limits the set of available interpretations. Furthermore, the introduction of a discourse referent by he means that it is the most salient individual available as a referent for the second occurrence of man. Thus we arrive at the appropriate reading where he and the second instance of man are coreferential.

3 Agreement and person features

Now that we have seen that definite personal readings range across all combinations of person and number, the question immediately arises whether this is the result of a lack of specification for ϕ-features, or a different specification for each interpretation with no overt morphological reflex. To tease apart the difference, we can look to subject agreement paradigms with the verb across speakers. This can be a little difficult to do, since English lacks rich agreement morphology, and it is especially difficult since there is a general dialectal levelling phenomenon which turns past tense be into either was or were across the board for MLE speakers (Cheshire et al. 2011). Present tense be, however, does still have an agreement paradigm identical to Standard British English.

As most of the examples of the use of MLE man as a subject above show, 3sg agreement on the verb is standard, but non-3sg agreement also seems possible for some speakers (21).

I suggest here that the difference can be put down to there being two different dialects, and that agreement possibilities absent from both dialects give us an insight into the internal make-up of the pronoun. Dialect 1 has 3sg agreement across the board, regardless of interpretation, and Dialect 2 has 3sg agreement for singular interpretation, and what appears to be 3pl agreement for plural interpretations. Taking both dialects together, in the present tense BE paradigm, is and are are the only possible forms, but am is never possible in either dialect, even with 1sg interpretation. Are is not possible in either dialect with 2sg interpretation, showing that it really is number that is doing the work in forms that surface as are. In summary, collapsing both dialects, we have the following possibilities:

These facts show that person is never relevant for agreement, and this can be put down to the lack of person features in the pronoun. I adopt the position here that morphological 3sg agreement represents default agreement (Malamud 2012), and that a [+atomic,−participant] specification is not required to give rise to 3sg agreement on the verb. Some support for this comes from the behaviour of expletive subjects. Ackema and Neeleman (2018), in their discussion of 3sg agreement on impersonal pronouns, point out that 3sg agreement occurs in English with expletive it subjects, and that this must constitute some kind of default agreement:

It could not be the case that it here includes a [−participant] person specification: if it did, then it would have to pick out some actual individual other than the speaker or addressee from the domain of discourse, although it does not. This means that even in the absence of any person information, 3sg agreement appears to be the default.²⁴

Since agreement is sensitive to number in Dialect 2, then it must be the case that a number head projects for those speakers, with a [−atomic] feature giving rise to plural agreement on the verb. The following insertion rules for present tense BE capture the full agreement paradigm across the two dialects:²⁵

A piece of evidence that Dialect 1 has no number features comes from its availability with reciprocals, even with apparent 3sg agreement on the verb. Malamud (2012) uses reciprocals to show that 3sg agreement morphology with German man involves default agreement, and not a singular feature specification, and her test here shows that MLE man has the same property (German example originally from Cabredo Hofherr 2004:6):²⁶

Hoekstra (2010:43) and Ackema and Neeleman (2018) take this fact (or, at least, related facts from Frisian and Dutch) to be evidence that impersonal pronouns are “semantically plural.” They mean by this that regardless of the morphosyntactic properties of the pronouns (such as giving rise to singular agreement), the semantic content does not preclude a plural interpretation. My contention here is that this is evidence instead of number neutrality, and being able to appear with a reciprocal does not mean that these pronouns are featurally plural rather than number neutral, and that a plural interpretation is possible because pluralities are not excluded from the set of possible referents.

Some evidence for this comes from Mandarin Chinese, where bare NPs, which are number neutral (in that they can refer to atomic entities or pluralities) can also felicitously appear with reciprocals in the appropriate context. Imagine a world in which all men are not very generous, but women are, and during a gift giving season this vice of men and virtue of women becomes very apparent. In that case, one could utter the following:²⁷

A bare noun nüren, which is number neutral, is used, but it is free to combine with huxiang, ‘each other’. Number neutrality appears to be straightforwardly compatible with the diagnostic for semantic plurality that Hoekstra and Ackema and Neeleman employ.

Second, Malamud (2012) shows that German man can also combine with a reciprocal, but is morphosyntactically constrained to the singular, resisting plural nominal and adjectival agreement:

The conclusion is similar to mine here: being able to combine with a reciprocal is the result of number neutrality, and not evidence of syntactically present feature specifications. I take the singular only agreement to be a default, following Malamud.

We can summarise the facts in the following way:

So MLE man for speakers of dialect 1 has no number or person features, meaning that we only ever see man is. For speakers of dialect 2, man are is only possible as a reflex of [−atomic], but not person. Since MLE man never has person features across dialects, there is no dialect in which man am is possible. Man are is blocked in a 2sg environment because there is never a person feature present.

The complete lack of person features will become relevant to the analysis in Sect. 5, but for now I turn to another property of MLE man, namely its resistance to binding.

4 Binding man

In this section I introduce the focus of the paper: MLE man’s behaviour with respect to binding. My main claim is that MLE man’s resistance to binding in a variety of situations means that its generic impersonal interpretation cannot arise from binding by a generic operator.

Overall the picture that arises from speaker judgments is clear: MLE man does not behave like a bound variable. The following subsections present the specific findings in detail.²⁸

4.1 Simple quantificational binding

First, judgments suggest that MLE man cannot be bound by a universal quantifier: man in (29) cannot be bound by every guy or everyone, and hence man here can only refer to some specific individual(s). This is true in local configurations and across clause boundaries, although with embedding the bound reading is not as strikingly degraded as in local contexts.

It’s important to note here that impersonal one also seems to be resistant to binding by a universal quantifier. A Google search finds some examples which suggest that one can be bound by a universal quantifier. Some examples are given in (30).

An interesting fact about these examples is that the first three all come from Indian or Pakistani English web pages, and the last comes from an English translation of a Hindi text. It’s possible that this is a feature particular to Indian English. A reviewer points out that there are examples of online discussions about the use of one in this kind of environment that suggest that it cannot be bound there. A quote from one such discussion: “Everyone loves one’s mother. You’re kidding, right? One doesn’t say this even in England.”²⁹ Another comment was the following: “Even British English speakers avoid constructions like [Everyone loves one’s mother].”³⁰ Some native British English speakers that I have consulted find such cases borderline acceptable, but it seems that for at least some speakers, binding with a universal quantifier is completely out for one.³¹

4.2 Anaphoric binding

By ‘anaphoric binding’ here I mean binding by a coreferential DP (pronoun or r-expression). This is the domain in which the judgments at first do not appear to be so clear cut, but upon closer inspection the pattern that appears is the one that is expected if man does not act as a variable, and thus cannot be bound. First, it appears that man is very strongly judged to be unacceptable as a local domain form, where we would expect a reflexive anaphor.³²

In this example man has to pick out some contextually salient individual other than the speaker, resulting in an interpretation along the lines of ‘of course I like him/her/you/them.’ When the antecedent also takes the form of man, ratings improve slightly, but it is still strongly degraded for most speakers.

Déchaine and Wiltschko (2010) point out that Reinhart’s (1983) generalization that locally bound pronouns “are only interpretable as bound variables” allows us to use local domain forms as a diagnostic for bound variable anaphora. We see here that man is not a local domain form. This does not provide evidence that man cannot ever be a bound variable, but the fact that it is unacceptable in a local environment is explained if it is true that it cannot be bound.

Note that coreference between a fully phi-specified pronominal antecedent and man is, however, possible across a clause boundary.

I claim here, though, that this example does not involve binding: rather, the possibility of coreference is simply the result of man being able to pick out any contextually relevant individual, which can include the speaker (i.e., this is a case of ‘coreference’ rather than ‘binding,’ in the sense laid out in, e.g., Grodzinsky and Reinhart 1993).

An unusual fact that arose from speaker judgments is the following: when man is possessive, and takes a local antecedent, there is a strong preference for reference to be disjoint.

Again, the most readily available reading here involves John loving either the speaker’s mother, or some other salient individual’s mother, so long as it is not John’s own. The question arises of why judgments should be so strong in the case of coreference between John and man in (34). If man can pick out any contextually salient individual, as I will argue in Sect. 5, then why can it not just pick out John in this case, in the absence of binding? I suggest that in fact coreference can be forced, but generally this is overridden because binding is the preferred method of achieving coreference in such local cases, and thus the sentence is judged to be bad even though it is well formed. This condition can be thought of as something akin to Grodzinsky and Reinhart’s (1993) Rule I. One piece of evidence that coreference is actually possible in such cases is that judgments improve when John is replaced by man, as in (35).

4.3 Bound only

Another environment in which man resists a bound variable interpretation is that exemplified in (36) using only, based on an example from Grodzinsky and Reinhart (1993).

(36a) is ambiguous between two different readings: one where John is alone in thinking John is a great cook, and the other bound variable reading where John is alone in considering himself a good cook (i.e., no others consider themselves good cooks). These two readings are summarised in pseudo-logical form in (37).

Again, the bound reading represented by (37a) is not available where man replaces he, (36b). It can only be interpreted (on the relevant reading) as meaning that John alone thinks that John is a good cook (or again, that John thinks some other relevant individual is a good cook).

4.4 Binding in strictly generic impersonal environments

Many of the examples above show that man cannot have a bound variable interpretation in many environments, but I further want to show that bound readings are also blocked in the kind of environments where they are possible for English one.

Moltmann (2006) and Cabredo Hofherr (2010) note that one can be bound in a possessive in local and non-local configurations, as long as the antecedent is identical, i.e., also one. The examples in (38) from Moltmann (2006) illustrate a binding environment with a Q-adverb.

Example (a) has the bound reading which can be paraphrased as “for many relevant persons x, x loses x’s belongings on the train.” MLE man interestingly does allow coreference between two instances of man in the kind of configuration exemplified in (38a), but does not allow the bound variable generic reading.^33,34

This example can only have the interpretation where man refers to some specific individual, and that individual often loses their belongings. The generic reading is not available. This suggests that whatever mechanism that gives rise to the generic reading with one in these contexts is either not available or is somehow blocked when man is used.

4.5 Sloppy readings under VP-ellipsis

I further tested the possibility of sloppy readings under VP-ellipsis. The first example sentence was judged to be strongly unacceptable with a sloppy reading for man.

In this example, the interpretation in which Andrew also did his own homework was judged not available. This again suggests that man is not acting like a variable, and cannot be bound.

There is a potential problem with this specific example. As stated above in (34), the possessive man’s is generally judged to fail to be coreferential with a local antecedent anyway, so this is likely responsible for the strong unacceptability of the sloppy reading. Embedding man so that there is no effect of locality helps clear up this problem. Consider example (41).

Judgments on this example showed that a sloppy reading with man was degraded, even though speakers fully accepted a sloppy reading with a normal 3sg pronoun in the same context. For transparency’s sake, I report that one speaker that I consulted (out of five) did in fact accept the sloppy reading with man, but only with strong stress on man. I think that what is happening here is the following. Stressing man makes it more readily interpretable as a plural definite personal form with the meaning of us/them. This means that it picks out a plurality including both of the referents in the sentence, and thus the full sentence is interpreted as something like the following:

In this case, the interpretation of the elided constituent along the lines of ‘Andrew knows that Andrew is going to get fired’ becomes acceptable, because Andrew is a member of we or they. This explanation gains support from the comments of another speaker, who judged that a sloppy reading was available for us man, but not man alone. Us man is a form (existing as part of a paradigm with you man and them man) which is used to refer to a plurality in all cases (it functions as a normal 1pl pronoun), and is analogous to us lot (and you lot and them lot) in more traditional varieties of English in London. The example would then be equivalent to the following:

Again, as with the example above, the ‘sloppy reading’ is in fact not a sloppy reading at all: when asked whether the second conjunct could mean that Andrew knows that Andrew is going to get fired, the speakers who were reinterpreting it in this way of course think it is possible: and so does Andrew can be interpreted in a way such that it means to the hearer that Andrew knows that Andrew is going to get fired (because he’s one of ‘us man’).

The data on VP-ellipsis is in line with the other variable binding cases, and adds further evidence that MLE man does not behave as a bound variable.

4.6 Quantificational Variability Effects with adverbs

Hoekstra (2010) points out that, unlike definite descriptions, but similar to indefinites, the Frisian impersonal pronoun men can give rise to Quantificational Variability Effects with some temporal quantification adverbs.³⁵

In (4.6b), the adverb quantifies over a variable introduced by the indefinite, so that the interpretation is that most smart students are proud. This kind of effect is not present for definite descriptions as shown in (4.6a), which only has the unusual interpretation that a particular student’s intelligence and pride pattern together. QVE is present for the impersonal pronoun men in (4.6c), suggesting that the impersonal can exhibit bound variable behaviour.

Interestingly, MLE man does not give rise to QVE.

Similar to the definite description example above (if the student is smart, he is usually proud), the only interpretation available in (45) in an out-of-the-blue context is the awkward one noted above. This suggests that another kind of variable binding behaviour is absent for MLE man.

However, things are not quite as straightforward as this. Chierchia (1995) points out that the standard blocking of QVE with definite DPs only applies in out-of-the-blue contexts, and with the appropriate context provided even a definite DP can show QVE (example (46) is taken from Malamud 2012, where she credits it to a reviewer):³⁶

Chierchia argues that definites can be involved in QVE only if a relational variable is introduced, the value of which must be something salient in the context. The introduction of this relational variable is also used to account for contextual domain restriction of the type illustrated in (47).

The same variable is introduced in the interview situation exemplified above, and this variable can be bound by a Q-adverb. Malamud (2012) uses this fact as a diagnostic for (in)definiteness of impersonal pronouns in English and German. She points out that in contexts that do not make available such variables over admission interviews, no QVE is predicted with definite DPs, and thus that “susceptibility to QVE with Q-adverbs can serve as an empirical test for (in)definiteness in English and German” (ibid, p. 35).

When it comes to MLE man, we see the pattern we would expect from a definite description.

This suggests that MLE man does not have the QVE profile of a normal generic impersonal, and in fact has the same profile as a definite DP, only allowing a bound reading where a contextually salient relational variable can be introduced. This point will become very relevant in Sect. 5, where I argue that MLE man is fundamentally a full DP with a definite D head.

4.7 Summary

MLE man has the following properties:

• It can have some kind of generic impersonal reading.

• It can have a definite personal reading, spanning any person (1, 2, 3), and number (sg, pl) combination;

• It does not give rise to person sensitive agreement.

• It cannot act as a bound variable across a variety of binding configurations;

• It behaves like a definite description with respect to quantificational variability effects.

What I conclude from these generalizations is that MLE man has no person features (but can have number features for some speakers), and it behaves something like a definite description, which allows for definite personal pronominal interpretations.

There are two ways to approach the different possible interpretations of man. One could suggest that impersonal man and definite personal man are two distinct, homophonous elements (which potentially share some internal parts). This would by no means necessarily be the wrong approach; remember we have already seen that there is a homophonous bare nominal form man which exists alongside the pronominal form(s), although with a different syntactic distribution. There is also man used as a form of address and pragmatic marker (see fn. 9, and Cheshire 2013). I think that attempting to provide a unified analysis of all of those items would be wrong-headed.

However, the impersonal interpretation and the definite personal interpretations are so close in their general use and distribution that it’s reasonable to suggest that they arise from the same element, and that some extraneous syntactico-semantic factors determine how the pronoun should be interpreted in context. I believe that this is the correct approach. I argue in the next section that the mechanism that gives rise to the definite personal readings of the pronoun can also capture the apparently generic impersonal reading, without appealing to binding by a generic operator. The proposed analysis therefore accounts for all of the properties of MLE man, and shows that there is a mechanism which can produce generic-like interpretations instantiated in example (16), in the absence of variable binding.

5 Analysis: From π to i

Before moving on to the precise analysis of the different interpretations of MLE man, I lay out my assumptions about the make-up of pronouns in general. For concreteness’s sake, I adopt the theory of pronouns (and features thereon) from Harbour (2016). The main reason for this is that Harbour’s theory of pronouns gives us a straightforward system where a lack of specification for person features means that a potential referent can be picked out of a set containing all possible human referents, and where a “bare” person head can project even if it contains no person features.³⁷ What is important to us from Harbour’s system, is that the possible reference of a pronoun is determined by the action of a set of features on a ‘person lattice,’ which is a power set of all combinations of the individuals which can be picked out by pronouns, i.e., me, you, and others. This means we can think of the pronoun in its barest form as introducing a lattice of possible atomic and plural referents, akin to Link’s (1983) semantics for plural nouns, but in addition (and crucially) the pronoun lattice includes the speaker and the addressee.

The mental ontology of person proposed by Harbour includes the following members, with their corresponding notation:

Each element, i, u and o, symbolises an individual that can be referred to. The author and the hearer are both assumed to be unique,³⁸ whereas there are any number of others, each distinguished by prime marks. We can think of different persons (1st, 2nd, 3rd) in linguistic expressions as reflecting sets including different members, and combinations of members, taken from the power set of the members of the list above. For example, second person ranges over the set $u_{o}$, which is an abbreviation of the set containing u alone, plus u in combination with any number of others (o, $o^{\prime}$, $o^{\prime\prime}$ and so on). This means that $u_{o}$ abbreviates the singleton {u}, dyads {u,o}, {$u, o^{\prime}$}, {$u, o^{\prime\prime}$}, ..., triads {$u, o, o^{\prime}$}, {$u, o, o^{\prime\prime}$}, {$u, o^{\prime}, o^{\prime\prime}$}, ..., and so on. This reflects the fact that 2nd person not only has to be able to pick out u, the unique addressee, but also others associated with the addressee, particularly in the case of 2nd person plural expressions.

The entire set of possible referents is the atomic join complete semilattice containing all combinations of the possible members discussed above, i.e., a lattice containing i, u, o, $o^{\prime}$, {i,o}, {i,u}, {i,u,o} etc. This is the entire set of all possible individuals that could be referred to. This can be simplified with the notation in (50):³⁹

This is akin to the denotation of a bare (mass/plural) noun in Link (1983), developed in Chierchia (1998a,b) et seq., but crucially it includes the speaker and addressee (and not just individuals for which a particular property holds), since 1st person and 2nd person pronouns need to be able to refer to them.

Different persons (in the sense of 1st person, 2nd person, 3rd person) are subsets of this total ontology, which are carved out by the functional application of person features on the entire person lattice. Person features are themselves lattices, and the interaction of the content of person features with the person head is expressible as the application of sets on other sets (or rather lattices on other lattices). I put aside discussion of the full technology here, because the way that specific person features work is not relevant to our discussion, since MLE man has no person features. What is important is that all pronouns project the person head π in the syntax. The categorial head π has as its denotation the π lattice $\mathcal{L}_{\pi}$.

A π head which bears no features just introduces all of the possible referents in $\mathcal{L}_{\pi}$.

The root node ϕ does the job of introducing a variable over the set of possible referents (π), and has the following semantics:

ϕ is the root of the functional projection that the head π belongs to. This means that we have the following syntactic representation (53a), and semantics (53b):⁴⁰

The notation π$_{\text{F}}$ simply represents the person head π with some feature F on it. The final line in (0b) says that the meaning of the πP is the set containing x, where x is a member of the person lattice with any person features F applied to it. In our case, since there is no application of person features, this can simply be replaced by $\lambda x. x \in \mathcal{L}_{\pi}$.

A number head ω can be projected above the π head, carrying number features which act on the output of the person features applied to π, further whittling down the contents of the set. The only feature on the number head that is relevant to us here is [±atomic], which takes the input set and returns the set containing only those elements which are non-atomic (i.e., the pluralities) or atomic (i.e., the atoms):

The pronoun ultimately picks out a (possibly plural) individual, and not a set, so the question remains of how that individual is picked out. Harbour does not discuss how we get from the set of possible referents to an individual of type e, but I propose here that this is done by merging a D head above the number head (adopting a definite description account of pronouns; see, e.g., Elbourne (2005) et seq.). In the next section I come back to explaining the properties of MLE man, and give an explicit proposal for the nature of the D head that picks out an individual from the set of possible referents.

5.1 Generating man

In this section I will propose that definite readings of man are the result of the merger of a D head above π (or ω). The D head introduces an epsilon operator, in the sense of von Heusinger (2004), which picks out a contextually salient member from the π set. I argue following on from this that all of the properties of man discussed above can be explained if we assume that the epsilon operator always picks out a contextually salient member of the featureless π set; in short, that D obligatorily projects.⁴¹ Therefore man has no person features, but can still appear to take on an interpretation associated with a particular grammatical person, and the ‘generic’ interpretation is not a binding relation at all, but is produced by precisely the same mechanism that gives rise to the definite readings, and that in that case the set of salient individuals that is picked out is particularly large. Since the variable introduced by the pronoun is obligatorily bound off by the operator introduced by D, there is no free variable available for binding by an operator outside of DP: it is essentially a definite description of type e, differing from normal definite descriptions in that it can directly refer to both speaker and addressee (and other plural individuals including speaker and addressee).

Recall that MLE man has the following properties:

• 1. It can have something like a generic impersonal reading;

• 2. It can have a definite personal reading, spanning any person (1, 2, 3), and number (sg, pl) combination;

• 3. It does not give rise to person sensitive agreement.

• 4. It cannot act as a bound variable across a variety of binding configurations;

• 5. It behaves like a definite description with respect to quantificational variability effects.

In the terms of the theory of person presented above, properties 1 and 2 taken together mean that the set of referents that man can pick out cannot exclude $i_{o}$, $iu_{o}$, $u_{o}$, or $o_{o}$. In its definite personal use, MLE man can pick out any and all possible referents that other pronouns can pick out. In its impersonal use, it can pick out a general group which includes the speaker and the hearer, and possibly others. We saw earlier that $\{i_{o}, iu_{o}, u_{o}, o_{o}\}$ is just the set introduced by π with no person features on it. Coupled with the agreement facts reported in Sect. 3, we can say that, at a syntactic level, the π head of man must have no feature specification, and is minimally represented as in (55), or as (56) for those speakers that have number sensitive agreement with the pronoun:

Ideally an analysis of the internal structure of MLE man will provide us with an explanation of how the different definite personal interpretations are picked out, but also points 4 and 5 from above: the lack of bound variable readings, and the behaviour akin to that of a definite description. I introduce a mechanism which captures all of those facts in the next section.

5.2 Picking out individuals

To account for the variety of different possible definite readings that man can have, there needs to be some mechanism by which some member of π is picked out. Here I suggest that mechanism is the introduction of a D head, which is interpreted through a global choice function that picks out the contextually most salient entity in the discourse. For concreteness sake, I assume that the D head introduces an epsilon operator, in the sense introduced in Egli and von Heusinger (1995), and von Heusinger (2004). I elaborate on the properties of this operator in this section, and show how it applies in definite descriptions. The following is a summary of parts of von Heusinger (2004).

The epsilon operator is a selection function that assigns to each non-empty set one element of this set. Unlike ι, it carries with it no uniqueness presupposition as a condition of interpretation; in this way, it is completely acceptable for the potential set of referents to be non-unique (as is the case with MLE man), but still arrive at an interpretation. The epsilon term is then interpreted relative to a global choice function, which von Heusinger calls Φ. This choice function is relativized to a specific context, so that it can pick out the appropriate member. Epsilon terms are interpreted according to the following rule (the model M is the pair 〈D,I〉, D the domain of discourse, and I the interpretation of constants):

Each context c has its own corresponding choice function $\Phi_{c}$, which represents the salience structure of the discourse. Uniqueness is understood as “the ‘unique availability’ of the referent” (von Heusinger 2004:315) rather than as a requirement that there be a unique relevant individual that corresponds to the descriptive content of the NP.

An example from von Heusinger (2004) will help to illustrate. Imagine a situation in which we are visiting Lake Konstanz. In this situation, the property island holds of three entities: Mainau, Reichenau and Lindau. The definite description the island could pick out different entities in different situations. If a Reichenau fisherman utters it, then it likely refers to the island of Reichenau; if a tour guide on Lindau utters it, then it will refer to Lindau; and if Earl, the owner of Mainau, utters it then it refers to Mainau. Each of these situations will have a contextually relativized choice function $\Phi_{c}$ assigned to it, which represents the salience structure of that particular context c.

This approach to definite descriptions makes sense of situations in which there is more than one potential referent, but yet where the definite description is still clearly interpretable. The difference between the interpretation of a definite description under an iota operator analysis and an epsilon operator analysis is given in (59)

The iota operator requires that there be a unique available referent, whereas the epsilon operator only requires that some referent can be selected, and then interpreted by the appropriate choice function.⁴²

I propose, therefore, that this is how the appropriate interpretation of MLE man arises. In a definite context, man is represented as in (56), where a D head projects and introduces an epsilon operator, turning the pronoun into an epsilon term, which is interpreted by a global contextual choice function.

I assume that the nominal content of the pronoun that is interpreted is simply represented as human(x), since man has to pick out a human individual (and cannot refer to non-human animates or inanimates). Since there are no features on π, the set of individuals that can be picked out by the contextual choice function is unrestricted (unless the speaker is a speaker of Dialect 2, in which case the number head can restrict the potential referent to only atomic or non-atomic entities). This means that any contextually salient (human) individual, regardless of person and number, can be picked out.

The question then arises: from where does the generic interpretation arise, and why can man otherwise not be bound? If we assume that a generic interpretation involves binding of a variable by a Gn operator, as is standard, then it is surprising that man cannot otherwise undergo binding, when a D head is not introduced. I suggest that it is the case that D obligatorily projects, and therefore man can never undergo binding, since the epsilon operator binds off the variable over the set of individuals introduced by the pronoun.

If the above is correct, then binding of the pronoun by Gn is never an option. Instead, I propose that the apparently generic interpretation of man does not involve binding by a Gn operator at all, and instead arises purely as a consequence of the nature of the action of the choice function. Recall that the π set is unconstrained, and therefore that the epsilon operator can potentially pick out any member of the full person lattice, including very large members, potentially members which contain all of, or the majority of, all possible referents (in an appropriate discourse context).

Harbour’s theory of person is ideal for capturing these facts because a lack of feature specification in that system provides us with precisely the entire set of possible referents, without any additional stipulations or ad hoc principles. The impersonal reading and the breadth of possible interpretations available for definite readings of MLE man together make immediate sense in this system. The context dependence of the interpretation of man falls out naturally once we assume that D introduces the kind of operator that allows for contextually determined selection from a set of possibilities.

The obligatory projection of D also explains why binding is never possible for man: D always projects, and picks out an individual, in essence acting as a binder of the variable lower in the pronominal projection.⁴³ The QVE effects discussed in Sect. 4.6 are particularly revealing in this regard, since they suggest that MLE man is behaving very much like a definite description, giving rise to and blocking QVE precisely where a definite description would. The fact that QVE is generally blocked in out-of-the-blue impersonal contexts also suggests that the “generic” impersonal interpretation associated with MLE man is not quite the same as that associated with other impersonal pronouns. This difference arises because of the different mechanisms which give rise to the “generic” interpretation.⁴⁴

If a Gn operator were present higher in the clause, it would find no variable to bind. Other impersonal pronouns that I have discussed above, however, do not behave in this way. Some allow binding by quantifiers, or quantificational variability with certain adverbs in out-of-the-blue contexts (like indefinites), and this is expected if those pronouns are not interpreted in the same way as MLE man.

Take for example English one. To repeat the discussion in Sect. 4.4, one can act as a bound variable, bound by Q-adverbs as the examples in (38) from Moltmann (2006) illustrate.

Example (a) has the bound reading which can be paraphrased as “for (almost) every relevant person x, x loses x’s belongings on the train.”

What about the unavailability of arbitrary existential interpretation for MLE man? I pointed out in Sect. 2.2 that Moltmann (2006) and Malamud (2012) argue that the unavailability of this reading with one is down to a Gn feature on one which restricts it to contexts in which it is bound by a sentential operator or a Q-adverb. I think that this explanation does not extend to MLE man: unlike one, it doesn’t allow QVE in out-of-the-blue contexts, and so behaves more like a definite description. The lack of arbitrary existential interpretation can again be explained by the obligatory projection of a D head containing an epsilon operator. Fundamentally MLE man is behaving like a definite description.

6 Summary and conclusion

The aims of this paper were twofold: first to provide an analysis of the unusual properties of the recently innovated pronoun man in MLE, and second to argue that its apparently generic impersonal interpretation could not be explained under a traditional Gn-binding approach. On the first point, I argued from agreement facts that the pronoun is devoid of person features, and proposed an analysis where the pronoun gets its interpretation from the projection of a D head hosting an epsilon operator, which can pick out any (possibly plural) individual from the lattice of all appropriate individuals (including speaker and addressee), introduced by a person head π.

On the second point, I make the claim that the generic reading of MLE man does not arise from the same mechanism as other impersonal pronouns, i.e., by binding of the pronoun (acting as a variable) by a generic operator. It is important to note that I am not making the stronger claim that a Gn-binding approach to generic readings should be ruled out in the case of impersonal pronouns: indeed, as we have seen, it is still the case that such an approach suits English one, and other impersonal pronouns. However, I am claiming that this is not the only way to get what looks like at least one type of generic reading.

On the empirical side, I have laid out the properties of MLE man, a pronominal form that is (as far as I know) unique in the breadth of possible referents that it can pick out: it can be interpreted (pseudo-)generically, but can also have a definite personal interpretation across all person and number combinations. While many impersonal pronouns can pick out definite referents in the right context, this is usually restricted to first person singular or plural, and more rarely third person singular. MLE man seems to be somewhere in between a personal and impersonal pronoun, and its ability to have such a broad range of interpretations comes from its lack of feature specification, and the obligatory projection of D.

I have also attempted to show that a theory of person which fundamentally treats the person head in pronouns as a head that introduces a Link-like lattice containing the full set of person referents (including speaker, addressee and others) makes an explanation of the referential properties of MLE man straightforward. I have, for the sake of concreteness, adopted Harbour’s theory here, but any theory where some operator can pick out individuals from an appropriately constrained set is suitable.

Finally, I have suggested that a theory of definiteness that can take into account contextual salience is best suited to accounting for the wide range of possible referents that MLE man can pick up in definite contexts. The reference of the pronoun really is about context, and an approach to definiteness which employs the iota operator, and therefore imposes uniqueness/maximality, would have a harder time accounting for the flexibility exhibited without the introduction of more complex semantic machinery.