2. Prior discussion on lexical be

Very little has been said about lexical be in the traditional literature. The first mentionFootnote ³ we find is Erades (Reference Erades1953), who cites the following example.Footnote ⁴

Quoting Wood, Erades says that this is ‘certainly not a solecism’ and that ‘its meaning is not the same as why aren’t you reasonable?’. He further observes that Why don’t you + inf. ‘is quite frequently used as a kind of rhetorical question which is, in effect, not a question at all but a suggestion or a piece of advice’. For verbs other than be, he suggests, the structure is ambiguous, with only the context making the intent clear. Wood also comments on a change in the meaning of be: the verb itself has ‘a slightly different meaning […] suggestive of purpose, effort, or activity directed toward the specified end. Be means something like “take steps to become” or “bring yourself to be”’. He cites, for instance, a case like Why don’t you be a doctor?. We see here the first recognition that lexical be, unlike auxiliary be, explicitly expresses a directive or desiderative force.Footnote ⁵

Denison (Reference Denison, Fischer, Rosenbach and Stein2000, §3.3.1) points out some examples that are quite different from (1), such as (5) (his examples (50) and (52)):

He also cites the classical cases, including (6) from Wharton’s Age of Innocence (his example (57)), which he calls the ‘the quasi-imperative Why don’t you be …? construction’, which he claims is ‘now fully accepted in standard’.

He further states that ‘The typical syntactic contexts for tensed do + be are non-assertive, frequently both negative and either interrogative, quasi-imperative, or conditional. If be is lexical, it usually forms a nonstative group-verb with its complement’.Footnote ⁶

Huddleston et al. (Reference Huddleston and Pullum2002, 114) propose a detailed description of lexical be. They provide cases similar to (1) and note that they are ‘virtually restricted to the negative’ as shown by the unacceptability of the variants without not, as in (7).

They also note that in cases like 1, the utterance pragmatically conveys a directive ( $ \approx $ ‘Be the judge. / Be quiet.’). They even provide 3sg variants (‘Why doesn’t he be …’). And they point out a second subtype of the construction with if, illustrated in (8), noting that it again conveys a directive (‘you/he should be quick’):

It is important to stress that the acceptability of such cases of lexical be is truly surprising. In typical cases, attempting to use supportive do with be is highly unacceptable to the point of being unintelligible, even when the predicate is easily interpreted as actional, as seen in (9).

With stative predicates, attempting to use supportive do leads to even stronger unacceptability (even with stage level predicates, as in (10c) and (10d)).

The only substantive theoretically oriented discussion of lexical be, to our knowledge, is that of Collins (Reference Collins, Deumert and Durrleman2006), couched in a Principles and Parameters perspective. He bases his discussion on a set of naturally occurring data collected on the internet through Google searches. He cites classical cases like (1) and (8) above (e.g. his (5), (12), (13), (14)), and he also mentions examples like the following, where lexical be occurs with past tense supportive do in an if conditional, and in a main clause polar interrogative, respectively:

Collins’ central claim is that the construction is only possible with agentive be. He notes that lexical be is never inflected in his variety of English (‘*If he be’s/bes/bees careful, …’ [his (35b): 207]).Footnote ⁷ He also notes that the construction ‘seems more acceptable in the context of an implicit command’, as in (13).

We take up Collins’ account of lexical be in Section 5 in our discussion of the conclusions to be drawn from the possibility of lexical be.

Finally, Kjellmer (Reference Kjellmer2009) provides a corpus-based investigation of lexical be in British English, based on the British National Corpus (BNC), the CobuildDirect Corpus and the internet. He points out the predominance of Why don’t you be XP? and suggests (p. 81) an analogy with negative imperatives noting the semantic proximity between the two. He also notes (p. 83) the predominance of the present tense. But, beyond this, his examples of lexical be are mostly taken from the internet and consequently, as with Collins’ data, do not offer a clear picture of the relative frequency of the various subconstructions.

Before closing this section, let us explicitly make clear that the negative imperative with be (e.g. Don’t you be too hard on yourself.) is not an instance of lexical be. As discussed below at the end of section 4, do in the negative imperative does not mark tense and is hence not a genuine case of do-support, contrary to what is the case with lexical be.

3. Corpus investigation of lexical verb be

In order to get a more detailed and representative picture of lexical be, we investigated its use in the COCA and in the SOAP corpora.Footnote ⁸ We used various queries such as [do] n’t [pp*] be (any form of do followed by n’t followed by any personal pronoun followed by be) which were designed so as to retrieve the great majority of occurrences.Footnote ⁹

Complete list of queries:

We eliminated noise and a certain number of cases which provided clear evidence of not being standard English. This provided a total of 208 (65/143)Footnote ¹⁰ occurrences. Table 1 provides an overview of the results.

Table 1.

Frequency of subconstructions involving lexical be in COCA and SOAP

The most striking property of these data is that the distribution over construction types is extremely skewed. First, we see that the main clause why-interrogative subconstruction (WDYB) is by far the most frequent, comprising more than 80% of occurrences. Two further subconstructions occur comprising between 5 and 10% of occurrences, specifically (i) the if-conditional subconstruction (IYDB) identified by H&P (6% of occurrences)Footnote ¹¹ and (ii) the main clause how-interrogative subconstruction (HDIB), not noticed by H&P, with 9% of occurrences, illustrated in (14).

The other cases found have a frequency close to 1%. Furthermore, numerous imaginable configurations were not attested at all (e.g. those in (9) above).

Second, we note that subjects are heavily constrained depending on the subconstruction, as shown in Table 2. Second-person you is present in 90% of the cases in the WDYB subconstruction and 75% in the IYDB subconstruction, whereas first and second person appear to be balanced in the HDIB subconstruction. Third-person subjects are only marginally represented overall.

Table 2.

Frequency of subjects by subconstruction in COCA and SOAP

Third, polarity is constrained by subconstruction: all occurrences of the WDYB construction have n’t, whereas all of the HDIB constructions are positive.Footnote ¹²

Fourth, the form of do is highly constrained. It is always do or don’t except for one occurrence of doesn’t with the subject she and two of did with subject he and one of didn’t with the subject they. Of special interest is the absence in our corpora of past tense uses with first- and second-person subjects, as in (15), despite the fact that the form makes perfect sense (viz., ‘{I/You} should have been quiet/the judge’).

As for speech acts, as suggested by H&P, directives dominate.Footnote ¹³ This is centrally the case because the WDYB construction always clearly has this interpretation with a you or we subject, as in (1a) above. Similarly, most cases of the IYDB construction have clear directive force, as in (8) above.

Other cases, illustrated in (16), do not have actual directive force. However, all of these involve a deontic or desiderative stance that is closely linked to directive force, in the sense that the subject should be or wants to be the agent of the action, as indicated in the glosses.

Finally, Table 3 provides information about the category of the predicate in lexical be constructions. Clearly, AP and NP predicative complements strongly dominate because they can easily denote actional predicates. The low number of PP predicative complements can be explained by the fact that typical PP locative complements are not actional (??Why don’t you be in the garden). This can be compared with the actional PP predicate in our corpus example Why don’t you be more like me? ( $ \approx $ ‘Why don’t you behave more like me?’).

Table 3.

Category of the predicate of lexical be

4. Explaining the range of meanings and the differences in acceptability of lexical be

In this section, we tease out the fine-grained conditions that license the use of lexical be. The first point concerns the form of the clause (i.e. whether it is a why/how-interrogative or some other interrogative, an if-clause, etc.). The second concerns polarity, the third concerns the person and number of the subject, while the fourth concerns tense and aspect. The subtle properties of the various constructions involving lexical be all contribute to the argument for a constructional account, and against a derivational account such as Collins (Reference Collins, Deumert and Durrleman2006).

We start with the most frequent construction, Why don’t you be XP (WDYB), which accounts for over 80% of occurrences of lexical be in our corpus data. Our hypothesis is that this construction is a blend inheriting properties of the more general Why don’t you VP construction (WDY) and the negative imperative with subject (NIS) construction Don’t NP VP, where NP is usually you. Independently of lexical be, the more general WDY construction is widely used as a means of expressing weak directives (i.e. suggestions or invitations). With the imperative and deontic should, it is a member of a set of constructions expressing a gradient of decreasing deontic force.

We suggest that the WDYB construction arises from the pressure to make the full gradient available to actional predicates naturally expressed with be as the verb:

We also propose that this is facilitated by the fact that the sequence don’t you be independently exists in the negative imperative with subject construction (NIS, for example, Don’t you be late!). Blending the two constructions is further facilitated by their semantic similarity.

It is important to note that the gradient deontic force affects the way the directive is presented by the speaker to the addressee rather than the nature of the directive itself. As pointed out by Huddleston et al. (Reference Huddleston and Pullum2002, 929ff), imperatives can convey a wide variety of directives (from commands and requests to advice, warnings, instructions, invitations, permission and wishes). It is, in fact, the properties of the directed action which determine how the directive is interpreted. In particular, if it is beneficial to the speaker, a command or request is most likely; if it is beneficial to the addressee, advice, warnings, instructions, invitations, permission and wishes are more likely.

The WDY construction can similarly express a broad range of directive meanings,Footnote ¹⁴ but it presents them as not having the force of a command. In a case like Why don’t you be quiet!, where the resulting situation is clearly beneficial to the speaker and not the addressee, the WDY construction presents, under the guise of an invitation, what can in fact only be understood as a command or request.

As stated above, our proposal that the WDYB construction results from a blend of the WDY and NIS constructions receives initial support from the presence of the negative auxiliary don’t, which we claim is inherited from both the NIS and the WDY construction. The latter requires negation; positive why interrogatives do not have the directive force of WDY. Compare (17c) with Why do you sit down?. Further support for the idea of a constructional blend can be obtained by a close examination of the subjects in WDY and WDYB. If WDYB was simply an extension of WDY, we would expect the range and proportions of pronominal subjects in both constructions to be the same.

A closer look at the data shows that this is not in fact the case. Consider Table 4:

Table 4.

Person and number of subjects in WDY and WDYB in the SOAP corpus

In order to establish a useful comparison between WDYB and WDY, we have to eliminate cases where Why don’t you … ? is actually an information question (and hence not a case of WDY). A close look at the data shows that the information question interpretation typically arises with stative verbs (Why don’t you like them?).

In this light, we searched the SOAP for occurrences of main verb do in WDY (i.e. the pattern why do|does n’t PRON do). Though it is in principle possible for such sequences to denote information questions, an examination of the first 100 occurrences in the SOAP shows that they are all, in fact, relevant instances of weak directives – that is, cases of WDY. Further evidence that information question interpretations are not present here stems from the fact that third-person subjects are almost absent.

Both the WDY and the WDYB constructions show strong skewing toward second person and, to a lesser extent, toward first. However, it is interesting to note that there is a striking difference between the two in the distribution among first- and second-person person pronouns. The bias toward second person is stronger for WDYB than for WDY.Footnote ¹⁵ Since the weak directive interpretation is the same for both constructions, it alone does not explain this difference. We suggest that the bias toward second person in WDYB supports the hypothesis that it is a blend with NIS, where the subject is almost always you. Footnote ¹⁶

Consider now the restriction to present tense we found in the WDYB construction. If WDYB were a simple extension of WDY, we would expect the same distribution of tenses in both constructions. But the data from the SOAP show that this is once again not the case. For WDY, the SOAP provides 74 occurrences of Why didn’t PRON do?. This is unsurprising as these cases implicate a counterfactual deontic reproach interpretation similar to ‘You should have’ – for example,

These 74 occurrences in the past tense are then clearly instances of WDY.

We thus obtain a proportion of 74 past tense occurrences of WDY for 742 occurrences in the present tense (see Table 4). This should be contrasted with 3 occurrences in the present tense of WDYB for 208 in the past tense (see discussion just before example (15)). These are clearly different proportions.

We suggest that the constraint on the present tense is inherited from the NIS construction where the form is always don’t in the present. The rare cases where we find the past tense didn’t can then be attributed to further ad hoc extensions of the construction, where the formal link to the negative imperative is overridden.Footnote ¹⁷

Once lexical be is established in the WDYB construction, it can extend to further constructions where it keeps most of the same formal and semantic properties. Only two such constructions show any degree of productivity in our corpus data. Consider first the If you don’t be (IYDB) construction, noted by Huddleston et al. (Reference Huddleston and Pullum2002, 114). As they point out, it is used in contexts where the utterance indirectly conveys a directive as in:

The IYDB construction thus preserves the central semantic property of WDYB. It also preserves the formal properties: second-person subjects represent 75% of occurrences, and all occurrences are in the present tense and negative with the form don’t.

Second, consider the How do I/you be (HDIB) construction (see also above examples (14)):

From a semantic point of view, these examples appear to have diverged a little with respect to the directive meaning of the WDYB construction. There is no actual directive expressed. Rather, the occurrences express deontic modality ((21a) $ \approx $ ‘I should be the real me’) often combined with an element of desiderative modality (‘I want to be the real me’). Formally, the construction shares wh-interrogative status with the WDYB construction. Furthermore, all of the examples are in the present tense with the form do. However, there is a significant change with respect to subjects. First, 10 of the 19 examples are first person. But beyond that, the 9 occurrences of second person do not have the same interpretation as in WDYB and IYDB: rather than referring to the addressee, they clearly receive a generic interpretation, applicable to the case of the speaker.Footnote ¹⁸ Clearly, the speaker in (21b) is generalizing over her own desire to be a (good) mother. There is no attempt to act on an addressee (this explains our choice of I in the HDIB acronym, rather than Y as in WDYB and IYDB). This change in interpretation with respect to the subject is easily understood given the shift from directive (influence on the addressee) to deontic/desiderative (attempt to influence oneself).Footnote ¹⁹

Four additional cases appear in the corpus, comprising 7 occurrences. These diverge from the central constructions that we have been discussing. They appear in Table 1 and include the examples in (22).

Interestingly, the first three of these express a deontic meaning. Note specifically the deontic should in the initial clause of (22a). Clearly, the two following clauses should be understood as expressing the same modal judgment. Note also that all of these have first- and second-person subjects and are in the present tense with the form do, following the preferences found in the three central constructions. Example (22d), where there is clearly no deontic component present, seems to illustrate a different situation which can enhance the acceptability of lexical be – namely, cases where there is no other simple and obvious way of conveying precisely the same content. As we have seen, lexical be is restricted to actional predicates. Here, without a predicate, it seems to keep this actional meaning, making it preferable to the obvious alternatives (You aren’t / You don’t exist) which would tend to receive a strictly stative interpretation.

Similarly, it is interesting to note that the single instance of the WDYB construction with past tense didn’t found in our corpus is clearly a case of sequence of tense in free indirect discourse:

Clearly, the internal speech of the protagonist would be the standard Why don’t you just be quiet?.

It is, of course, possible to find occurrences of WDYB with a past tense that is not the result of sequence of tense by randomly searching the web, even in the third person (though these are clearly much less frequent than second person) for instance:

More generally, other types of wh-questions can be found (though they are clearly rare, and some occurrences are obviously produced by non-native speakers). These seem much more acceptable when they can be interpreted in context as involving a deontic stance.Footnote ²²

In (25a), which comes from a paper published in a scientific journal, there is a clear deontic stance that is inferrable from the context (note the importance of your role and the ease of paraphrasing with should: So when should you be an active citizen?), and the example feels very acceptable. In (25b), however, there is no deontic stance at all. The utterance is a strict information question (that receives answers in the discussion forum about the dating game being referred to). The example seems less acceptable to us, and there is independent evidence of sloppy writing in the post (the dates that pop up of him). It is possible that be is used here instead of become.

In sum, if one conducts random searches on the web in general, one can find examples that diverge to varying degrees in meaning and form from the three central WDYB, IYDB and HDIB constructions we have discussed. The advantage of working on a well-defined corpus, as we have done here, is that it provides some idea of which cases are in fact relatively frequent collocations and which are much less frequent. Our proposal is that once lexical be is established in the three central constructions, with their related semantic and formal properties reviewed above, it can give rise to ad hoc extensions. The closer such extensions are to the three central constructions in their semantic and formal properties, the more likely they will be to be produced and the more natural they will sound.

Throughout this discussion, we have assumed that lexical be is only possible for copular uses of be. The following example from the COCA suggests that passive helping verb be can be lexical:

This is clearly a true passive and not a copular construction. Be here is not agentive, it is meaningless (the corresponding active without be is truth-conditionally equivalent: Why did someone not declare him unfit?). The sentence is acceptable because it has clear deontic force ( $ \approx $ ‘he should have been declared unfit’), despite the fact that the agent who is the target of that force is left unexpressed in the passive.Footnote ²⁵

This leads us to a final issue concerning the occurrence of be in imperatives. Since be systematically requires supportive do in the negative imperative construction (NI), should we consider that be is always lexical in imperatives? Although they are marginal, there are cases of the passive (and much more rarely the progressive) auxiliaries in imperatives, as in (27).

When negated these also require do, and some of these clearly allow readings as actual passives rather than as copular clauses.

And be occurs as well with emphatic do.

The difference between the imperative and WDY is that the occurrence of do in the imperative does not mark tense, and hence is not a genuine case of do-support; it is in the tensed question WDY. Do in the imperative is tenseless, as shown by the fact that there is no agreement with a third-person singular subject, as in (30) (see Culicover Reference Culicover1971).

So bare be occurs in the two constructions but for different reasons, and it satisfies different interpretative constraints in the two constructions. In the imperative, be must be understood as ‘active’ (Partee Reference Partee and Zampolli1977), and as directly or indirectly under the control of the addressee. In WDY, it has the feature of being under the direct control of the subject.

We conclude that lexical be is licensed by the three clearly productive constructions WDYB, IYDB and HDIB. Other uses are ad hoc extensions of these that are more likely to be produced and to be acceptable if they conform to the semantic and formal properties we have uncovered, especially the presence of deontic force.

5. Collins’ minimalist approach to lexical be

We consider now the question of how to account for lexical be in a grammar. The only account in the contemporary generative literature we are aware of is that of Collins (Reference Collins, Deumert and Durrleman2006). Collins proposes to account for lexical be by assuming that there are two verbs be: auxiliary be and agentive be. Following Chomsky (Reference Chomsky1995), he argues that only semantically vacuous verbs can move to Infl in English. On the assumption that auxiliaries are semantically vacuous, this provides an account for the fact that they do not show do-support in the NICE constructions. In a nutshell, Collins proposes that agentive be has an actional meaning that makes it unable to raise to Infl. Thus, it requires do-support because it is not semantically vacuous.

We argue that this approach is unsatisfactory for three reasons. First, we show in Section 5.1 that the agentive/ non-agentive contrast is not a property of be, but of the predicative complement (see Rothstein Reference Rothstein1999). Second, we argue in Section 5.2 that the more general derivational approach lacks explanatory value. Third, we show in Section 5.3 that that the derivational approach is not able to account for the systematic differences in frequency and acceptability between the different uses of lexical be observed in Section 4.

5.1. Should one distinguish agentive and non-agentive be?

Is the distinction between agentive and non-agentive be warranted? Rothstein (Reference Rothstein1999) argues against the assumption that there is more than one copular verb be (contra Partee Reference Partee and Zampolli1977 and Dowty Reference Dowty1979). We follow Rothstein here in claiming that differences in agentivity with be are due entirely to the properties of the predicative complement. A central piece of evidence for this is that the relevant differences in interpretation appear in various constructions that involve a subject–predicate relation without any overt verb at all. This can be seen by noting the parallelism between the a. and b. versions of the following pairs, independently of the presence or absence of the copula.

The subject control construction with try in (31) implicates that the subject is the intentional agent of the predicate – hence the felicity of adjectival predicates which allow an agentive reading like quiet and brave and the incompatibility with ILP predicates like tall and French. Similarly, the implied goal-directed action in (32) is embodied in the predicates without be. Despite the absence of be, the NP–AP complementation of the perception verb in (33) and the with NP XP construction in (34) allow predicates with an agentive reading but not ILPs.Footnote ²⁶ It thus appears that the relevant interpretative differences are properties of the predicates and that the range of possible predicates is constrained by each construction. Ockham’s razor suggests that there is no reason to assume that different versions of be are in any way required to account for them.

In this light, let us have a closer look at the central data on which Partee (Reference Partee and Zampolli1977) bases her argument in favor of distinguishing a specific agentive be (her (50)–(57), 306).

Partee argues that there is an agentive be that requires a volitional subject. This is clearly not a stative verb, as evidenced by its compatibility with the progressive in (35b). The volitionality condition explains the impossibility of (35d), where the subject is inanimate (Partee uses the variants with makes/is making to establish that volitionality is not simply a property of the progressive, but is linked to be itself).

We argue that even if one assumes that Partee is correct about the need to distinguish a specific agentive be, it would be irrelevant to our discussion because that be would be an auxiliary, rather than a lexical verb. It is impossible to use the NICE constructions to see this if one starts from (35b) because of the presence of the progressive auxiliary. But if we put the sentence in the past tense, both the property reading of (35a) and the agentive reading of (35b) are available:

Now consider what happens if we try to apply negation or subject auxiliary inversion to (36b):

Clearly, the most natural step is to treat be as an auxiliary and simply add n’t and invert as appropriate. Note that this is the case even under the volitional reading of these sentences, where be is clearly Partee’s agentive be. The variants with do-support are similar to the cases of lexical be given in (22) cited earlier, but they are much less acceptable given the lack of any deontic stance.

We conclude that Partee’s arguments do not bear on our discussion of lexical be because, in most cases, there is no reason to consider her agentive be as anything other than an auxiliary from a morphosyntactic point of view.

Beyond these considerations, Collins’ proposal fails to account for the fact that in many cases, the auxiliary version of be, with the associated raising to Infl, has the same interpretation as the lexical version. For example, (38a) can have the same agentive interpretation as (38b).

Examples like (38a) show that with appropriate predicates, such as quiet, either the non-agentive or the agentive interpretation is possible. Only when there is do-support is the agentive interpretation required. If agentivity resides in be, it should not be vacuous in (38a) under the agentive interpretation and should consequently not be able to raise over negation, contrary to fact.

5.2. Does the minimalist account actually explain anything?

Collins (Reference Collins, Deumert and Durrleman2006, 212) explains the possibility of do-support with be by invoking a proposal of Chomsky (Reference Chomsky1995): in English, a verb with semantic content does not raise to Infl in NICE constructions. By assumption, auxiliary verbs do not have semantic content, so they are the only verbs that raise to Infl in English. Under standard assumptions, if Infl does not have a verb attached to it, do is inserted. Thus, we get alternations such as (39), where Infl is realized as Pres(ent Tense).

Presumably, lexical be has semantic content on Collins’s account because it selects an agentive subject. So it does not raise to Infl, triggering do-suport in these constructions. The hypothesized difference between auxiliary and lexical be is illustrated in (40), paralleling (39).

Besides the fact that this proposal is simply a notational device for marking the auxiliary/lexical distinction, it has a number of other shortcomings. First, Collins has to make a number of otherwise unmotivated ad hoc assumptions about empty heads and verb raising to get the correspondences between morphosyntactic form and meaning to come out right. Since this is a characteristic of virtually all analyses that assume empty heads and verb raising in order to derive surface forms, we do not dwell on it here; for extensive discussion, see Culicover and Varaschin (Reference Culicover and VaraschinIn press).

Second, Collin’s proposal is falsified by the fact that English have, which functions both as an auxiliary and a lexical verb, can show NICE properties even when it expresses possession (and consequently is not semantically vacuous). The examples in (41) show that possessive have can participate in subject Aux inversion and precede negation.

But the variants with do-support are equally well-formed.

It is not plausible to claim that have is semantically vacuous in (41) but not in (42), since the sentences are synonymous.

Similarly, the claim that modal auxiliaries are semantically vacuous also requires a neological understanding of semantic vacuity. Compare the two following sentences:

Both may and seem express a second order modal judgment (epistemic for may, evidential for seem). There is no reason to claim that may is semantically vacuous while seem is not. Both for have and for verbs expressing modality, such a claim would involve turning the morphosyntactic distinction between auxiliary and lexical into an unmotivated semantic distinction between vacuous and non-vacuous.

6. A constructional analysis of lexical be

We turn now to our constructional analysis. Section 6.1 provides a brief overview of the formalism, and Section 6.2 spells out the particular constructions that license lexical be and their inheritence relationships.

6.1. Constructions

A grammar comprised of constructions exemplifies the model-theoretic approach to grammar of Pullum and Scholz (Reference Pullum, Scholz, de Groote, Morrill and Retor’e2001). We assume that the formalization of a construction has (at least) three corresponding tiers (Jackendoff Reference Jackendoff2002; Culicover and Jackendoff Reference Culicover and Jackendoff2005; Culicover Reference Culicover2021; Culicover and Varaschin In press): phonology (phon), syntax (syn) and meaning (sem). An individual linguistic expression – a construct – is licensed by a set of constructions if it satisfies the licensing conditions imposed by the constructions. These conditions constrain the possible correspondences between the terms in the tiers.Footnote ²⁷

On this view, a single lexical item is a construction. The construction for the verb kick is given in (44). For simplicity of representation, we use the spelling or the index of a constituent in phon rather than the phonetic form. Cosubscripting indicates the correspondences.

This construction says that kick is licensed in a linguistic expression if the phonetic form of the expression is /kık/ and it has the appropriate meaning. By the same token, a construct with the phon representation /kık/ and the sem representation λy.λx.kiss’ (agent:x,patient:y)]₁) is not licensed because kick does not mean ‘kiss’.

A somewhat more general construction that also has some idiosyncrasies licenses the idiomatic VP sell NP down the river is (45). $ \oplus $ signals ‘immediately precedes’.

Strictly speaking, the phon tier does not have to be specified here, since the ordering is licensed by the head-initial VP construction in (46) below. And in fact, other VP orders are possible, for example, with heavy NP shift (They sold down the river [_NP all the people who depended on them]).

Other constructions state more general correspondences. For example, the English VP is licensed by (46). This construction says that V precedes the arguments and adjuncts of VP. The interpretation is left unspecified, since it is a matter of the particular lexical constructions such as (44) and (45). $ \ll $ signals ‘precedes’.

6.2. Analysis

Now back to lexical be. We characterize the contexts in which lexical be appears in terms of the constructions that license it. There are two constructions to take into account, Negative Imperative and the why don’t you construction WDY. Negative Imperative licenses tenseless don’t be with directive force, and tensed WDY licenses weak directive force. The tensed WDYB construction is parasitic on these two constructions, allowing tensed don’t be with weak directive force. Further extensions of the construction license the subconstructions IYDB and HDIB.

The details of the negative imperative construction are given in (47).

(47) states that in the negative imperative, there is do-support with the form don’t, followed by an optional subject – you or a general addressee such as anyone/everyone – followed by the VP. The special case with the optional subject is what we have referred to as NIS. The position of the subject aside, we do not need to stipulate the linear ordering properties of phon, since they are licensed by independent constructions, such as (46).Footnote ²⁸

Crucially, the negative imperative requires its own construction, since its morphosyntactic properties are idiosyncratic and do not follow from those that govern the imperative – it is a syntactic nut in the sense of Culicover (Reference Culicover1999). The licensing of don’t in this construction is an idiosyncrasy, since the imperative is tenseless and thus does not license productive do-support, as already noted (see (30)).Footnote ²⁹

The WDY construction must stipulate the exact form why don’t you as well as directive force.

This is all simply bookkeeping, in a sense – keeping track of the unpredictable small details. But these formulations are necessary, since they take account of the irreducible idiosyncrasies of the language. They are, again, ‘syntactic nuts’, in the sense of Culicover (Reference Culicover1999).

What is critical from the perspective of lexical be is that these two constructions share three properties: they have directive force, they license the form don’t, and the verb has the bare form. The bare form is an idiosyncratic property of the imperative and negative imperative, and appears in the VP when there is an overt auxiliary do, as in the WDY construction. The robust correspondence between these morphosyntactic and semantic properties is recruited by other constructions that share these properties, thereby licensing don’t be in a broader set of syntactic contexts.

Recall that on the present analysis, there is no construction that specifically licenses agentive be. We assume, however, that the licensing conditions for using the bare form with don’t to express directive force in the well-established constructions (47) and (48) of English are available to speakers to express directive force in other contexts. The possibility of bare don’t be in the negative imperative and the weak directive force of WDY together contribute to the novel construction why don’t you be that inherits the licensing conditions of (47) and (48). This construction is stated in (49).

Further extensions of (47) and (49) preserve the weak directive force but relax or modify one or more licensing conditions of the source constructions; the results are IYDB (50) and HDIB (51). We do not attempt to specify the interpretive details.

Ad hoc extensions of the directive force of these constructions allow for the occurrence of cases such as why don’t I/we/they be …; why doesn’t she/he be …, and even why didn’t I/you/she be …, if she/he doesn’t be …, how does one be …, etc. The extensions that are judged more acceptable are those that preserve more properties of the prototypical cases, and clearly support a directive interpretation.

The approach sketched out here goes beyond a derivational analysis such as Collins (Reference Collins, Deumert and Durrleman2006) because it explicitly addresses the correspondences between form and meaning at a level of detail where the actual phenomena can be accounted for. The capacity to refer to fine-grained licensing conditions is a hallmark of the constructional approach. In contrast, Collins’ analysis predicts that all uses of lexical be that satisfy the two constraints mentioned above (agentive predicate and no inflection) will be equally acceptable.

Moreover, a constructional account is compatible with the view that judgments of acceptability reflect not only satisfaction of formal licensing conditions imposed by the grammar – that is, the set of constructions – but frequency, usage and constructional similarity (see, for example, Bybee Reference Bybee2010, Reference Bybee, Hoffmann and Trousdale2013; Culicover et al. Reference Culicover, Varaschin and Winkler2022). So, for example, the question why didn’t he be more careful does not satisfy the explicit condition of (48) that the agent be the addressee and that the form be don’t. Extension of the directive force with respect to a third person in the past is marginally possible, as illustrated by the corpus examples in (24) above. The relative difficulty of constructing a plausible context accounts for both the judgment of unacceptability out of context and the very rare occurrence of such a form.

7. Theoretical implications

To conclude, let us take a step back and consider briefly the theoretical implications of lexical be. One might, in fact, wonder why it is worth studying such a minor construction, since its frequency in spoken English is on the order of one occurrence per million words. We believe that lexical be is significant because it constitutes a classical instance of a ‘syntactic nut’. That is, it is an idiosyncratic construction that licenses a highly restricted form with a certain interpretation or function.

While syntactic nuts are idiomatic, they may bear close relationships to the more general and productive constructions of the language,Footnote ³⁰ as we have argued for WDYB, IYDB and HDIB. A very different case that illustrates this point particularly clearly is the acceptable (52). What is idiosyncratic in this construction is that the direct object precedes rather than follows the verb, contrary to the general licensing conditions of the English VP in (46).Footnote ³¹

It is possible to adapt the pattern of (52) to novel situations, with a distinctive rhetorical force and gradient acceptability. For example,

Given the acceptability of (52), examples like those in (53) can become more acceptable with repetition, or if they are understood as wordplay. Of course, corresponding examples in which the verb precedes the direct object are all undistinguished.

Constructions such as these are theoretically significant because speakers have clear intuitions about their acceptability, although they can be extremely idiosyncratic. The challenge is to explain how they fit into grammatical theory. It is not sufficient to claim that syntactic nuts like the lexical be construction are not part of the ‘core’ but rather belong to the ‘periphery’. Indeed, these constructions raise the typical learnability problems that are claimed to characterize core phenomena. Specifically, in the absence of negative evidence, there is no reason for a child exposed to (1a) not to simply assume that be can be used as a lexical verb with actional predicates in general, which would lead her to conclude that all the cases in (9) are perfectly well formed. Why is it then the case that speakers do not produce the full range of possibilities?

This approach rests on the view that judgments of acceptability reflect more than satisfaction of the licensing conditions imposed by the grammatical constructions. The constructions require that an expression satisfy not only narrow licensing conditions on syntactic form, but also conditions governing correspondences between syntactic form and phonological and semantic representations. So, for example, why didn’t he be careful is not strictly speaking ill-formed; don’t+V[bare] is a licensed syntactic form in English. But this example does not satisfy the more idiosyncratic licensing conditions imposed by the constructions WDYB, HDIB and IYDB and so will be judged as unacceptable out of context. However, some examples may be judged as marginally acceptable in context by extension of the broader licensing conditions of the well-established constructions, by assigning an interpretation of directive force and dropping the condition that the subject be the addressee, both conditions on why don’t you be.

As noted in our discussion of Collins’ account of lexical be in Section 5, syntactic theories couched entirely in terms of general principles are not capable of accounting for what speakers actually know. Such phenomena provide strong evidence for a non-modular view of grammar in which constructions are ontologically central. On this view, knowledge of language consists of constructions that express correspondences between sound and meaning, mediated by syntactic structure. These constructions range from the very general to the very idiosyncratic. The natural state of affairs on such an approach is that grammatical constructions of a language are related to each other, in that more specific constructions acquire properties from more general ones – they take on the shape of non-idiomatic phrases of the same category, or deviate minimally from them (Culicover and Jackendoff Reference Culicover and Jackendoff2005; Culicover et al. Reference Culicover, Jackendoff and Audring2017). We believe that addressing knowledge of language at this level of specificity is essential to ultimately achieving a full understanding of how language works, how it is acquired and how it changes, and how it is represented in the human mind.