2 Close and loose apposition

The aim of this section is to briefly collect the scattered arguments which show that so-called close and loose appositions are in fact two radically different things. In particular, it will be seen that paradigmatic loose appositions like Chaucer, the medieval genius, are constructions which specialize in arresting syntax to code only pure schematic emphatic presuppositions (not assertions) of the form X [is] Y. In that sense, such appositions are not even syntax as we ordinarily know it. By contrast, so-called close appositions can only be seen as instances of more or less ordinary NP structure.

Koktová (Reference Koktová1985: 39–40) was among the first modern grammarians who proposed the view that canonical loose appositions ought to be seen as pragmatic ‘insertions’ which do not belong to the syntactic structure of a sentence and which, accordingly, do not contribute to the proper assertion expressed in the sentence (see also Burton-Roberts Reference Burton-Roberts1975: 409–10). Such appositions code autonomous messages that abandon the ‘main stream of communication’ (Koktová Reference Koktová1985: 41). This explains their logical properties, in particular their not contributing to the truth conditions of the sentence. In (29B) below the pronoun that may refer to any of the embedded predications of speaker A's turn (the thinking, the saying). However, the autonomous predication that Jacques is a distinguished writer is not affected by the negation in B's turn:

1. (29) A. Joseph thinks that Sofía said that Jacques, a distinguished writer, is an impostor.

2. B. No, that is not true.

As (30) below shows, quantification illustrates the same informational encapsulation of the second constituent in paradigmatic loose appositions: it is not within the scope of subject NPs (Emonds Reference Emonds1979; McCawley Reference McCawley1982). Finally, in the face of (31)–(33), Hannay & Keizer (Reference Hannay, Keizer, Mackenzie and Gómez-González2005) argue that loose appositives are just ‘holophrases’, that is, autonomous units in discourse:

1. (30) *One/*Every/*No friend of mine, the owner of a Cadillac, is coming for a visit.

2. (31) Midland, once the greatest bank in the world, has forfeited its independence.

3. (32) What will Mary, John's wife, say when she hears about this?

4. (33) Don't forget to invite Sue, Peter's brother, to the party.

In (31) once suggests that the segment containing it is a full predication (Doron Reference Doron1992; McCawley Reference McCawley1998: 467–78). In (32) the matrix clause is interrogative while the appositive predication has a different illocutionary force. Something similar may be said of (33). It is of course difficult to think of one constituent whose internal structure accommodates subconstituents with independent illocutionary forces. In short, loose appositives are strongly characterized by pragmatic autonomy. What all the previous examples show is that such an autonomy grants them a level of representation that is distinct from that of the containing sentence (with which it must be inevitably linearized anyhow). The level in question is that of the speech act itself, that is, the same metalinguistic stratum occupied by sentence adverbials like between you and me, honestly, or in a nutshell.

None of the previous signs of informational encapsulation apply to close structures like the poet Burns or the painter Lorraine. Regardless of the constituency of these strings, it seems clear that their integrated nature (Huddleston & Pullum Reference Huddleston and Pullum2002: 447ff. and 1350ff.) causes all of the previous tests either to turn non-operative, or to yield different results. In principle, relative to (29) above, for instance, (34) below might also pass the negation test:

1. (34) A. Joseph thinks that Sofía said that the poet Burns is an impostor.

2. B. No, that is not true.

However, the segment the poet Burns is not meant to emphatically predicate of a previously identified poet that his name is Burns. Rather, the construction seems to have as its job the natural and established (phrase-level) identification of an individual by his prominent social role. In this sense, the fact that the negation in B's turn does not affect such an identification is not different from its not affecting great Burns in the great Burns either, which is of course not an apposition. The point is, the apposition in (29) involves an emphatic predication, as the acceptance of all kinds of verbal modifiers like once and an independent illocutionary force in similar structures show. And even so, the predication is not touched by negation. This is what makes the construction in question unique. By contrast, as (35)–(37) below show, the two nominals forming close appositions are integrated into an indivisible whole that is picked up by pronominalization, quantification and VP ellipsis in toto. The question of whether the poet Burns has two parts does not even arise:

1. (35) John sold the poet Burns a lemon, and Max did too (sell the poet Burns a lemon; ?sell the poet a lemon; ?sell Max a lemon).

2. (36) One/Every/No friend of mine loves the poet Burns (compare (32)).

3. (37) The poet Burns has announced his refusal to join the bill. He is the one that brought up all the issue in the first place.

4. B. *No, it was the one Smith (meaning the poet Smith).

5. C. *No, it was the actor one (meaning the Burns who is an actor).

Additionally, (38) below clearly illustrates that close appositives do not code predications of the same calibre as those in LA since they do not normally admit predicational adjuncts like once (McCawley Reference McCawley1998: 469).Footnote ³ (39) makes it clear that so-called appositive markers (Burton-Roberts Reference Burton-Roberts1975; Acuña-Fariña Reference Acuña-Fariña2000; Lekakou & Szendröi Reference Lekakou and Szendröi2007) are also barred from the construction. And (40)–(41) show that LAs can appear in series, while CAs cannot (Fugier Reference Fugier1973: 97ff.). Finally, the separate illocutionary force of the two nominals that was evident in (32) and (33) above cannot even be tested in close types.

1. (38) *The (once) poet (once) Burns has not said anything.

2. (39) *The poet namely/to wit Burns.

3. (40) Sean Rambo, the wrestler, the pride of New Orleans, the man with more ties with the audience, presented a horrific performance.

4. (41) *The wrestler pride personality of the year Sean Rambo.

As already suggested, the reason why the previous tests either do not apply or produce different results in CAs is that in order for them to apply one would need to conceive of two parts of the appositive construction: one affected by quantification or ellipsis, the other not. Then, if two parts are needed, one of them would have to be ‘the poet’. But, to quote Burton-Roberts (Reference Burton-Roberts1975: 399), unlike in (29), in the poet Burns ‘In what sense can Burns be described as “THE poet”? Which poet?’ It seems clear that in (29), Jacques, a distinguished writer, Jacques is grounded in the discourse (that is, presumed to be an active reference and a topic) before the appositive comment (that he is a distinguished writer) is processed. This has nothing to do with the first unit being a proper noun. If we say Have you met the poet, Burns?, apart from the fact that Burns is not questioned (i.e. has its own illocutionary force), the encoder is presenting ‘the poet’ as a grounded referent via anaphora. That is, the determiner points backwards in discourse in search of specificity for its reference. In (34), by contrast, the definiteness of the whole phrase the poet Burns is not reached backwards, via anaphora: rather, it is the construction as a whole that builds reference ex novo. In short, the different referential properties of CA and LA are appropriately captured by Lekakou & Szendröi (2007: 6): ‘not only do close appositives lack a uniqueness presupposition for their first subpart; what picks out a unique referent is the construction as a whole’ (emphasis added; see also Kolliaku Reference Kolliaku2004: 274–5).

One must conclude, therefore, that the intonational detachment that characterizes LA, but not CA, is not a stylistic extra but something integral to the construction. The pauses separate the speech act conveyed by the sentence from the interpolated speech act conveyed by the apposition.Footnote ⁴ In this sense, the apposed segment simply arrests the otherwise fluent development of the sentence structure to add, on a different level of speech, that the grounded nominal just mentioned is something. Once that clarification is made, the second pause signals that the sentence structure can now be resumed. Unlike other interpolations, LAs are specialized in somehow foregrounding the backgrounded information that U1 is U2 (see Potts Reference Potts2005 for a pragmatic analysis of the parenthetical nature of LA). In sum, LAs arrest syntax to code pure emphatic presuppositions. This is the schematic meaning of the construction, and everything else that specific LAs may express must be worked out in context. Since intonational detachment is absent in CA, none of the twelve types or subtypes of CA in the introduction can code the same meanings as their corresponding LA lookalikes, even if these exist.

3 The poet Burns

The aim of this section is to discuss the constituent structure of the construction illustrated by the poet Burns or the writer Thomas Pynchon, a construction which has acted as a ‘prototype’ for the class of close apposition structures in the literature during the past century. Traditional grammarians devoted quite a lot of attention to a clarification of the internal structure of these seemingly bipartite or ‘binominal’ strings (Keizer Reference Keizer2007a: 22). Interestingly, the semanticist orientation of these pioneering approaches reveals how easy it is to argue, on notional grounds, for any one of at least three diametrically opposed positions: namely, that this construction has the first common noun as head, that it has the second (usually proper) noun as head, or that it has two heads. Usually, whatever structure is assumed for this kind of CA is extended to all the other close appositive types like my friend Steve, the opera Carmen or we boys. It is noteworthy that only very recently have some linguists felt the need to insist that maybe not all CAs are structurally the same (Acuña-Fariña Reference Acuña-Fariña1996; Keizer Reference Keizer2007a). Here I assume that whatever characterizes the poet Burns in particular need not necessarily characterize other close appositive types such as we architects or the word ‘golf’.

Poutsma (Reference Poutsma1904: 183ff.) discussed close appositions in a section of his grammar entitled, revealingly, ‘Attributive adnominal adjuncts’, where he argued that the first member of constructions like the poet X is the head. Lee (Reference Lee1952: 269) shared the same view on the grounds that U2 is ‘necessary to limit or define the limits of the first’. In a reply to Lee, however, Haugen (Reference Haugen1953: 165–6) insisted that the head of the construction is not the first but the second noun because the latter denotes ‘the thing we are talking about’. Finally, in an influential paper, Hockett (Reference Hockett1955: 99ff.) made a theoretically controversial move in suggesting that close apposition is an endocentric structure with two heads, instead of only one. Accepted in passing (more often) or explicitly (less so), this soon became the classic formulation of apposition, whether close or loose. It had the following criterial requirements:

1. i The constituent must be endocentric.

2. ii The ICs must belong to the same major form-class.

3. iii There must be no more justification for taking the first IC as attribute to the second one than for the reverse.

4. iv The ICs must refer to the same entity.

If Hockett's criterial properties for close apposition could clearly be met by a sizable number of specific constructions, and if these were distinctly different from other constructions, the indeterminacy surrounding the notion of apposition would never have arisen or reached very far. The problem has always been to find a large enough pocket of real instances of such constructions, and to demarcate them tightly from similar but distinct constructional types. In particular, conditions (iii) and (iv) have proved too open to interpretation, to the present day. In essence, the same three basic theories of close apposition exist today. I will briefly review them in turn now, as each position reflects a different point of view, and it seems likely that the century-old issue that CA has become will have to be contemplated from as many angles as possible if we aim to reach a more balanced view of it.

Burton-Roberts (Reference Burton-Roberts1975) tried to derive close appositions like the poet Burns from underlying relative clauses of the form: [det] Burns [det] Burns be poet → [det] Burns WH be poet (relativization, obligatory). Although the evidence he adduced was internal to the theory of generative grammar of the 1970s, his basic claim that the head of the construction is the proper noun was adopted in Acuña-Fariña (Reference Acuña-Fariña1996). The idea is that the proper noun imposes conditions on the whole construction, a fact that usually indicates head status. Acuña-Fariña used (42)–(45) below to show that when proper nouns are modified they usually take the definite article in a mould that seems to easily accommodate the proper-noun-as-head view (compare: the excellent Burns/the poet Burns). He added (46)–(51) to show that the definiteness of the proper noun imposes restrictions on indefinite additions (even though instances like (50) are acceptable in a marked, contrastive way):

1. (42) Burns

2. (43) The excellent Burns

3. (44) The poet Burns

4. (45) The Burns who came here yesterday

1. (46) *The Burns

2. (47) *A Burns

3. (48) *Poets Burns

4. (49) *A poet Burns

5. (50) *An excellent Burns

6. (51) *A Burns who came here yesterday

Consistent with the proper noun being the head, it follows that the ‘adjective-like’ common noun (Aarts Reference Aarts2007: 129ff.) in U1 position can hardly be elaborated (52)–(54), just like ordinary modifying nouns (55)–(56):

1. (52) *The poet of the decade Burns will host the ceremony

2. (53) *The poet distinguished by critics Burns . . .

3. (54) *The author of ‘Go Down Moses’ William Faulkner . . .

4. (55) A white rubber hanger

5. (56) *A white rubber of high quality hanger

On this view, the only difference between the poet Burns was white and the poor millionaires were white is that, if we apply the omission test to both, the poet and Burns happen, quite by chance, as it were, to yield two acceptable strings (the poet was white + Burns was white), whereas the poor millionaire does not (the poor were white+ millionaires were white). In fact, the poor were white and the millionaires were white do happen to be valid strings, but, as their meaning shows, they are not the same strings as those found inside the specific construction the poor millionaires. Acuña-Fariña (Reference Acuña-Fariña1996: 29ff.) claims, like Burton-Roberts (Reference Burton-Roberts1975), that if we extract Burns from the poet Burns we change the scope and thus the function of the definite article, making it refer backwards to a previous context, instead of forwards to Burns. In effect, we are changing the meaning of the construction and thus saying nothing relevant about the poet as it stands inside the specific construction the poet Burns. Of course, if we omit the proper noun, all the restrictions to the expansion of the common noun that were evident in (52)–(56) disappear (e.g. the poet of the decade will host the ceremony). Only grammatically motivated pauses can salvage such deviant forms, provided it becomes clear that the NP refers to a unique, and as such presumably identifiable, referent: the poet of the decade, Burns, will host the ceremony. That is, provided we end up using a different construction: loose apposition.

Keizer (Reference Keizer2007a: 31–2) agrees that ‘whether or not a sentence is acceptable, syntactically speaking, after the proper noun has been left out is irrelevant: it may be acceptable, but the structure of the construction will have changed, with the determiner and the element poet appearing in a new configuration’. However, for her the head of the structure is not the proper noun but the common noun, which does not even project reference: it is rather the combination of the two that does so: the + [poet Burns]. Specifically, Keizer rightly points out that it is not always true that proper nouns refer uniquely, since they are often used as common ones, which means that the restrictions that Acuña-Fariña highlights (46)–(51) may easily disappear. In the first place, she offers examples like (57)–(58) below to show that indefinite determiners can occur with proper nouns when these lack a specific reference:Footnote ⁵

1. (57) Every now and then a Forsyte would come up, sidle round, and take a good look at him.

2. (58) Do you know an Alice?

Secondly, she adduces instances like (59)–(60) to show that structures containing a modified proper noun are also possible:

1. (59) An embarrassed Sir Patrick Mayhew, QC, the Attorney-General, . . . had to cancel it after only 48 hours . . .

2. (60) By 1918, Croatia's political class yearned for a Yugoslavia where an extended Serbia would have been balanced by a consolidated Croatia.

Finally, and more importantly, she found in her corpus some examples of indefinite close appositions like (61)–(62):

1. (61) He also has to put up with a soppy elder brother Robert who is forever mooning about some girl or other and a sister Ethel who has all the brisk no-nonsense superiority of a true Wodehouse gel.

2. (62) I have a friend John who's in Linguistics.

As for the restrictions on the expansion of the common noun (see 52–56 above), Keizer finds that Acuña-Fariña's observations are incorrect since ‘the first element in these constructions can be modified, by PPs as well as by relative clauses’ (2007: 14). These are some of the examples she adduces:

1. (63) Uhm let me bring in the Conservative leader of the county council Ken Thomson.

2. (64) The most influential writer on the English constitution Walter Bagehot warned that daylight should not be let in on the magic of the monarchy . . .

3. (65) And medical attention will immediately go out to the rider who fell there causing the stoppage of that race Mullet.

There are a number of qualifications to make regarding Keizer's observations. In the first place, in my view, if one uses a structure like a Forsyte to make a point about the grammar of proper nouns, then instances like (57) and (58) do not in theory undermine the proper-noun-as-head view since in a Forsyte it is clear that no other (common) noun can be the head (so Roberts could still be the head in the hero Roberts, for instance). Then, in theory, headedness aside, what examples like (57) and (58) would prove at most is that there is no restriction on definiteness imposed by the proper noun. In the second place, however, even this is not such an inevitable conclusion. Keizer points out that the proper-noun-as-head theory, which is based on the restrictions imposed by U2, fails because the examples she adduces (all with indefinite determiners) are acceptable. Yet, she must at the same time make the point that, in order for them to be acceptable, the proper noun must be recategorized as a common noun (see Aarts Reference Aarts2007: 130 on decategorization). If that is the case, the fact remains that all the restrictions still come from the second noun.Footnote ⁶ In the third place, some, if not most, of the examples she has found in her corpus strike one as somewhat odd if they are meant to be close structures like the architect Norman Foster. This is particularly true of the specific examples of CA that she mentions, like (63)–(65) above. Keizer does not mention how often such examples appear in her corpus, but in my view such examples merely show a rather deviant use of punctuation (more appropriately: a lack of it). One would have to compare how many times those long NPs are pronounced in the way that the lack of commas suggests that they should be pronounced, to the number of times such heavy units incorporate some kind of pause or tonal boundary (it is not simply a question of pauses).Footnote ⁷ In my view, most of these examples are only natural if the second unit in them is set off from the first by some demarcating use of tonicity. In the absence of a fine-tuned analysis of the prosodic contours of such strings and the absence of fine-tuned statistics of the use of such forms relative to all other appositive forms, it seems appropriate to maintain the view that such sporadic evidence ought not to be given such a decisive role in the determination of the final analysis of this construction. This by no means entails viewing corpus linguistics as an invalid form of linguistic analysis.

The issue of whether the first unit in CAs can be elaborated or not is in fact a little more interesting. One of the reasons why Acuña-Fariña suggested that U2 is the head is that not only is the first unit − more often than not − incapable of elaboration, but, when elaboration seems possible, it must be compatible with U2 (even though it occurs adjacent to U1). Thus, it seems that we can say the excellent poet Burns or the great painter Picasso because we can also have the excellent Burns and the great Picasso, but it sounds odd to say ??the deaf poet Burns, or ??the short painter Picasso because we then seem to be driven to construe ?the deaf Burns and ?the short Picasso, which, without adequate pragmatic remediation, strike one as strange. Thus, in the face of this, the fact that we can say an embarrassed Sir Patrick Mayhew (see (59) above), as Keizer points out, merely confirms the fact that proper nouns can be modified (= the excellent Burns) by a restricted set of modifiers. This is in principle independent of the grammar of CA. However, I can think of two reasons why the grammar of CA is a different issue: in the first place, it is hard to imagine anyone saying ??an embarrassed actor Tom Cruise (compare: an embarrassed Tom Cruise). This shows, among other things, that in the presence of indefinite grounding, not even adjective compatibility with the proper noun is enough, and suggests that the construction as a whole is rather fixed (see next section). In the second place, the restrictions on the elaboration of the first noun that we saw in (52)–(56) are – in my view − very real for most speakers, since cases like (63)–(65), adduced by Keizer, are either extremely sporadic or deviant in the use of punctuation or tonicity, as argued above. So we do still need to account for the restrictions on the elaboration of the common noun, and contemplating such syntactically unmalleable units as heads makes all those restrictions even more mysterious. Momentarily, therefore, one must conclude that the second constituent of CAs is given more profiling power than is expected of modifiers (Langacker Reference Langacker1987: 183–9) – or, alternatively, that the first is given less than is expected of heads.Footnote ⁸

Finally, a modern version of Hockett's double-head theory of close apposition is defended by Lekakou & Szendröi (Reference Lekakou and Szendröi2007) in a recent account of close apposition and Greek polydefinites. Lekakou & Szendröi base their thesis on a distinction between LA and CA. Since in LA only the first ‘part’ comes with a uniqueness presupposition and the second part is a parenthetical predicate, it makes sense to believe that in CA reference to a unique entity is determined by the two NPs (DPs in their approach) together. In CA what picks out a referent in the world, according to Lekakou & Szendröi, is the construction as a whole, and both parts contribute to reference. Since neither NP is a predicate,Footnote ⁹ the two NPs must be combined in an operation that Lekakou & Szendröi call ‘complex argument formation, whereby the R(eferential) roles associated with the two DP's are identified’. This results in a ‘symmetric’ relationship as illustrated in (66) below:

1. (66)

In their own words:

So in their view the ‘two R-roles undergo identification – an operation which Higginbotham independently introduces for adjectival modification’. The reader may have noticed that the line of reasoning underlying Lekakou & Szendröi's theory is the same as in the classic work by Hockett (Reference Hockett1955): whenever there is ‘no more justification for taking the first IC as attribute to the second one than for the reverse’, it seems reasonable to suggest that both are heads.

One important feature of Lekakou & Szendröi's analysis is that they take the branching the poet + Burns for granted (cf. ‘the intuition that both DPs would be independently well-formed’). It is important to bear in mind that in their account both NPs are referends, because two important predictions (which they themselves make) follow from this: the first is that since the essence of CA is R-role identification, it must involve nominal phrases, and only nominals have an R-role. The second is that since in CA both units jointly refer through R-role identification, CA will not be possible when the two NPs are identical (that is, when they pick out the same referent). The first prediction might be contradicted by instances like here in London, or now in the summer, which can clearly be pronounced without intonational detachment, and which are usually regarded as appositive (see Taylor Reference Taylor2002, for instance). The second prediction is a more important one, and is actually a restatement of Burton-Roberts's (1975: 396ff.) comments on the difference between CA and noun compounds like fighter-bomber. According to Burton-Roberts, unlike unacceptable instances of CA like *man homo sapiens or *the Bard Shakespeare, a compound like fighter-bomber is acceptable because both members of the compound form a superordinate entity that is distinct from both of them. This synthesis is impossible in the case of unambiguously co-referential NPs in CA because:

These observations also bring to mind Williams's (Reference Williams1982) i-within-i principle, which, according to Chomsky (Reference Chomsky1981), is not allowed within arguments. Notice that in (67) below the whole NP cannot have the same referential index as one of its parts. That is possible in direct predications like (68), as well as in the appositive predication in (69). Notice, interestingly, that LAs behave like main predicates, and not like arguments, in allowing i-within-i:Footnote ¹⁰

1. (67) *[His_i own worst enemy]_i lost the election again.

2. (68) John_i is [his_i own worst enemy].

3. (69) John_i, [his_i own worst enemy]_i, lost the election again.

It seems to me that the restriction on identical reference of the two Ns is more of a stipulation than an actual constraint, the reason being that reference to the same individual is not, in and of itself, incompatible with a linguistic (i.e. formal) construal of the same individual from two parallel angles (e.g. that someone is a poet by profession, that s/he has a name, etc.). In fact, reference, an inherently discoursal phenomenon, is usually built up in that way (Levy Reference Levy and Givón1979: 193). It might therefore be useful to make a distinction between identical reference and identical denotation. The latter would be subject to Burton-Roberts's and Lekakou & Szendröi's restrictions on ‘NP synthesization’, and it would also violate Goldberg's (Reference Goldberg1995) Principle of the No Synonymy of Form (Bolinger Reference Bolinger1968: 127; Givón Reference Givón and Slobin1985; Wierzbicka Reference Wierzbicka1988). The former would not. The actual restrictions on usage (*the Bard Shakespeare, *the author of the century Burns, *the man over there Smith) are thus best explained if we assume that the grammar of the NP, whose primary objective is to project reference, is such that only one referential projection may exist in it at the top. Seen in this light, the restrictions follow not from the coexistence of two NPs with identical reference in an integrated construction (Huddleston & Pullum Reference Huddleston and Pullum2002: 1350ff.), but rather simply from the existence of two referential NPs. In the next section, I will argue that the restrictions on the expansion of these forms, together with their strong conventionalization, are the way grammar has of protecting itself against a fuzzy internal referential constituency. That is, grammar either imposes constraints aimed at avoiding referential competition inside the same nominal phrase (*the poet of the decade Burns), or it sanctions forms which are very constrained formally (like most close appositions or structures like the poor) to ensure object recognition despite weak, and even conflicting, internal elaboration. Additionally, given our comments above on the scope of the determiner and the absence of backwards anaphora, a branching like the poet + Burns in which the poet is a full referent seems highly problematic.

4 Inchoate NPs

The three views sketched and commented on in the previous section (the + poet Burns, the + poet Burns, the poet + Burns) do not in theory exhaust the possibilities of analysis. A fourth view would be to defend Lekakou & Szendröi's branching approach but analysing the poet as a predicative NP and Burns as the head. This would entail symmetry in constituent structure but asymmetry in referential power. Needless to say, definite NPs cannot refer in certain configurations, such as U2 in LA (Collins, the prey of violent and obscure emotions, threw himself at the gun; see Burton-Roberts Reference Burton-Roberts1975). However, it would be difficult to find evidence for such a novel analysis of (integrated) argument NPs.

A fifth potential view would be to proceed in the opposite direction and, while keeping the NP + NP branching, maintain at the same time that non-anaphoric reference is projected from the common noun, while the proper noun simply serves an appellative function without proper reference. On this view the proper noun would simply be used to restrict the reference of the head, like a typical modifier. However, the same effect is obtained if we assume an analysis like Keizer's, one in which the branching is rather the + poet Burns, and this has the at least superficial advantage of resting on habitual NP structure (the + black poet; the + poet of the century; the + kitchen table).

While it is possible that, after a century of disagreement, any one of the previous analyses may end up being preferred over the others, this does not seem likely. I suggest that rather than solid evidence for any of those analyses, we only have rather solid evidence for the lack of clear constituency in paradigmatic close appositions like the poet Burns. To be more precise, we seem to have scattered evidence for the view that U1 is the head and also scattered evidence that U2 is the head. This alone is often taken as ‘evidence’ that both are heads, but that is a nonsequitur. There is another option that no one seems to have considered, and it consists in claiming that the constituency of these strings is simply unresolved. Before I suggest how and why this anti-Aristotelian state of affairs might have arisen, let us briefly review the facts regarding the CA under analysis and add a few more:

1. a) The profiling power of the proper noun, U2, is suggested by:

   1. 1. The fact that it imposes selection restrictions on the entire NP. These restrictions are clear in the case of accompanying adjectives, in that they tend to be the more easily accepted the more they unambiguously point to U2. These restrictions are in fact even clearer regarding definiteness in the sense that the article must be definite. In the much more infrequent cases where the indefinite article is used, the proper noun must be recategorized as a common one (Keizer Reference Keizer2007a).

   2. 2. The fact that the first noun, U1, is severely constrained in its elaboration potential (e.g. *the poet of the century Burns, *the poet Burns of the century; */?the deaf poet Burns). In this, it behaves like ordinary modifying nouns like kitchen in the kitchen utensils we all need (*the kitchen of design utensils we all need). U2, by contrast, is not so constrained by the construction: it co-occurs with determiners and adjectives with which it tends to occur outside CA: theBurns that you know, the excellent Burns.

   3. 3. The stress pattern is secondary + primary (Haugen Reference Haugen1953; Keizer Reference Keizer2007a: 24), as in the cotton shirt or the blue ray.

2. b) The profiling power of the common noun, U1, is suggested by:

   1. 1. Pluralization: it is much more natural to say the biologists Huxley than the biologist Huxleys (the latter preferentially being acceptable in a contrastive interpretation; Burton-Roberts Reference Burton-Roberts1975; Keizer Reference Keizer2007b: 35ff). This is unlike clear endocentric structures like the lady president, which, when pluralized, yield strings like the lady president and vice chancellor (not *the ladies president and vice chancellor). Also, when instead of one proper noun there are two, the common noun takes the habitual plural inflection: the Lockhead sisters Venus and Beatrice.

   2. 2. Agreement: consistent with the pluralization facts, if a phrase like the biologists Huxley occurs in subject position, formal agreement links the verb and the common noun: the biologists Huxley were invited.

   3. 3. Order of the parts: it seems reasonable to say that just as in Burns the poet, the theme Burns has thematic prominence, so in the poet Burns, the poet receives that prominence as well.

   4. 4. Inheritance and taxonomic ties (Fillmore & Kay Reference Fillmore and Kay1993; Goldberg Reference Goldberg1995). This would require more space than we can afford here (see the introduction, and especially Acuña-Fariña, submitted), but suffice it to say that, if we view the entire collection of CAs as constituting a dynamic network organized around a number of inheritance and taxonomic hierarchies (a network of associations), then attractors in the vicinity of this construction which have rather clearer [head + modifier] constituency would be expected to exert (via partial, shared co-activation) some influence on the constituency of the construction under analysis here (Petitot Reference Petitot, Port and van Gelder1995: 227ff.). The most relevant types include my friend Burns (notice the inadequacy of *my Burns who is a friend) and marked but frequent constructions with indefinite determiners like this bloke Steve (*this Steve who is a bloke), or Keizer's I've a friend John who's in linguistics (*I have a John).

3. c) Evidence that does not distinguish between the two nouns as head:

   1. 1. Pronominalization: in the kitchen table, table functions as head because, among other things, it can be pronominalized: the kitchen table / the bedroom one. So in the red table / the blue one. But, as noted, in CAs like the poet Burns one can only pronominalize the entire phrase using he/she: the poet Burns / *the one Smith; the poet Burns / *the doctor one (meaning: the Burns who is a doctor). Pronominalization is usually a clear indicator of X-centricity, so this fact has great significance.

   2. 2. Distribution: there seem to be equal distributional grounds for integrating both nouns as head in existing productive paradigms: the poet Burns may be seen to illustrate the same structure as that in, for instance, the excellent Burns, but also that in the poet over there.

In the face of such conflicting signs of constituency, it makes sense to conclude that, barring a traditional NP +NP analysis where both NPs have ‘the same privileges of occurrence’ (Hockett Reference Hockett1955: 99ff.) and both have equal referential power (see above),

I suggest that what we see in these strings are traces of constituency pointing in conflicting directions, and that these traces are inchoate due to lack of strong functional pressure. Why would there not be a strong functional pressure to fully elaborate their internal structure? There are three reasons for this: 1. because the construction has as its job the activation of a social referent, and in the social world that we inhabit this is usually done either by name or profession (i.e. not by height, size, colour of the eyes, etc), with no reason to prefer one over the other, and no logical incompatibility between the two; 2. because the construction is a hybrid of distinct and more elaborated templates, which act as attractors (Zanone & Kelso Reference Zanone and Kelso1992; Langacker Reference Langacker2006) and pull constituency in opposite directions; and 3. and more importantly, because the construction is easily identifiable as such ‘from the top’. This makes it unnecessary to have to spend valuable cognitive resources (like creating, storing and deploying inaudible, abstract, constituent structure) when, somewhat metaphorically, we can reach the final destination of that journey (last stop: meaning) directly, as it were, with no changing of trains (Haiman Reference Haiman and Pagliuca1994; Boyland Reference Boyland1996; Hay Reference Hay2001).

The fact that the functional job of close appositions like the poet Burns is to activate a referent in the social world seems intuitively straightforward. There are numerous accounts of the discourse functions of close appositions (Quirk et al. Reference Quirk, Greenbaum, Leech and Svartvik1985; Meyer Reference Meyer1989, Reference Meyer, Aijmer and Altenberg1991, Reference Meyer1992; Keizer Reference Keizer2007b) and they are all compatible with this view. In the most recent account, Keizer (Reference Keizer2007b: 42ff.) speaks of a ‘descriptionally identifying’ use,Footnote ¹¹ with the descriptive element providing ‘information which allows the hearer to relate the referent of the construction to his/her “knowledge base” by anchoring this referent in the (linguistic or extralinguistic) discourse situation. As such, the information provided in the descriptive element of the apposition is meant to forestall questions on the part of the hearer of the type “Who is he/she?”.’ This is the reason why these structures always code professions, occupations and the like. They prosper in the press genre because here writers often have the need to introduce a referent to a community of readers (with little shared knowledge), and social occupation is of course one of the most relevant ways to do that (Meyer Reference Meyer1992).

That the construction is a hybrid of distinct and more fully elaborated templates which pull constituency in opposite directions requires more consideration, but the incompatible features of the construction that we reviewed above do point in that direction. In particular, the selectional restrictions, the strong constraints on the elaboration of the first unit, and the stress pattern may be taken as inheritance ties (Goldberg Reference Goldberg1995, Reference Goldberg2006) from structures of the the famous Burns type. Alternatively, pluralization, agreement and constituent order are like those of mundane NPs with mundane nouns like poet: e.g. the poets of the revolution, the poet distinguished by critics. A positive aspect of a constructionist approach to grammar is that in a representational map of the constituent features of a given construction, the features which are suggestive of a certain analysis do actually co-exist with those suggestive of another, different analysis. If we view these structural biases as activation paths (on the neural basis of meaning, see Feldman Reference Feldman2006: 105ff., and references therein; also Langacker Reference Langacker2006: 141), then the meaning, and the structure, of the construction is quite straightforwardly the entire collection of activation paths. This is a basic form of Hebbian thinking (Hebb Reference Hebb1949): neurons that fire together wire together and thus form a category. In these close appositions the activation paths simply happen to present a more scattered topography than the one we may hypothesize for more frequent and more strongly canalized nominal forms, of the kind captured in X′ accounts.Footnote ¹²

In fusing [the + Modifier + Proper noun] and [the + Head Noun + Modifier] a third construction arises which is different from these two. This is the phenomenon described in Goldberg (Reference Goldberg1995: 72 ff.), whereby two parent constructions with partially conflicting features feed into a daughter construction in such a way that the latter may display selective features of the former in what can only be regarded as a third, unique structural creation. So the poet Burns has all the selective features of the famous Burns, on the one hand, and of the poet of the revolution, on the other, that we reviewed above, but it also has unique features of its own, and these emerge with the construction. The new features, which are the formal reflex of a new symbolic package (a new construction), guarantee functional survival in an ecological niche composed of similar – yet subtly distinct − network companions. In general, hybridity tends to come with rigidity in the sense that the inheriting construction tends to be subject to constraints which did not affect the parent constructions (Aarts Reference Aarts2007: 228ff.). In an ecological environment, such constraints ensure the right object is being recognized. This is evident in the construction under analysis, which is characterized by the following constraints:

1. 1. A very restricted set of nouns available (names of occupations and the like): ??the friend Burns, ??the cousin Burns, ??the man Burns all seem wrong. Compare: my friend Burns, my cousin Burns, where there is no ambiguity as to the scope of the determiner (my friend + Burns). The possessive helps instantiate a different construction, so the restriction on the nouns disappear.

2. 2. A constructional ‘freezing point’ that depends on the referential saturation of the first unit (Acuña-Fariña Reference Acuña-Fariña1996):Footnote ¹³the unique poet Burns is fine (notice the unique Burns), and so is the French poet Burns. The Jaguar boss Tom Walkenshaw seems to be all right as well,Footnote ¹⁴ especially because it is easy to conceive of Jaguar boss as a unit. However, the stubborn Jaguar boss Tom Walkenshaw contains an initial nominal constituent (the stubborn Jaguar boss) with enough elaboration to activate a referent by itself. That is, it starts sounding like a referential NP. In that case, the prosodic dissociation of the newly-created NP from the proper noun becomes more necessary given the syntactic weight U1 has. Hence the restrictions on the expansion of this unit that were mentioned above (Acuña-Fariña Reference Acuña-Fariña1996, Reference Acuña-Fariña1999): ??the Jaguar boss at the time Tom Walkenshaw, ??the increasingly desperate Jaguar boss Tom Walkenshaw, or ??the fans’ favourite Jaguar boss Tom Walkenshaw are, in my opinion, awkward if pronounced exactly like the poet Burns, that is, in a fully integrated manner. Heavier expansions are clearly barred: *the boss of the British formula 1 racing team Tom Walkenshaw. Notice that, as long as the language user intuits that there is only one centre in the phrase, even proper nouns admit rather heavy modification: the rich, distinguished and arch-famous Burns; so the constraints in question are unique to the appositive construction, which involves nouns with referential potential occurring after a form – the definite article − that specializes in activating referential projections.

3. 3. Further restrictions characteristic of conventionalization and idiomaticity: *the writer the poet is ungrammatical despite the fact that the two NPs are far from being identical, which means that there should not be any restriction to ‘synthesization’ on logical or informational grounds. One could conceive of one such restriction in the case of *the poet the writer, as it is easier to use poet to narrow down writer than the other way round. However, both options are prohibited: it is therefore a restriction set by the construction, which does not accommodate two full NPs – at least not with the same grounding mechanism (determiner the). Again, one can say my friend Burns and even my friend the poet, with two different grounding mechanisms (leaving aside now whether the poet is grounded in that construction), but the possessive determiner has scope only over the first noun, which suggests that the constituency of these strings is different: my friend + the poet. This entails the need to list a (related but) different construction. Incidentally, in my friend Burns / my friend the poet there are comparable restrictions on grounding: ??that friend Burns / ??that friend the poet; *a friend the poet; *two friends the poets. Notice also that the can occur in this construction only in the second NP (*the poet my friend vs. my friend the poet). This is because the possessive determiner must project reference (and therefore occur beside the referential head) but the definite article may not do so in ‘predicative’ positions (like after a copula, expressed or not; see below).

   Additionally, as noted, not only is the first common noun incapable of growing to its right, like ordinary modifying nouns (*the marble of Galicia table = *the author of Emma Jane Austin), but even the proper noun, which can usually take PP dependents (the Burns of post-war fame, ‘I'm the Picasso of pain’),Footnote ¹⁵ and appears equally unconstrained to its right in close apposition (that is, at a phrasal limit), seems not to accept such dependents in this particular configuration (*/?I'm the painter Picasso of pain). This suggests a strong degree of conventionalization and entrenchment.

4. 4. Reversibility. Although in a constructionist approach to grammar, such as the one advocated (though not fully pursued) here, the move from the poet Burns to Burns the poet might be seen as a change of construction (especially if it comes with an associated change of meaning), the fact remains that this CA allows such a possibility. It is tempting to argue that there in London cannot be transformed into *in London there because there is a restriction to the effect that the more specific subconstituent cannot come before the less specific one. That that does not happen in the poet Burns shows the ‘traces of equality’ that have long characterized the notion of apposition. Notice that most close appositions cannot in fact reverse their order: we boys / *boys we; this bloke Pete / *Pete this bloke; my friend the doctor / *the doctor my friend (in which case the force of the deictic first person possessive overrides the requirement that the second unit be more specific). We are, again, dealing with emergent properties of one specific construction. It is interesting to notice that when the common noun is plural (the biologists Huxley), reversibility is not possible (*Huxley the biologists; see Burton-Roberts Reference Burton-Roberts1975: 403–4; Keizer Reference Keizer2007a: 38). This is yet another indication of constructional fixity.

Finally, the fact that CAs like the poet Burns can be easily recognized ‘from the top’ also seems straightforward. Bybee & Thomson (Reference Bybee and Thompson1997) point out that routine often makes it unnecessary to specify internal structure, and that it also makes such structure unavailable for analogical extension. In an ecological niche such as the one that might be contemplated for the whole class of CAs, however, having no fine-grained internal structure counts as structure anyway since constructions form an interconnected lattice where each dot in the system is set off from the rest by whatever is unique about it − including the fact that it may not have a fine-grained internal structure, unlike the surrounding dots. In any case, both the token frequency and the type frequency of the CA under study provide one of the clearest indications of constructional status. Thus, the number of times words (tokens) like poet, painter, or writer appear in CA in the press genre is hypothesized to be very large indeed (by comparison with other genres). However, the number of words that appears in the construction (in the type) is extremely low: only those of occupations and the like, and not even all those really show up, as occupations like butcher or carpenter usually lack the social dimension that justifies treatment by the press. If we take type frequency as an indication of productivity (Bybee Reference Bybee1985), it is clear that CAs in general, and especially the construction under analysis, are not very productive as a pattern. The other side of the lack of productivity is that these conventionalized word sequences must be listed more or less like idioms, or like irregular pieces of morphology, a fact which reinforces their holistic representational status and freezes their form (Bybee Reference Bybee1985; Ogura Reference Ogura1993; Smith Reference Smith2001). If to the scarce internal variability of the construction one adds the small number of actual words that make it up (only three in a large number of cases), recognisability ‘from the top’ – the last stop en route to meaning – is never a problem for decoders who face such rather frozen forms (Haiman Reference Haiman and Pagliuca1994; Boyland Reference Boyland1996; Hay Reference Hay2001).

Generally, one can see the need for layered componentiality in language as the need to build composite meaning which is internally heterogeneous. In a phrase like the red table, for instance, it is cognitively essential to count on clues that distinguish between predication and kind; otherwise red might be construed as a kind and the whole phrase projected from there. Valuable cognitive resources must therefore be devoted to untangling the different categorial make-up of the phrase. Aarts (Reference Aarts2007: 233) points out that languages tend to avoid true hybridity because ‘cases where the categorial scales are perfectly balanced are presumably hard to process mentally’ and he mentions instances like I am tired of that feeding the animals all day as a case in point (with the -ing string a mixture of clause and NP).Footnote ¹⁶ He also points out that intersective gradience (defined as the existence of cases which display properties of two distinct categories which are therefore said to converge) can usually be approached discretely by counting the properties that the convergent structures share with the more prototypical ones. Usually, he adds, a perfectly balanced number of properties is avoided for computational reasons, as pointed out above. Now, it is not easy to define a priori which properties two different constructions may have, and especially whether all properties ought to have the same value, but in general it does make sense to believe that language users are better off developing language systems which avoid true hybridity. Ultimately, componentiality is of course very useful − and probably unavoidable for large syntactic objects with intricate internal structure (like many -ing segments). However, language users develop language systems that maximize the expression of meaning, so meaning – not componentiality − is really the final purpose of language. In a short conventional segment like the poet Burns, when poet Burns occurs right after the grounding determiner the, both nouns project kinds of a mutually compatible nature, so there is no strong need to adjudicate headedness to either. Additionally, and importantly, adopting Hawkins's (2004: 32ff.) terminology somewhat loosely, all these strings construct NP by Det. That is, they start with the quintessential determiner, and this automatically projects NP at the top of the structure, thus facilitating the CRD (‘constituent recognition domain’) by EIC (‘early immediate constituency’). This ensures instantaneous object recognition, which means that Aarts's concerns about the pernicious effects of hybridity really do not apply to this class of structures. After all, as Aarts himself points out (2007: 123), ‘if at any point in a particular grammatical process or derivation an element α is required, it does not matter whether α is a good or a bad example of the category it belongs to’, provided its relative ‘badness’ does not percolate upwards in the tree. In fact, in a construction that is conventionally made up of just three parts (the + noun + proper noun), right after the the top of the structure can be reached automatically via neuromotor routinization, and the top is after all what we need to make sense of the meaning. In a way, in the press genre where almost all these highly frequent constructions occur, we process a sequence like the poet ‘something’, or the writer ‘something’, or the architect ‘something’ as an avalanche, that is, as a string of words which usage congeals or precipitates together (MacWhinney Reference MacWhinney2001; see also Acuña-Fariña Reference Acuña-Fariña2005; on the role of specific contextual effects on the activation of exemplar constructions, see Bybee & Scheibman Reference Bybee and Scheibman1999; Scheibman Reference Scheibman2000). Avalanches tend to be rather rigid neural routines. CAs like the poet Burns are in fact rather rigid linguistic forms. The psycholinguistic literature makes it clear that a short phrase, such as for instance the writer Saul Below, which is pronounced in well under a second, is read even faster, and that only two or three saccadic movements of the eyes are needed to go past the entire phrase (Rayner Reference Rayner1998). In fact, with ordinary parafoveal processing (some seven characters to the right of the last landing position), it is probably impossible to fixate the segment the writer without at the same time ‘reading’ the segment Saul Bellow, which means that there is no real scope for starting to construct an NP on the evidence of the writer only.Footnote ¹⁷

Coupled with the form, there is no doubt that all these constructions project one referend at the top too – not two or more. All the restrictions on the expansion of the construction which we have reviewed, and its fixity as a constructional type, testify to the fact that the grammar of English is set to avoid referential mishaps: for instance, the moment the first noun of a structure like the writer Thomas Pynchon receives even minimal elaboration, the language-users’ processing system starts activating a referential path, and this is incompatible with another active referential path ‘under the same nominal roof’ when the name appears on the scene (*the writer of the century Thomas Pynchon, vs the writer of the century, Thomas Pynchon). And, revealingly, while my friend the poet (another construction) is all right, *the friend the poet is not, the reason being, I suggest, that the launches an instruction to locate the reference of the adjacent noun friend in one's knowledge base. However, right after friend, instead of some predication which might help project that nominal against such a base (such as, for instance, over there, or that I told you about yesterday), another instruction of the very same kind both: 1. prolongs the initial search and 2. is compatible with the launching of yet another instruction to locate another referent, while the first instruction is still open and unresolved. In my friend the poet, by contrast, initial my anchors the reference of the phrase to a deictic centre by EIC (early immediate constituency), thus giving it as much discourse grounding as one can possibly expect to find, and thus forces the second determiner to activate a predicative (nonreferential) reading relative to the already-established referential centre. It comes as no surprise, then, that, in line with Aarts's comments on the difficult processability of hybrid strings, the grammar of English judiciously proscribes such indeterminacies as *the friend the poet.

Absence of full, step-by-step, bottom-up specification in the grammar of the NP usually involves two compensatory mechanisms: either listing, which explains why English NPs such as the poor and the Smiths must have a very narrow range of application (the poor people and the Smith family respectively), and redundancy, which explains why in Spanish, for instance, the same constructions can refer to all sorts of things (‘las blancas’, meaning ‘the white ones-(feminine)’ can refer to people, objects, etc.) − as in inflection-rich Spanish, the determiner itself provides the needed nominal cues (so the noun can be dropped rather freely; Hawkins Reference Hawkins1994, Reference Hawkins2004). Most close appositions in English (not just the pattern under scrutiny) must be listed. The listing includes a set of formal peculiarities and constraints that guarantee unambiguous object recognition, as well as a specific discourse function for each (Keizer Reference Keizer2005, Reference Keizer2007b).Footnote ¹⁸ Linguists often argue about the raison d'être of seemingly arbitrary constraints (compare we girls and *they girls). For present purposes, it is surely worthwhile to point out that in a network of interconnected dots (as I believe can be posited both for close apposition in general and grammar in general), all that matters is that each symbol be associated with a specific form. The constraints and the formal fixity thus prevent the system from generating and processing wrong forms by mistake. Two other things are needed to avoid massive ambiguity: in the first place something similar to Goldberg's (Reference Goldberg1995) Principle of the No Synonymy of Form (Bolinger Reference Bolinger1968: 127; Givón Reference Givón and Slobin1985, Wierzbicka Reference Wierzbicka1988); and, in the second place, a memory far more powerful than was assumed in the 1960s, when, with acquisition in mind, it seemed reasonable to generate absolutely everything using componential, step-by-step, bottom-up rules (Lieven, Pine & Baldwin Reference Lieven, Pine and Baldwin1997; Tomasello Reference Tomasello2000). Today, given what we know about memory and the role of frequency in cognition in general (MacDonald, Pearlmutter & Seidenberg Reference MacDonald, Pearlmutter and Seidenberg1994; Hare, McRae & Elman Reference Hare, McRae and Elman2004), it makes sense to believe that it is actually more wasteful for the mind to generate the same percept all over again, especially if what is to be generated has already been generated a million times in the past. In such circumstances, all that is needed is recognition and retrieval, of structure, for notice that absence of fine-grained internal structure by no means entails absence of structure: all these strings are clear, unambiguous NPs at the top. In sum, what is needed is a system much like our reading system: that is, one which normally accesses words in toto while still preserving – perhaps in parallel – the ability to access them phonologically (see Altmann Reference Altmann1997: 160ff., and references therein). In such a flexible system of grammar, when meaning can be reached safely anyway, componentiality need not be obligatory, and may even be cumbersome.

In short, nothing in the previous remarks entails the view that componentiality, no doubt the default (though surely not the only) route towards meaning in grammar and in syntactic processing, is not essential and therefore unresolved in all kinds of short constructions. It simply is not inevitable, especially when concurrent circumstances militate against investing in it. In the case at hand, the frequency of the construction tied to a specific genre, its hybrid componency and short size, its rather frozen form, and its easy formal recognizability do militate against it. In this sense, CAs like the poet Burns need even less internal structure than many idioms like to pull somebody's leg or to kick the bucket, in the sense that in these, regardless of the final composite meaning of the construction, the relationships among the constituent parts are still transparent and momentarily ambiguous between a literal and an idiomatic interpretation (Smolka, Rabanus & Rösler, Reference Smolka, Rabanus and Rösler2007). In CAs like the poet Burns, they are not. To end on a metaphorical note again, such strings can afford the luxury of dispensing with the nuisance of most of the connecting stations by using direct trains most of the time.

5 Summary and conclusions

Close appositions are probably best understood if they are seen as forming a rich ecological niche where each construction relates to other constructions forming a dense network of taxonomic and inheritance hierarchies. The structure of that network, however, falls outside the scope of the present work, where I have focused on the concept of close apposition itself, on how it deviates from that of loose apposition, and on an analysis of the quintessentially close appositive construction typified by the example the poet Burns in the literature (Curme Reference Curme1947; Lee Reference Lee1952; Fries Reference Fries1952; Haugen Reference Haugen1953; Hockett Reference Hockett1955, Reference Hockett1958). The thesis that these strings are formed by a doubly endocentric structure where the first segment (the poet) is a grounded nominal (e.g. Hockett Reference Hockett1955; Langacker Reference Langacker1991; Taylor Reference Taylor2002) has been rejected. Instead, it was argued that these highly conventionalized close appositions are instances of ‘inchoate’ noun phrase structure, and that the internal composition of such strings is not fully elaborated due to lack of strong functional pressure. Additionally, it was argued that a lack of rich internal structure by no means entails a lack of structure since all these constructions project an unambiguous NP node at the top of the tree. Since NPs are primarily reference-activating forms, fuzzy internal constituency is less of a problem once the larger object (the superordinate phrase) has already been identified. Object recognition is guaranteed by Early Immediate Constituency signals (Hawkins Reference Hawkins1994, Reference Hawkins2004), a short size and a highly constrained form.