Andrew Radford is well known for his textbook introductions to what Culicover & Jackendoff (Reference Culicover and Jackendoff2005: 3 fn.1) call Mainstream Generative Grammar (MGG), ‘the line of research most closely associated with Noam Chomsky’.Footnote ^[1] For nearly 40 years, his books have made some complex ideas accessible to students. But unlike many working in MGG, who favour a little data and a lot of theorizing, Radford loves getting his hands dirty with data. Thus, for example, he is well known for his work on children’s early syntax (Radford Reference Radford1990). In the present book, he focuses on a very specific kind of data: live unscripted radio and TV broadcasts, especially commentary on soccer and cricket, supplemented by Internet data. The book consists of four chapters: an introductory chapter setting out the theoretical background, a chapter on relative clauses containing some kind of resumptive element, a chapter looking at preposition doubling and related phenomena in relative clauses, and a chapter on relative clauses which seem to contain neither a gap nor a resumptive element. It concludes with a brief epilogue. The book is based on a fascinating body of data, which should be of interest to anyone interested in naturally occurring English. The discussion of the data will be of considerable interest to proponents of MGG. Outsiders, however, may well feel that it has some important weaknesses and limitations.

The main focus of Chapter 1 is the approach to relative clauses that is assumed in the following chapters. Radford adopts the cartographic version of MGG, in which the left periphery of the sentence involves a complex hierarchy of phrases – RelP, ForceP, ModP, WhP, FinP and others. Among other things, this makes it to quite easy to handle examples like the following, where a relative clause contains an interrogative and an imperative:

However, it means that simple relative clauses have a mass of invisible structure, a position which would not find favour outside MGG. Following standard MGG practice, Radford refers here to interrogative and imperative ‘force’. This is unfortunate terminology. Interrogative is a sentence type which can have a number of illocutionary forces: question, request, suggestion, etc., and much the same is true of imperative (see e.g. Huddleston & Pullum Huddleston et al. Reference Huddleston and Pullum2002: Chapter 10).

Radford also assumes, following Kayne (Reference Kayne1994), that the antecedent of a relative clause may originate inside the clause as a complement of the relative pronoun. In support of this idea, he cites examples like the following:

Here, himself is not c-commanded by its antecedent on the surface, but it will be in its original position if it has originated inside the clause. The problem with this argument is that there are examples where a reflexive is not c-commanded by its antecedent for which no movement analysis is plausible. Pollard & Sag (Reference Pollard and Sag1992: 272) cite examples like the following:

There may be good arguments for the idea that the antecedent of relative clause can originate inside the clause, but this is not one.

In Chapter 2, Radford looks at relative clauses containing not a gap but a resumptive element of some kind. This may be a pronoun, as in (6), a demonstrative, or a nominal, as in (7).

As Radford notes, there has been a long debate about whether resumptive elements are grammatical in English or performance errors of some kind. His view is that they are grammatical but ‘stigmatised in much the same way as other colloquialisms such as gonna/gotta/wanna’ (55–56). This seems questionable. Prescriptivists clearly can and do object to contracted forms like gonna, but resumptive elements are not one of the things they standardly object to. It is not obvious how they would object. Clearly not by using the term ‘resumptive’, which means nothing to most people.

Assuming that examples like these are grammatical, Radford considers what sort of analysis would be appropriate. He argues convincingly against a movement analysis, noting among other things that they would require some ‘spellout magic’. However, this suggests that such examples are rather different to superficially similar examples in languages in which resumption is a normal feature. There is considerable evidence that gaps and resumptives involve the same mechanism in such languages, which in MGG means movement; see e.g. Aoun, Choueiri & Hornstein (Reference Aoun, Choueiri and Hornstein2001) and Alotaibi & Borsley (Reference Alotaibi, Borsley and Müller2013) on Arabic, and Willis (Reference Willis and Rouveret2011) and Borsley (Reference Borsley2013) on Welsh. It looks, then, as if what we have in Radford’s data may not be ordinary resumption of the kind that occurs in these languages. The fact that the data involves resumptive nominals as well as resumptive pronouns also points to this conclusion. As far as I am aware, languages which have resumption as a normal feature have resumptive pronouns but not resumptive nominals. It seems to me that this casts some doubt on the assumption that the examples that are the focus of this chapter are grammatical.

In Chapter 3, Radford turns to preposition doubling, as in (8), examples with two different prepositions, such as (9), and examples where a preposition is missing, such as (10), where, as indicated, by seems to be missing.

He concludes that examples like (9) and (10) are performance errors but suggests that this is not the case with examples like (8). He notes that they might be viewed as blends, errors which arise where a speaker has two different target structures in mind and produces a hybrid structure with properties of both. However, he argues against this view, citing the research of Radford, Felser & Boxell (Reference Radford, Felser and Boxell2012), whose informants found such examples acceptable. He proposes that they are grammatical and that they involve what he calls ‘redundant spellout’. Normally all the copies left behind by a moved constituent are deleted. But here the lowest copy of a moved PP is not deleted in its entirety. Rather, just the complement of the preposition is deleted, and the preposition remains. If the findings of Radford et al. (Reference Radford, Felser and Boxell2012) are seen as decisive, then some such analysis seems necessary. However, Rui Chaves has pointed out to me that it is well-known that high frequency and highly predictable words are more likely to be skipped during reading (Rayner & Duffy Reference Rayner and Duffy1986, Rayner & Pollatsek Reference Rayner, Pollatsek and Coltheart1987, Henderson & Ferreira Reference Henderson and Ferreira1990). Hence, it is possible that subjects systematically failed to notice the repeated prepositions. This suggests that the findings of Radford, Felser and Boxell do not necessarily establish that these examples are grammatical. If they are seen as grammatical, then something like ‘redundant spellout’ is necessary. But this is a rather obscure notion. It is not really clear how it is supposed to work or why it doesn’t happen elsewhere. Are there cases where a VP is moved and the verb within the copy that is left behind is not deleted? If not, why not? In contrast, if these examples are seen as errors, then the notion of a blend provides a plausible explanation. It seems to me, then, that these examples could well be performance errors.

In Chapter 4, Radford considers what he calls gapless relatives, relatives with no obvious gap and also, unlike those discussed in Chapter 2, no resumptive element. Here is a typical example:

He assumes that examples in which a normally obligatory preposition is missing represent processing errors. An example is as follows, with about missing:

However, he assumes without much argument that other gapless examples are grammatical and considers what sort of analysis might be appropriate. He considers and rejects a number of analyses in which they in fact contain a gap which is the object of an invisible preposition. Instead, he argues for an analysis in which the relation between the relative clause and its antecedent is determined by pragmatic inferencing (216). It is worth noting that whether or not such examples are grammatical, they must be interpreted when they occur, and it seems reasonable to propose that pragmatic inferencing is crucial here.

I have suggested in the preceding paragraphs that many of the examples that Radford sees as grammatical may in fact be performance errors. It would be interesting to see more investigation of this matter. It would be good, for example, if some of the commentators could be asked about some of their utterances. Would some of them agree that some of their utterances are not what they would have said if they had been under less pressure? It would also be interesting, if possible, to compare the utterances the commentators produce in commentary with the utterances that they produce in less pressured situations. For example, it might be interesting to compare the language that cricket commentator Jonathan Agnew produces when commentating with the language he produces when interviewing guests, something that he does regularly. There might well be interesting differences in the area of relative clauses.

Obviously, if the various types of example are grammatical, analyses are necessary. But analyses are also valuable if the crucial examples are performance errors of some kind since it is important to consider exactly how such examples differ from broadly similar grammatical examples. As noted above, Radford considers what sort of analyses are appropriate. However, he does not provide analyses in the sense of sets of formal rules or constraints which identify precisely what is and is not possible. In other words, he does not do the sort of thing that Sag (Reference Sag1997) did for relative clauses in Standard English. It would be interesting to see analyses in this sense for the phenomena.

How well equipped is MGG to provide analyses for the data? For the framework assumed in Sag (Reference Sag1997), a grammar involves a complex system of phrase types or constructions. Within this framework, the relative clauses discussed in Chapters 2 and 4 might involve one or more phrase types similar to those proposed by Sag but differing at least in not involving a gap. MGG of course has no time for constructions and claims that the grammar is just a few general mechanisms such Merge, Agree, and Move. What actually happens in syntax is determined by the lexical elements with which these mechanisms interact. In the case of relative clauses this is mainly invisible functional heads. Thus, it is these functional heads that would be central to a real analysis. It would presumably be possible to provide such an analysis with a precise specification of the properties of the functional heads, but this is something that never seems to be done in MGG work, and this book is no exception.

Thus, I have a number of reservations about this book. Nevertheless, it is an important work. Even if one is sceptical about the value of MGG, the detailed exploration of the analytic options that MGG makes available is of considerable interest, and even if one doubts whether various types of example are grammatical, the data that is presented here is fascinating. This may well be what is of most lasting value in the book, and it may well be that psycholinguists will derive most benefit from it. The nature of speech production is a central concern for psycholinguists, and one question here is: what happens when speakers have to communicate complex ideas in pressured situations? This book shows in great detail some of the things that happen.