Nikolas Gisborne's book deals with various questions raised by perception verbs within a Word Grammar (WG) framework. It consists of eight chapters, which I will summarise briefly below, followed by a short glossary. However, my synopsis will only cover some of the many points made by the author since the book introduces and discusses a rich collection of data and is often very detailed in its analyses, which therefore cannot be reported in all their complexity here. Finally, I will add, where appropriate, a few critical comments to the summary of each chapter and then will deal with more general issues at the end of the review.

Chapter 1 (‘Introduction’) sets the scene for the rest of the book. Gisborne explains that he intends to offer a detailed description of perception verbs, using WG as a theoretical framework, as well as to address certain issues in lexical semantics such as argument linking, polysemy and evidentiality. It must be pointed out that although Gisborne subscribes to the usage-based thesis (see p. 4), his data are mainly based on introspection. In the Introduction, Gisborne distinguishes three kinds of perception verb, the Listen-class, which includes agentive verbs (as in I listened to the tenor), the Hear-class, which has to do with experiencer verbs (as in I heard him struggle), and the Sound-class, which involves percept verbs (as in The high C sounded flat, where the Subject is a percept). The Introduction, like the rest of the book, is rather dense in the sense that many observations are made and many questions are asked. Luckily, Gisborne offers a useful summary at the close of the chapter where he lists the main issues to be dealt with in his monograph, namely (1) ‘How can we capture both the generalizations and the idiosyncrasies that . . . exist in the domain of argument linking?’, (2) ‘How do we represent relationships between different verb senses?’, (3) ‘How do we analyse modal semantics?’ (4) ‘What are the theoretical issues raised by non-finite complementation?’, (5) ‘How does a network [i.e. a network representation] negotiate the underspecification of the temporal semantics of hear-class verbs?’.

Chapter 2 (‘Word Grammar’) is also introductory in nature. It consists in an introduction to WG, where Gisborne emphasizes that language is seen as intra-mental in WG, i.e. WG espouses a mentalist approach to language. Gisborne also stresses that dependencies in WG are not simply syntactic relations, but rather ‘gestalts . . .; they involve semantic information. In particular, a dependency is an argument-linking construction’ (p. 32). This amounts to saying that WG, like Goldberg's Construction Grammar but unlike Langacker's Cognitive Grammar (CG), distinguishes between semantics and syntax and then has to face the problem of explaining how the two are related to each other. The reader not familiar with WG will also learn, among many other things, that, contrary to e.g. Langacker's CG, both nouns and verbs are analysed as non-relational categories in WG, i.e. as nodes in a network. Relations do exist in WG but they comprise the dependencies and some semantic relations like Result (see below). In this chapter, Gisborne also discusses the notion of (default) inheritance and introduces the XComp dependency, which is found in predicative complementation, e.g. to be nice in Jane tried to be nice and We expected Jane to be nice. Both concepts will figure prominently in the rest of the book. Chapter 2 also introduces Er and Ee, which are the schematic semantic relations associated with the Subject and Object syntactic roles (they are also dealt with in more detail in the subsequent chapter), and the notion of sub-lexeme, i.e. ‘a subtype of a lexeme’ (p. 52). Although the chapter should be clear enough for readers not familiar with WG and equips them with a basic understanding of the theory so that they can grasp at least the gist of the WG analyses presented in the following chapters, the reader would nevertheless benefit greatly from reading ‘primary’ sources such as Hudson (Reference Hudson2007, Reference Hudson2010) before tackling Gisborne's book.

With chapter 3 (‘Causation and relations between events: an introduction to WG semantics’), Gisborne leaves the introductory part and delves into the issue of causation for, mainly, two reasons: because it is a central topic in lexical semantics across various different theories, and because it is needed for the discussion of some of the perception verbs in the subsequent chapters. He discusses three main approaches to causation, the localist approach, the aspectual approach and the causal/force-dynamic approach. He then goes on to propose that causation involves both a Result relation and an Initiator and an Endpoint. Gisborne claims that a Result relation is not enough to make a verb causative since a Result relation is found for example in inchoative open (e.g. The door opened), which is not causative in Gisborne's view. What is needed is not only a Result but also a force-dynamic interaction between what Gisborne labels an Initiator (cf. Talmy's Agonist) and an Endpoint (cf. Talmy's Antagonist). Gisborne also makes the important claim that, contrary to e.g. Jackendoff's (Reference Jackendoff1990) treatment, ditransitives involve no consistent force-dynamic transfer (see e.g. pp. 103 and 108); hence they are not to be treated as causatives, unlike in e.g. Goldberg (Reference Goldberg1995). Further, Gisborne points out that, as is well known, not all ditransitives involve a Result relation (i.e. an actual transfer as in Ann gave Beth the box) because some ‘lexicalize a Purpose relation’ (p. 106) as in She baked her children a cake (but the dog got to it), where the transfer is intended but doesn't necessarily take place. The chapter concludes with an analysis of the polysemy of make, a description of show and of how event structure is handled in WG (it is handled through a default inheritance hierarchy). As for show, it is important to observe that Gisborne analyses it as a type of giving, i.e. in WG parlance, ‘showing’ ‘Isa’ ‘giving’ (p. 112). Gisborne assumes that a sentence like Jane showed Peter the sculpture means something like Jane gave a look-see at the sculpture. Although Gisborne's analyses offer a lot of food for thought, I have some concerns about the material presented in this chapter. Gisborne's lexical semantic analyses are based on his own introspection – in fact, this point should be made for the other chapters as well – but there is no a priori reason to assume that all speakers have identical mental lexical representations. In fact, within the usage-based model approach, to which Gisborne claims to subscribe, analysts have underlined that speakers may have different ‘grammars’, see e.g. Dąbrowska (Reference Dąbrowska2010). Related to this point is the fact that the reader may find some of the author's analyses counterintuitive. Although Gisborne's efforts to tease apart notions like force dynamics, causation and result are commendable, to claim that causation involves both a Result relation and a force-dynamic interaction brings about the terminologically counterintuitive consequence that the very verb cause should not be described as a causative verb because it doesn't (necessarily) involve a force-dynamic interaction, as illustrated by his example Jane caused a cat to be killed under the reading that ‘Jane chased a cat out of her garden and it ran under a passing car’ (p. 88). Although this is essentially a matter of definition, I think that it would be simpler to align labels to existing lexical items, i.e. to use a different term for ‘causation’ in Gisborne's sense, so as to avoid such terminological confusion. More importantly, the author contends that ditransitives are not causative (in his sense). But the fact that some ditransitive examples are not amenable to a force-dynamic analysis doesn't necessarily mean that all examples should be analysed as non-force-dynamic. In an example like Sally gave Tom a punch, which is not used by Gisborne but is on a par with his Jane gave Peter a look (p. 108) in that both instantiate what is commonly regarded a light verb use of give, I struggle to see why there should be no force-dynamic interaction between Sally and Tom, for example.

In chapter 4 (‘Network structure and the polysemy of see’), Gisborne discusses examples such as the following:

1. (1) Jane saw Peter.

2. (2) Jane saw the dog cross the road.

3. (3) Jane saw into the room.

4. (4) Jane is seeing stars.

5. (5) Jane saw that Peter was right.

6. (6) Jane suddenly saw that Peter was right.

7. (7) I saw in the paper that there was another government scandal.

He claims that such examples illustrate five senses of see. (1) and (2) code the prototypical sense of see (‘perceive physically’), which Gisborne claims, following e.g. Jackendoff (Reference Jackendoff1990), includes two subparts, a ‘gazing’ part (which Gisborne calls ‘seeing’), and an ‘image-forming’ part. The former subpart accounts for the sense of see in (3), where the directional prepositional into can only be explained by appealing to some motion component in the meaning of the verb. The latter subpart is relevant to (4), where there is a mental image but no physical perception. (5) and (6) point to a meaning of see related to ‘understanding’. (7) is similar to the previous two examples but is also evidential. Gisborne's analyses are based on evidence involving ‘selection restrictions (the ontological class of the verb's complement [e.g. the distinction between event vs. proposition in (2) vs. (5)], Aktionsart [the fact that e.g. prototypical see is underspecified for its Aktionsart while image-forming see is dynamic], and semantic similarity to other verbs [the fact that see in (3) behaves like reach, see p.137]’ (p. 149). Further, he models his analyses within WG using the notion of sublexeme (as well as the Result relation, see p. 138) and stresses that WG can model polysemy by using just inheritance links. In other words, unlike Goldberg's Construction Grammar, there is no need for four different types of inheritance link. Curiously, however, Gisborne doesn't refer to Langacker's CG where similarly there is also one type of inheritance link, namely what Langacker calls a ‘categorizing relationship’ (see e.g. Langacker Reference Langacker2008: 37). It is difficult not to agree with Gisborne's analysis of his data in this chapter, although I would have liked some explanation of what he means by ‘prototypical’, as well as some discussion of why we should aim to (necessarily) simplify the lexicon (this point is made with reference to the analysis of see and understand as having the same concept as their sense, see p. 142). Within usage-based accounts there is no such concern because redundancy is part and parcel of the model (see e.g. Langacker Reference Langacker1987: 29). Also, inclusion of authentic examples would have been particularly useful in the discussion of the various senses of see. For example, Gisborne claims that the ‘seeing’ subpart of see resists both the progressive and the simple form, but one can come across examples such as (8), although, admittedly, see could also be paraphrased as ‘probing’.

1. (8) In this image, dark regions represent areas where Cassini is seeing into deeper levels in Saturn's atmosphere. (saturn.jpl.nasa.gov/photos/imagedetails/?imageID=824)

In chapter 5 (‘Perception verbs and the semantics of content’), Gisborne claims that the senses of look/a (i.e. agentive look, as in I looked at the painting) and prototypical see are hyponyms of the same concept, which he calls ‘eyeing’. He then goes on to show that ‘the PPs that alternate with the Direct Object of see are complements [e.g. Jane saw *(the picture) into the room, p. 166], the PPs that can occur with the Direct Object of see are adjuncts [e.g. Jane saw (the picture) through the keyhole, p. 166], and all the PPs that follow look are adjuncts [e.g. Jane looked into the picture at the details, Jane looked at the picture into the eyes of the Subject, p.166]’ (p. 157). In his discussion, Gisborne also points out that see has two themes, the gaze (which is unbounded and coextensive with its path) and the image of the percept, while look has only one theme, the gaze, which, in this case, is bounded and not coextensive with its path. This chapter also explores the force dynamics of see and look/a. Gisborne seems to conclude that in the case of see ‘there is no straightforward force-dynamic relation between’ its Subject and its Object (p. 178). Similarly, he claims that with look/a ‘there is no force-dynamic relationship between [the] Subject and Object’ (p. 180). The remaining part of the chapter is devoted to the analysis of the Aktionsart of see, which expands on the claim made in chapter 4, i.e. that ‘see/“seeing” is underspecified for its Aktionsart’ (p. 186). Although the motion analysis of see and look/a proposed in this chapter is very interesting, the idea being that the Subject's gaze is the theme for both verb types, this approach is not entirely new, as Gisborne himself points out; see e.g. his references to Goldsmith (1979) and Jackendoff (Reference Jackendoff1983, Reference Jackendoff1990), and see also Broccias's (Reference Broccias2003: 63–70) discussion of ‘emission verbs’. More importantly, what I consider to be a weakness in this chapter is how Gisborne introduces his motion approach. He does so by mentioning examples such as The A72 goes from Peebles to Glasgow. He claims they are metaphorical in nature since the road does not literally go anywhere. However, I would have liked to see some reference here to Langacker's fictive analysis of such cases (see e.g. Langacker Reference Langacker2008: 528–9, for a recent summary, and also Langacker Reference Langacker1990: ch.5).

Chapter 6 (‘Non-finite complementation’) discusses the difference between examples like We saw him go and He was seen to go. Gisborne claims that two different senses of see are involved. In the former case, the sense is strictly perceptual, i.e. ‘seeing’, and the percept (the referent of the XComp, go) is an action, ‘going’. He also makes the point that the two events of ‘seeing’ and ‘going’, prototypically at least, have to be co-temporal and the percept, like ‘seeing’ itself, has to be a stage-level predicate. However, if see is non-finite, the XComp can be an individual-level predicate (e.g. Seeing Jane own a car at last brings tears to my eyes). In He was seen to go, the sense of see is ‘understanding’ and the XComp denotes a proposition. Next, Gisborne tackles the big issue of why *Peter was seen go is impossible. He discusses ‘two alternatives, neither of which works’ (p. 220). The discussion is rather technical but can be summed up as follows. Under one analysis, *Peter was seen go is impossible because ‘the Ee of passive see [should be] mapped onto its Subject’ (p. 220), but the Ee is Peter go and only part of it, Peter (the Er of ‘going’, which is the Subject of the XComp) has been mapped onto the Subject. However, even if we map the Ee of passive see, i.e. Peter, onto the Subject (this is the second analysis), we get an impossible result because the Ee of the active verb (Peter go) should be mapped onto the XComp. It seems to me, however, that this is by and large a theory-internal explanation. Although Gisborne doesn't say so explicitly, he is aware that a deeper explanation is needed. Obvious questions are: where do such constraints on mapping come from? How can they be learnt? Gisborne just seems to try to model within WG the intuition that the problem with *Peter was seen go may be that an event like Peter go ‘has a basic integrity’ (p. 222) which cannot be broken up by the passive construction. In my view, the notion of ‘integrity’ needs to be explained in more detail so as to offer a truly cognitive account of the impossibility of cases like *Peter was seen go. Otherwise, one is merely paraphrasing descriptive facts using ad hoc theoretical machinery. Further, it would have been beneficial to discuss the impossibility of *Peter was seen go alongside the acceptable option Peter was seen going, which, surprisingly, is not mentioned in this chapter.

In chapter 7 (‘Sound-class verbs’), Gisborne claims that sound-class verbs have both epistemic and evidential meanings. The evidence given for the epistemic meaning of this class of verbs includes the observation that cases like Peter looks drunk are by default associated with the speaker as an experiencer if the experiencer is not specified by way of a PP (cf. Peter looks drunk to his boss). A major claim made by Gisborne is that such examples involve a force-dynamic relationship (which links into the context, i.e. with the speaker, when there is no explicit PP). However, it seems to me that the author simply asserts that a force-dynamic relationship is involved but never really explains its nature or why that should be the case. In this chapter, he also claims that sound-class verbs are polysemous. He distinguishes between three senses, a sensory evidential sense (also called evidential-1), as in The wine smells delicious, a reported evidential sense (also called evidential-2), as in (I've heard the forecast and) tomorrow's weather sounds fine, and an attributary sense, as in This food smells spicy. Both evidential senses can be paraphrased as ‘seem, with respect to a particular sensory modality’ (p. 245) and are therefore (subjectively) epistemic because they involve an ‘evaluation in the mind of an experiencer, in a force-dynamic transfer’ (p. 262), where the speaker is by default the Endpoint and ‘the evidential source is the force Initiator’ (p. 255). Gisborne contends that the attributary use can be paraphrased as ‘is, with respect to a particular sensory modality’ (p. 245) and, unlike evidential uses, cannot be followed by a phrase like ‘but it isn't’. Although this kind of evidence seems to work for most of the examples given by Gisborne, I don't see the alleged difference between the two options ‘The wine smells delicious but it isn't really’ (which should be possible) and ‘This food smells spicy but it isn't really’ (which shouldn't be possible). Next, he offers semantic entries for sound-class verbs in a WG formalism. The chapter concludes with a detailed description of like Xcomps of sound-class verbs (e.g. It sounds like Jane).

The last chapter (‘Conclusion’) focuses on some of the elements of WG which have been used in the analyses in the book and summarises the answers to some of the problems Gisborne posed in the Introductory chapter.

As I said at the outset, the book presents a rich array of data and problems and is certainly a must-read for all those interested in research into perception verbs. Still, at times the book reads more like a collection of separate case studies dealing with perception verbs than a truly coherent whole. What holds the chapters together is primarily the theoretical framework within which various problems concerning perception verbs are discussed. Further, the number of specific perception verbs taken into consideration is rather limited (and, perhaps, the title of the book should have specified that the perception verbs being investigated are English perception verbs). For example, the visual perception verbs discussed in the book are essentially limited to see and look; no mention is made for instance of watch, which behaves rather differently from see in terms of infinitival complement selection (see Egan Reference Egan2008: 146–9, for example). I also found the reliance on introspective judgments excessive at times and, since the author supports the usage-based thesis, I kept wondering why his analyses did not include authentic examples from text corpora. This is important because not all readers may agree with Gisborne's judgments and, as I pointed out above, different speakers may have different ‘grammars’ (e.g. why can't different speakers have different polysemy networks in their minds?). In this connection, since WG is a cognitive theory of grammar and Gisborne subscribes to the usage-based hypothesis, I would have liked to see more references to Langacker's work; instead, the major ‘cognitive’ reference point, apart from the theory of WG, is Jackendoff's work, which of course doesn't qualify as usage-based. Further, although theoretical modelling of the data is important, Gisborne could have put more emphasis on the cognitive principles behind his analyses – see my comments on the impossibility of *Peter was seen go above. Moreover, sometimes a possibly excessive amount of space is devoted to relatively straightforward and familiar notions, while more original ideas are not explored to the extent they deserve. This is the case for example in the discussion of the difference between Initiator and Endpoint, on the one hand, and Er and Ee, on the other, which boils down to the somewhat familiar point that not all subjects are agent-like (and not all objects are patient-like). By contrast, the alleged force-dynamic nature of cases like Peter looks drunk could have been discussed in more depth. Despite these criticisms, the book is an important reference point in the study of perception verbs because of the number of problems and challenges it raises in the pursuit of a cognitive treatment of such verbs.