In two relatively recent papers, Goldberg provides a valuable analysis of transitive constructions without an overt object, as in She contributed to the Leukemia Foundation. Goldberg correctly stresses that this construction, which she terms the Deprofiled Object Construction has a clear semantic value; she further discusses some of the factors that may bring about the deemphasizing of the object (repeated or habitual actions, politeness strategies, etc.) as well as some (lexical) factors that prevent object omission. It is especially the latter issue for which the present paper offers a more nuanced, and at the same time more general, alternative. In our approach, we retain the constructional value but place it in a larger paradigmatic perspective, the opposition between the transitive and the ergative paradigms that structure the English grammar of causative events. Our cognitive lexico-paradigmatic approach is corroborated by ample corpus data, which at the same time brings in important nuances, revealing the intricate and contextually variable interaction between constructional and lexical semantics and it does so in a way different from Goldberg’s analysis.

This paper discusses a particular construction in English that has already attracted much attention, i.e., the use of a lexical causative verb with omitted patient argument, as in Soldiers trained to kill Ø. [1] Most of the analyses in the literature focus their attention on the specific lexical properties of either the verb (cf. Rice 1988) or the object noun phrase (cf. Fillmore 1986; Cornish, to appear) as crucial factors allowing object omission. On the other end of the spectrum, two recent analyses by Goldberg (2001, to appear) present a constructional approach to the omission of the Patient argument, insisting on the semantics of the construction itself as an explanatory factor. Much to my appreciation, her analysis is remarkably compatible with my own study of these constructions (which I termed “objectless transitives”, see Lemmens 1998a). While Goldberg and my account are thus largely compatible with respect to the semantics of the construction itself, they diverge in the way they account for its distribution.

In fact, my analysis combines the two approaches by considering the meaning of the verb in interaction with the semantics of the construction whose distribution I explain following the systemic-functional analysis that argues that the grammar of causatives is governed by two different models of causality, the transitive and the ergative. The present paper expands my earlier work and shows that adopting such a lexico-paradigmatic view succeeds in refining Goldberg’s analysis as well as other, more lexically-oriented, analyses.

As will become apparent, this paper considers different kinds of one-argument constructions. The larger theoretical question that emerges from this discussion is whether it is in fact legitimate to postulate different kinds of formally identical constructions (i.e., taking only one argument) based on the assumed valence of a verb, if we only arrive at this valence through the abstraction over the constructions the verb occurs in. [2] This issue probes right into the interaction between constructional and verbal semantics, and their respective conceptual weight in the linguistic system. Related to this question is the one concerning the status of alternations (correspondences between constructions sharing the same verb) as opposed to the status of surface generalizations that Goldberg (2002) argues for saying that “each argument structure is best analyzed on its own terms, without relying on explicit or implicit reference to a possible alternative paraphrase” (2002: 329). Within the confines of this paper, a detailed discussion will not be feasible, but some critical thoughts will be offered in the concluding remarks.

The structure of the paper is as follows. In the first section of this paper, I will look more closely at the properties of the construction without over object, as they have been discussed in the literature. The second section concentrates on Goldberg’s constructional analysis of object omission. Section 3 briefly presents the paradigmatic opposition between (prototypically) ergative and (prototypically) transitive verbs, which allows further generalizations not present in Goldberg’s analysis. In the fourth section, I present a short corpus-based empirical evaluation of the suggested hypotheses.

In linguistic analyses, a lexical causative verb is typically regarded as encoding a change of state that affects, in one way or another, the patient argument. Examples cited are often of the type The cat killed the mouse, Floyd broke the glass or I dried the dishes where the mouse, the glass and the dishes change in the course of the event, i.e., they go from one state (alive, whole, wet) to another (dead, broken, dry). In these clear-cut cases, the general meaning of the causative verb can be paraphrased as “cause to become X”. However, as Goldberg (2001: 504, fn.1) points out, using the notion of change-of-state to define a causative verb might be tricky, as the change of state may not be the one expressed by the verb, which she illustrates with the example I dry-cleaned this shirt , but it is not really clean. In this case, the “cause to become clean” does not apply. [3]

Moreover, in the vast literature on verb alternations, the notion of change-of-state has often been “abused” to explain alternations that are mutually exclusive, e.g., the causative-inchoative alternation vs. the middle alternation. [4]

Even if the notion of change-of-state is somewhat slippery, let us accept for the time being that it nicely explains why the participant undergoing the change cannot be omitted from the construction, since it is crucially involved in the change encoded by the lexical causative. For instance, in reference to the same events as the examples mentioned above, object omission is ungrammatical: *The cat killed, *Floyd broke, *I dried. However, it has been observed in the literature that causative verbs can and do occur without an overt object, as in the following examples.

There is a consensus in the literature that even in case of omission, the object is nevertheless tacitly assumed. [5] Levin (1993: 33) nicely summarizes the issue:

Despite the lack of overt direct object in the intransitive variant, the verb in this variant is understood to have as object something that qualifies as a typical object of the verb; however, there is some discussion concerning how to best characterize this interpretation. There is also some debate about whether the understood object is or is not explicitly represented at some level of syntactic or lexical representation.

The unexpressed patient is interpreted as indefinite and non-specific, and as some say, predicted by verb. In (1b), for instance, the understood object is alcoholic beverages; in (2b), a human being. As Rice (1988: 203) puts it, “objects that can be omitted tend to be those whose lexical content is most probable given the meaning of the verb”.

As is apparent from this quote, Rice (as many others) looks at the issue essentially from the lexical end and tries to prove that the degree of specificity of both verb and object determines the possibility of object omission. She points out that very general verbs that sustain a wide variety of complements (e.g., make) tend to require overt objects. Also verbs that are quite specific with regard to their object or reveal something about the manner in which the action is carried out (e.g., John puckered his lips or Mary manicured her nails) almost always require overt objects. Thus, Rice says, object omission occurs mostly when the verb is moderately neutral and the range of objects is restricted to one or two possible semantic domains. Nevertheless, she does notice that transitive verbs that usually do not allow an objectless construction may occur with their objects omitted [6]:

To account for these examples, Rice observes that they “readily evoke general semantic frames or scenarios, the particular object is fairly unimportant as the pragmatic focus is on the activity itself” (1988: 206). Within a particular semantic frame, such as that of war and/or (organized) violence underlying (3d), it is possible to identify the omitted object. In earlier work I have taken this observation by Rice as the key to the semantics of an objectless transitive construction:

I argue that this ‘pragmatic focus on the activity’ is pivotal to the meaning of the objectless transitive. Hence, it comes as no surprise that, as a rule, they function as general statements, not tied to any specific event. […] For this reason, it can be claimed that, in a sense, the objectless transitive is the ‘Actor-focusing’ counterpart of the [Patient]-focusing middle construction, as both constructions are not tied to individual events but imply a general comment on the process. In essence, then, the objectless transitive focuses on the general characteristics of the act and cannot, in principle, refer exclusively to a specific event. Even when tied to a specific act, the objectless transitive positions the act in a more general perspective […] taking the participants’ actions to the more general plane of role (stereo)types within a given frame. (Lemmens 1998a: 142)

There are a number of typical contexts that give rise to such generalized encoding of events or to the evocation of (more general or more specific) frames and role stereotypes. One is that of (habitual) repetitions of the action, as in some of Rice’s examples above (notice the use of too much, always, and the like). Another context is the enumeration of actions as in Rice’s last example. Bordignon (2003) mentions a similar example in French that in its English translation works very well too:

French and English are not fully alike in their transitivity realizations, yet some parallels can be drawn, and this occurrence of Patient omission is strikingly similar. [7] Notice how the last verb in the series does not omit its object; and also in English it feels more appropriate to preserve it. The reason is that this use of squeeze is quite specific in its selection of arguments (only citrus fruits) whereas the other verbs do not have such high specification (basically their Patient can be any food that can be chopped, shredded, etc.).

In addition to exemplifying the ‘verb chaining’ that typifies the objectless construction, example (4) illustrates another feature that motivates the omission of the object – one not mentioned by either Rice or Goldberg, viz. the Instrument Agent, i.e. the promotion to Agent of what is typically (or factually) the Instrument, like appliances or tools. Here are two additional examples (English/French):

It should be noted that the distinction between Agent and Instrument is not always that simple. It is clearly beyond the scope of this paper, to enter into a full discussion here. I follow Nishimura (1993) who goes against the classical role preservation principle (once an Instrument, always an Instrument) and argues, quite convincingly, that subjects like those in (4) and (5) are construed as the Agents of the events, albeit non-prototypical ones. This choice is usually well-motivated, as Schlesinger (1989: 191) points out. One of the motivations for recategorizing Instruments as Agents is when the action’s progression crucially depends on the Instrument’s properties. Coding the gas or the knife as Agents in the examples above profiles their efficiency, or lack thereof, for the actions of killing and cutting respectively, even if our real world knowledge tells us that there is a primary (human) Agent initiating the action and manipulating the Instrument. Indeed, such sentences often contain explicit comments on the (un)feasibility of the process, cf. in minutes in (5a) or the negative particle not in (5b). The objectless construction further enforces the generic character of the utterance.

The contextual salience of the Agent is in sharp contrast to the non-salience of the omitted patient. This is why definite object NPs are less likely to be omitted (although omission may even be possible here), not because they are definite, but because their definiteness is the logical outcome of some degree of salience in the context at hand, a characteristic that is in strong conflict with the semantics of patient omission. A particular sub-case of the above mentioned enumeration of actions is that of contrasting actions by different Agents, i.e., patterns of the type [x VERB1 conj y VERB2] as illustrated by the following examples, taken again from Rice (1988: 206).

This pattern is often found in the characterization of roles (in the sense of ‘division of labour’) which can be seen as complementary, and thus positions the actions in a more general perspective.

To summarize, my earlier work sees Rice’s notion of pragmatic focus on the activity as pivotal to the semantics of the objectless construction, which can be characterized as providing a maximal focus on the Agent-Process unit, mostly implying a general statement which abstracts away from individual instances. As has been pointed out, there are a number of typical patterns that fulfil that function that are naturally compatible with (and conducive to) the construction’s meaning, such as repetition, action series and/or opposition, or various types of “modal” comments, e.g., the in minutes in (5a) or the negation in (5b).

Working in the framework of Construction Grammar Goldberg sets herself the goal of finding “a full understanding of the way rich lexical meaning interacts with discourse and constructional factors” (Goldberg 2001: 522). Fully in line with my earlier analysis (yet independent of it), Goldberg (2001: 514) formulates a principle of Patient Omission under Low Discourse Prominence:

Omission of the patient argument is possible when the patient argument is construed to be deemphasized in the discourse vis a vis the action. That is, omission is possible when the patient argument is not topical (or focal) in the discourse, and the action is particularly emphasized (via repetition, strong affective stance, discourse topicality, contrastive focus, etc.).

Goldberg takes issue with Rice’s analysis in terms of pragmatic focus, as she argues that

the increased emphasis on the actions is not captured by the notion of focus. […] It seems that a different dimension of discourse prominence is required; the action is somehow emphasized in the discourse, but not necessarily by virtue of being focal (2001: 513)

I agree with Goldberg’s observation that the action is not focal, but – at least as I understand it – Rice did not use the term “pragmatic focus” in this way, but rather as a notion similar to Goldberg’s idea of “emphasis”, i.e., “a cover term for several different ways in which an action is construed to be especially prominent in the discourse” (Goldberg, to appear). [8] It is precisely the compulsory co-occurrence of deemphasis of the Patient and emphasis of the action that is essential to the objectless transitive or the “Deprofiled Object Construction”, as Goldberg (to appear) calls it. (In this paper the term “objectless transitive” will continue to be used.)

As to the means of emphasizing an activity, Goldberg’s Principle of Omission mentions strategies similar to the ones already discussed: repetition, contrastive focus, discourse topicality, etc. However, she points out an additional motivation factor that may have contributed to the grammaticalization of this construction for certain verbs, viz. politeness strategies. One particular subclass for which this is true are verbs of donating, such as donate or contribute. [9] A construction not specifying how much one has contributed is often considered more polite than one that does, as shown by the following pair:

As Goldberg points out, “[t]he construction allows a means of making implicit an argument that would be indiscreet to mention” (Goldberg, to appear). It should be clear that also here the actual semantic import of the construction is essentially the same as in the other contexts, i.e., it renders the action component more salient and deprofiles the Patient. The politeness strategy “is proposed to explain why it is natural for such a construction to exist” (ibid). [10] Interestingly, Goldberg points out, while give generally does not allow an omitted object construction, it does allow it when the verb acquires the meaning “donate”:

Goldberg makes a case, correctly so, for the existence of the objectless transitive as a grammaticalized construction in the grammar of English, which in some cases may be motivated by politeness and semantic recoverability. She furthermore points out that “speakers need to learn that it is available and which classes of verbs can appear in it” (ibid.). Also, she observes, for some verbs, such as eat, drink, or sing, the frequent occurrence in an objectless construction in generic contexts with habitual interpretation (precisely one that sanctions this construction) “has apparently led to the grammaticalization of a lexical option for these verbs, whereby they could appear intransitively in less constrained contexts” (ibid.). In other words, the objectless uses are reinterpreted as an intransitive lexical option instead of being licensed by specific discourse constraints that motivate the objectless transitive. Taking this view, Goldberg arrives at an explanation why less frequent near-synonyms of the above verbs do not easily sanction object omission (indicated by #):

These verbs may occur in an objectless transitive, but clearly not as frequently as the other ones (since generally less frequent), and thus have not been subject to a reanalysis of the objectless construction as a lexical option. The idea of usage-based grammaticalization as argued for by Goldberg provides a unified account of why for some verbs the omission of the object is indeed more common than for others. In other cases, the semantics of the verb itself may render it more difficult to furnish a context that sanctions an omitted object. This is nicely illustrated by my earlier corpus-based analysis of a subset of causatives, verbs of killing (Lemmens 1998a, esp. 140ff). In that study, I analysed some 4,000 sentences with verbs of killing drawn from text corpora (fiction and non-fiction, AmE and BrE) and 11,000 attestations drawn from the Oxford English Dictionary. Verbs such as kill, murder, execute, butcher etc. can be regarded as prototypical causatives and it should thus be possible to devise contexts that sanction the omission of the object, as illustrated by the following examples:

However, while the construction is relatively common with kill (between 5-10% for the different corpora), it rarely occurs with the other verbs, the above examples nearly exhaust all the objectless constructions attested in the text corpora (there is only one more with murder). In the WSJ corpus (the only non-fiction corpus in our sample of text corpora), kill is the only verb for which an objectless construction has been attested. This clearly shows that lexical matters do count as well. Possibly, for verbs such as assassinate or massacre there is a stronger Patient-orientation, as they incorporate a more salient reference to Patients that are considered “important” in some socio-economical context (assassinate) or to a high number of Patients (massacre). In these cases, integrating the semantics of the verb with that of the objectless construction would require greater cognitive effort, which may explain the overall low frequency of the objectless construction in the corpus. The verb kill has a less specific semantic structure (not coincidentally, it is the common denominator of this lexical field) which renders it more readily available for this construction. Rice’s conclusion that “neither extremely schematic nor extremely specific verb-complement pairs encourage object omission” (1988: 207) can be interpreted as identifying the prototypical situation for this construction.

So, even though less common, the objectless construction is still possible with these verbs. Conversely, there are verbs, like open, break or starve for which the objectless construction is hardly ever possible even in contexts that would normally sanction the construction. The following examples, paraphrases of some of Rice’s examples mentioned earlier, are awkward, to say the least:

How can this be explained? Goldberg does it by drawing further on the lexical properties of the verb. She points out, for instance, that the verb recycle generally does not present any problems for the objectless construction (e.g., That man always recycles), since “[t]he patient argument, the garbage, is in many contexts less relevant in that we often don’t care what particular items are recycled” (Goldberg 2001: 512). However, the situation is quite different, she says, for break:

Bubbles, TVs, breadsticks, and hearts break in very different ways and with very different consequences. The patient argument supplies much of the relevant information. Thus it is hard to imagine a context in which there is a very strong discourse emphasis on the action of breaking and relatively little on what was broken. (ibid.)

Hence, it is generally impossible to omit the object here: *That man always breaks.

While it is clear that break generally does not permit the objectless construction, the question is whether Goldberg’s explanation is entirely correct. Two critical comments can be suggested.

First, if we look at the semantic coverage of a verb like kill, we also note a wide variety of caused processes: the interaction between participants is quite different for kill a human being or kill weeds, or a bill or a production process. Paraphrasing Goldberg’s argument, we could say that human beings, plants, bills and processes “die” in different ways and with different consequences. Nevertheless, this verb does allow omission of the Patient, and more so than more specific verbs of killing. At the same time, when the object of kill is omitted, it generally refers to an event in which a human being is killed, as in example (10d) Stop me before I kill again, or at most a living being; contexts where the omitted object concerns plants, bills or processes are much harder, if not impossible, to construe. Notice that this observation does not figure in Goldberg’s argument at all, where it seems to be of less concern.

Secondly, a verb such as drown is more specific in the particulars of the event, yet it does not allow omission of the patient either: John drowned the ants does not alternate with an objectless construction *John drowned. The latter construction can occur, yet will immediately be interpreted as the non-causative (often also somewhat infelicitously called “inchoative”) counterpart of Somebody drowned John. This already reveals that what is at issue here is not so much a lexical matter, as Goldberg suggests, but – once again – a systematic grammatical opposition. The next section will discuss this paradigmatic opposition in more detail, after which I will be in a position to examine corpus data revealing the intricate interaction between lexical and constructional semantics. By doing so, the objectless transitive is placed in a broader framework that augments Goldberg’s and Rice’s pertinent analyses.

The present study subscribes to Davidse’s (1999 [1991]) view that the English grammar of actions and events is governed by two distinct causative models, viz. the transitive and ergative paradigms. These two models represent different ways of conceptualizing causative processes, implying different conceptual centres and different participant relations (in Hallidayan terms, different “inherent voice” relations). The full details of this model should not concern us here; what is essential for our purposes is that it distinguishes two large families of alternations (paradigmatic perspective) and how they tie in with different verbs (lexical perspective).

Applied to the objectless construction, this boils down to the fact that in English (as in all the Germanic languages) there are two large classes of verbs: (i) transitive verbs like kill, pull, or make that may occur without an object and (ii) ergative verbs like break, open or burst that in principle cannot. [11] The transitive verbs encode an event where the Agent is the central participant, who targets his action against a Patient; in an ergative construal of an event, the second participant, which Davidse calls the “Medium”, is the central element, having the potential of self-instigating the event although a secondary Instigator can be added to the event (e.g., John broke the glass). [12]

There is thus no need to look for lexical specifics to explain why the verb break does not tolerate omission of the object: it is an ergative verb for which this construction is not sanctioned, because of a mismatch in the conceptual salience between the participant and argument roles. The alternation central to ergative break is the causative/non-causative alternation, just as with verbs such as open, burst, starve or drown, as illustrated in example (12)

Saying that the objectless construction is an exclusively transitive phenomenon does not mean that it will be equally likely with all transitive verbs. In fact, the construction is not the most prototypical exponent of the transitive model. The (transitive) objectless construction, but also the middle construction (e.g., These markers wash easily from children’s hands), are clearly less typical in terms of both frequency and naturalness and not all transitive verbs allow them. On top of that, there is a considerable degree of interaction with the verb’s semantics that may contribute to the typicality of a construction. This is exactly Goldberg’s point mentioned above about the objectless construction having become a lexical option for some verbs (e.g., drink or eat) but not for others (e.g., i mbibe or peruse), a point that was also illustrated by our analysis of verbs of killing. As already noted by Jespersen (1927), the objectless construction may entrench to the extent of the verb having become a fully-fledged intransitive, especially with originally reflexive verbs, e.g., pine (away) [from pine oneself] or recover [from either recover oneself or recover health].

The larger claim here is that while a verb may be typically (or largely) “transitive” or “ergative”, the paradigmatic character is in fact a property of the entire construction, where verbal and constructional semantics are intertwined. In other words, depending on its usage, a verb may oscillate between the two models. One of my parade examples is the verb abort which has a transitive-ergative split aligned with the distinction between literal uses that are transitive, illustrated in (13), and metaphorical uses which, quite surprisingly perhaps, continue the older ergative pattern, as illustrated in (14). (See Lemmens 1997, 1998a: 191-219 for details).

Such clear alignments as with the verb abort are unusual, but there is nothing inherently awkward about prototypically ergative verbs being used in transitive constructions, cf. (15) and (16):

It is difficult to conceptualize the tin as a participant co-participating or capable of self-instigating the event (a feature crucial to the ergative paradigm), which renders the non-causative unlikely. Nevertheless, it remains a coding possibility, even if a marginal one, whereas the non-causative construction is in principle never possible with transitive verbs. [13]

The interaction between verbal and constructional semantics has always been tricky to analyse and to formalize. Within our study of the objectless construction, it would be interesting to try to trace lexical influences more explicitly, by checking whether a prototypically ergative verb, which in principle does not sanction omission of its object, may nevertheless occur in an objectless construction. In an either-or approach to grammatical categories, such examples would immediately reject the paradigmatic oppositions outlined above. However, we subscribe to a cognitive view on categorization, which posits that each category (lexical or constructional) will have a prototype and that there may be less prototypical cases which violate the category’s general denominator. This is in fact precisely what the following section will reveal, discussing corpus material for three prototypically ergative verbs.

In order to validate the lexical-paradigmatic account, a complement to Goldberg’s analysis, we have carried out some corpus analyses of three typically ergative verbs: starve, suffocate and break. The present paper will only discuss the results of break in full detail, based on data drawn from the British National Corpus, a 100 million word collection of samples of written and spoken language from a wide range of sources, designed to represent a wide cross-section of current British English, both spoken and written. [14] The results for starve and suffocate will only be summarized very briefly here, as they have been presented elsewhere (see Lemmens 2005).

In this paper, I have fleshed out my earlier lexical-paradigmatic analysis of the objectless construction, by bringing in further evidence that the construction is in principle limited to the transitive paradigm and does not occur with ergative verbs, such as open or break. Its non-occurrence with these verbs is thus not a purely lexical matter but a paradigmatical one, since – due to a clash between the semantics of the verb and the construction – the construction is in principle excluded with ergative verbs. Combined with insights from Rice’s and Goldberg’s analyses, the paradigmatic account thus has a much wider applicability. At the same time, the analysis has provided clear examples of how specific usages interact with the paradigmatic potential in non-trivial ways. In some cases, these concern variations tied to more general patterns, such as for example the intransitive motion construction for break, typically combined with a particle further specifying the path of the motion. In other cases, particular uses of the verb may interfere with the verb’s prototypical paradigmatic behaviour (transitive or ergative).

As the exploratory contrastive study of break has shown, this also pertains to the type of entities that occur in different constructions with this verb, as the entities that are being broken are generally different from those that break. This (quite modest) type of analysis examining the associations between the construction(s) at issue and (some of) the words occurring in them is one that should be pursued on a larger scale for more attestations for a larger sample of verbs and with more statistical support. It is expected that such a study would further confirm what has been observed for break in the present paper, namely (i) that the type of affected entities are largely different in the causative versus the non-causative construction, (ii) that larger semantic schemas may be at work that trigger their own constructions (like the intransitive motion schema) and (iii) that the specifics of a given usage may overrule the usual constructional constraints of a verb.

A final question to be considered is how all this relates to the theoretical issue mentioned at the beginning of the paper, concerning the conceptual weight of surface generalizations as opposed to alternations. The question concerned the validity of postulating different kinds of construction that are formally identical based on the assumed valence of a verb, if this valence is the result of a generalization process (by the linguist or by the language user) over the different constructions that the verb occurs in. In other words, would it not be preferable to altogether abandon the notion of alternations (often used to prove underlying differences, as also in this paper), and posit instead surface generalizations, as do for example Goldberg (2002) and Yoshimura and Taylor (2004). The modest scope of the present study does not allow any definite conclusions, but some thoughts can be offered for further consideration and reflection.

First of all, in the usage-based approach, to which this study subscribes, such “perceived” valence, deriving from the integration of a construction and a verb, does not pose any problem at all. On the contrary, it is the repeated experience of such context-specific instances that gives rise to a certain valence pattern for a verb becoming the predominant one, or the one that is prototypically associated with the verb. In other words, the valence of a verb is the entrenchment of similar individual occurrences in a given construction. Obviously, the lexico-semantic structure of the verb itself will provide further semantic specifications for the schematic roles projected by the construction.

By way of example, let us consider the one-participant constructions as in These girls cried, These glasses broke and These books read well . Pushing for surface generalizations, Yoshimura and Taylor (2004) put these SU-V constructions all under one single heading. [19] While such unifying operation might still be workable for the first two, where the subject participant possesses characteristics of Agent and Patient, the third construction clearly stands out as a middle construction where the participant does not have any Agent properties. Yoshimura and Taylor – who regard their account as being “in conflict with core aspects of the traditional approach” (2004: 316) – group all SU-V constructions together as “middle constructions” whose essence they define as “residing in the special way in which a subject referent is conceptualized in association with the semantics of the verb phrase” (ibid).

Their arguments are open to some criticism, which will not be presented here, however. Suffice it to point out that by grouping all these constructions together into one super-category, they ignore semantic distinctions for which ample evidence exists in the literature. In the words of Croft (2001), they can be regarded as true “lumpers”. I agree with Croft when he says that “[t]he fundamental fact that is overlooked is that while difference of form entails difference in categorization, identity of form does not entail identity of categorization” (2001: 76). While I refrain from becoming a true “splitter” (the extreme opposite of the lumpers), I do argue for more refined subgroups, since the data clearly suggest that verbs and their semantic roles pattern in subcategories and that speakers of English are well aware of these and often exploit them creatively. This seems to indicate that the language user operates at a lower level of abstraction, which result from generalizations from different verbs (or verb groups) occurring in a given construction, rather than at the most schematic one possible.

As has become apparent from the analyses presented in this paper, I argue for two types of refinements. First, on the constructional level, I follow Davidse’s (1999) distinction between the transitive and ergative models of causation, which each project different semantic relationships between the participant and the process expressed by the verb, and hence represent different constructions. While I equally reject a derivational view on alternating constructions – each construction clearly exists in its own right and is not “derived” from another – I argue that alternations do reveal the more hidden regularities of grammar. It would not be fair to Goldberg to say that she would ignore those, as she, too, points out that broader generalizations can be obtained by looking at “all instances with a formal and semantic similarity” (2002: 335; my emphasis).Much hinges on how far one is willing to push these generalizations and at what semantic cost; as I see it, Yoshimura and Taylor’s analysis pays it far too dearly.

At the same time, I extend Davidse’s approach considerably by considering the semantics of the verb as well as the contextual specifications (such as type of participants). As such, regular patterns of use (units) can be distinguished at yet another level. This is precisely where cognitive-functional linguistics would benefit from the corpus linguistic tradition, where such more fine-grained analyses of large data sets have been common practice. The analyses reported on in this paper are but the beginning of such a synergism.

Alvarez, Gerardo H. (1968). Transitivité et économie de l’objet dans l’énoncé français. Thèse de doctorat. Paris: Nanterre.

Aston, Guy & Lou Burnard (1998). The BNC handbook. Exploring the British National Corpus with SARA. Edinburgh: Edinburgh University Press.

BNC: British National Corpus.

Bordignon, Frédérique (2003). Une approche cognitive du potentiel sémantique et constuctionnel du verbe casser. Thèse de doctorat. Université Clermont II, France.

Cappelle, Bert (2005). Particle Patterns in English. A Comprehensive Coverage. PhD thesis. K.U. Leuven, Belgium.

CASS: Contemporary American Short Stories.

Cornish, Francis (to appear). Null complements, event structure, predication and anaphora: a Functional Discourse Grammar account.

Croft, William (2001). Radical Construction Grammar. Syntactic Theory in Typological Perspective . Oxford: Oxford University Press.

Davidse, Kristin (1999). Categories of Experiential Grammar. (Monographs in Systemic Linguistics X). Nottingham: University of Nottingham. (reprint of Davidse (1991). Categories of Experiential Grammar. PhD Thesis. K.U. Leuven, Belgium).

Fellbaum, Christiane & Judy Kegl (1989). Taxonomic Structures and Cross-Category Linking in the Lexicon. Proceedings of the Eastern States Conference on Linguistics: 93-104.

Fillmore, Charles (1967). The Grammar of Hitting and Breaking. In Jacobs, Roderick & Peter Rosenbaum (eds.) Readings in Transformational Grammar. Waltham, MA: Ginn. 120-133.

Fillmore, Charles (1986). Pragmatically Controlled Zero Anaphora. Berkeley Linguistics Society 12: 95-107.

Goldberg, Adele E. (1995). Constructions. A Construction Grammar Approach to Argument Structure. Chicago: Chicago University Press.

Goldberg, Adele E. (2001). Patient Arguments of causative verbs can be omitted: the role of information structure in argument distribution. Language Sciences 34: 503-524.

Goldberg, Adele E. (2002). Surface generalizations: An alternative to alternations. Cognitive Linguistics 13: 327–356.

Goldberg, Adele E. (to appear). Constructions, Lexical Semantics and the Correspondence Principle: Accounting for Generalizations and Subregularities in the Realization of Arguments. In Erteschik-Shir, Nomi & Tova Rapoport (eds.) The Syntax of Aspect. Oxford: Oxford University Press. Available at http://www.princeton.edu/~adele/Israel2.doc accessed 24 February 2005.

Huddleston, Richard & Geoffrey K. Pullum (2002). The Cambridge Grammar of the English Language . Cambridge: Cambridge University Press.

Jespersen, Otto (1927). A Modern English Grammar on Historical Principles. Heidelberg: Carl Winter.

Keyser, Samuel J. & Thomas Roeper (1984). On the Middle and Ergative Constructions in English. Linguistic Inquiry 15: 381-416.

Larjavaara, Meri (2000). Présence ou absence de l’objet. Limite du possible en français contemporain. Helsinki : Academia Scientiarum Fennica.

LDC: Leuven Drama Corpus.

Lemmens, Maarten (1997). The Influence of World Conception on Transitivity and Ergativity: a Case Study. In Sweetser, Eve, Kee Dong Lee & Marjolijn Verspoor (eds.) Lexical and Syntactic Constructions and the Construction of Meaning. Amsterdam/Philadelphia: John Benjamins. 363-382.

Lemmens, Maarten (1998a). Lexical Perspectives on Transitivity and Ergativity. Causative Constructions in English . Amsterdam & Philadelphia: John Benjamins.

Lemmens, Maarten (1998b). The experiential basis of lexical and constructional flexibility: a diachronic and synchronic study. Leuvense Bijdragen 87: 79-113.

Lemmens, Maarten (2005). Des constructions causatives sans objet : un complément à l’analyse récente de Goldberg. In Delmas, Claude & Mireille Quivy (eds) 6 Etudes en linguistique anglaise, CERCLES Revue pluridisciplinaire du monde anglophone. Occasional Papers, Université de Rouen. 117-138

Levin, Beth (1993). English Verb Classes and Alternations. A Preliminary Investigation. Chicago: Chicago University Press.

Levin, Beth (1995). Unaccusativity. At the Syntax-Lexical Semantics Interface. (Linguistic Inquiry Monographs 26). Cambridge, MA: MIT Press.

Mittwoch, Anita (1971). Idioms and Unspecified NP Deletion. Linguistic Inquiry 2: 255-259.

Mittwoch, Anita (1982). On the Difference between Eating and Eating Something: Activities versus Accomplishments. Linguistic Inquiry 13: 113-122.

The Oxford English Dictionary on CD-ROM. 1993. Oxford: Clarendon Press.

Nishimura, Yoshiki (1993). Agentivity in Cognitive Grammar. In Geiger, Richard A. & Brygida Rudzka-Ostyn (eds.) Conceptualizations and Mental Processing in Language. Berlin/New York: Mouton de Gruyter. 488-530.

Rice, Sally (1988). Unlikely Lexical Entries. Berkeley Linguistics Society 14: 202-212.

Schlesinger, Izchak M. (1989). Instruments as Agents: On the nature of semantic relations. Journal of Linguistics 25: 189-210.

WSJ: Wall Street Journal.

Yoshimura, Kimihiro & John R. Taylor (2004). What makes a good middle? The role of qualia in the interpretation and acceptability of middle expressions in English. English Language and Linguistics 8: 293-321.