Printer Friendly

On the middle voice: an interpersonal analysis of the English middle *.


In this article we approach the phenomenon of the English middle voice from a functional-cognitive perspective. Assuming the distinction between representational and interpersonal layers of organization in the clause (Halliday 1985; Hengeveld 1989; McGregor 1997), we propose that the elements and relations that define the English middle are of an interpersonal nature. We argue that the characteristic relation between nonagentive subject and active VP in the middle cannot be generalized in terms of a specific process-participant relation, as has been attempted in the literature. It cannot be maintained that the subject entity is always affected by the action, nor that it brings about the action, which would obscure the difference between middle and ergative intransitive. Rather, we propose that the middle subject and finite are related to each other by what Talmy (2000) defines as a "letting" modal relation, which we characterize further as combining active and passive elements. Middles construe a subjective assessment of the subject entity, presenting it as lending itself to the action designated by the predicator, and as having properties that are actively conducive to that action. We also argue that this letting relation between subject and finite can be interpreted as the subjectification of the agentive-patientive relation between the lexical verb and the sole participant in an ergative intransitive clause. The shift towards subject and finite expressing a conduciveness judgment activates an inherent agent in the conceptual base of the predicate. This means that ergative verbs are always used transitively in middles and it explains why other transitive verbs can also occur in middles. In a language like French, which marks both ergative intransitive and middle reflexively, no further extension of middle predicates is possible. By contrast, both Dutch and, more marginally, English, which do not mark middles reflexively, also allow some intransitive and compound transitive predicates to be used in the middle voice.

0. Introduction

In this article we will approach a number of descriptive problems posed by the English middle construction from a functional-cognitive theoretical perspective. The theoretical claim most central to our analysis of the middle is the layered approach to clause structure found in the functional frameworks of Systemic Functional Grammar (Halliday 1970, 1985; Davies 1979, 2001; see also McGregor 1997), and Dik's (1991, 1997) Functional Grammar particularly as developed by Hengeveld (1989). More specifically, the middle raises a number of descriptive issues that are fundamentally related to the distinction between the representational and the interpersonal organization of the clause (1). At the representational level, the relevant layer is that called predicate formation by Hengeveld (1989), i.e., the selection of the lexical verb and the construal of semantic roles for the nominals related to the verb. At the interpersonal level, the constructional organization relevant to the middle centers on the modal value of the finite and its relation to the subject, which in turn determines the integration with the whole of the predication. In this article, we will argue that, whereas hitherto one has generally attempted to characterize the middle at the level of predicate formation, it is in fact a specific interpersonal construction.

Typical examples (2) of the English middle construction are:

(1) More than 50 years later, the book still sells well. (CB--times)

(2) The new Holden Berlina handles like a junior sports sedan (...). (CB--oznews)

(3) And it's latex paint, so it cleans up quickly and easily with soap and water.


The English middle combines an active verb form with a subject that typically has a patientive flavor. For this reason, its voice has been labeled "activo-passive" (Jespersen 1914-1929, vol. 3: 347) or "middle", with the latter term having become the most generally used label for the construction as such. (3) Semantically, the middle is generally assumed to comment on the facility (Fawcett 1980) of carrying out a particular process with respect to the entity construed as subject. It is said to express meanings such as whether the action can be carried out, and if so, whether easily or with difficulty as in:

(4) [alarm system:] Installs quickly and easily--no wiring, no mains connection!


(5) When no longer required, the discs remove easily.


(6) The Waves, whose poetic prose now reads very awkwardly....

(Google) (4)

However, this traditional characterization of the middle runs into problems when confronted with English middle examples whose subject is not obviously patientive, as in (7) and (8), or that specify other aspects than the facility of carrying out the process such as the typical spatial destination of an entity, like (9).

(7) You wouldn't give my tip a chance of landing Aintree's Martell Red Rum Chase [...] but with the ground riding slower, he should improve dramatically.


(8) The top quality rods from Montegue, H&I, South Bend etc are good rods, fish well and look good.

(Google) (5)

(9) [about shoe chest] Stows on floor or shelf.

(Fagan 1992: 54)

As we will see in Section 1, in attempting to account for the constructional link between a nonagentive subject and an active VP, existing analyses have proposed generalizations pertaining to the REPRESENTATIONAL semantics of the process depicted in the middle. In the literature about English middles, the dominant generalization is that the subject is obligatorily affected. However, as we will show, this claim is falsified by the occurrence of English middles with intransitive verbs. Alternatively, some authors have proposed that middles confer a kind of agentive status on their subject, thus portraying the event as taking place autonomously. We will argue that this analysis fails to indicate clearly in what way the middle differs from the ergative intransitive.

In Section 2, we will make our fundamental point that the characteristic properties of middles reside in their INTERPERSONAL organization, i.e., in their subject-finite-predicate structure. In Section 2.2, we analyze the middle finite as expressing a modal letting value, which we link, in Section 2.3, to the conducive nature of the subject. Having established the letting finite and conducive subject as core properties of the middle construction, we propose in Section 2.4 a typology differentiating between subtypes of middles on the basis of complement and adjunct types.

In Section 3, we will bring together the elements of the preceding discussion by interpreting the middle as a SUBJECTIFICATION of the ergative intransitive. The ergative intransitive is a representational configuration, depicting an event whose single participant has an activopassive relation to the process. We will propose that the middle reinterprets this activopassive process-participant relation as the activopassive letting modality between subject and predication.

1. The representational approach to English middles in the literature

Surveying the literature, one finds that the general approach to middles, across linguistic schools, has been a representational one. The main attempts to typify middles have either formulated constraints on predicate selection (Section 1.2) or have claimed that the subject in middles participates in a specific way in the process designated by the verb (Section 1.3). Before discussing these, however, we will specify the theoretical assumptions within which we interpret these descriptive claims. As the exact nature of the representational relations that can be expressed in English middles is crucial to our argument, we will explain in Section 1.1 how we conceive in general of the representational layer that expresses process-participant relations and circumstances.

1.1. The layer of process-participant and circumstance relations

As also argued in Laffut and Davidse (2002), we view the process of predicate formation as the assembly of a conceptually motivated dependency structure, comprising both complementation and modification relations as defined by Langacker (1987: 306-310). In this section, we will first discuss Langacker's complementation analysis of process-participant relations (1.1.1), and then his modification analysis of circumstantial relations (1.1.2) (6).

1.1.1. Participants in processes as complements of verbs. Reacting against traditional valency approaches according to which the verb "determines" all its valents, Langacker (1987: 308-310) has proposed that the meaning of a lexical verb is "elaborated" by nominal complements. The verb, Langacker (1991: 14) emphasizes, is conceptually dependent. Verbs designate processes, i.e., events, energetic interactions, relations, etc., but these processes cannot be conceptualized separately from the participants in the process, i.e., the interacting or interrelating entities. Take an example like

(10) Mary stowed the shoe chest on the shelf. (cf. Fagan 1992: 54)

The meaning of the verb stow includes schematic reference to an agent doing the "stowing", an entity being stowed somewhere and the spatial destination of that entity. In Langacker's (1987: 304) terminology, the verb contains schematic "elaboration sites" (e-sites), which are elaborated by specific complements, such as Mary, the shoe chest and on the shelf in Example (10).

Importantly, Langacker's complementation model enables one to model the crucial fact that the nominals in the configuration undergo a change of semantic profile as a result of their complement-relation to the verb. Nominals such as Mary and the shoe chest as such designate entities, but when they function as complements of a verb as in (10), they become part of the "action chain" (Langacker 1991: 291-294) depicted by the verb, and come to designate an entity "stowing" or "being stowed", i.e., an "agentive" or "patientive" participant in the process. (7) Following Halliday (1968: 198), such agentive or patientive relations between participants and process can be thought of as "inherent voice relations". Halliday introduced this notion to discuss the differences and similarities between the examples:

(11) The prisoners marched.

(12) John marched the soldiers.

(13) John made the soldiers march. (Halliday 1968: 198)

According to Halliday (1968: 198), the use of the analytical causative made in (13) keeps intact the inherent voice relation of the prisoners to march as agent of an intransitive process. In (11) the inherent voice relation of the prisoners to intransitive march is also agentive. By contrast, the lexically causative use of march in (12) does code a different inherent voice. In (12) the prisoners are both patient of the general's forcing them to march and agent of the process of marching itself. The relation between marched and the prisoners is thus agentive-patientive (see also Davidse and Geyskens 1998).

The "inherent voice" relations defined by a specific complementation pattern of a lexical verb remain the same irrespective of whether the clause is active or passive. For instance, in both Mary stowed the shoe chest on the shelf and The shoe chest was stowed on the shelf by Mary, Mary is agent and the shoe chest patient of the stowing. This shows that inherently agentive or patientive participation in the process has to be distinguished from the active and passive options offered by the system of clausal voice (see Section 3). As we will argue below, a central element in our characterization of the middle is precisely that middle voice is not a matter of a specific type of participant voice but constitutes a special type of clausal voice.

1.1.2. Circumstances as modifiers of process-participant relations. Participants conceptually complete the process expressed by the verb (V) in that they "instantiate" (Langacker 1991: 33) it: they realize the processual chain that starts with entities V-ing and that may also extend to entities being V-ed. Circumstantial relations do not conceptually complete the process. They do not participate directly in the action chain, but "modify" elements of the action chain (Langacker 1987: 308; McGregor 1992, 1997). Circumstantial meanings often pertain to a structure consisting of the process and at least one participant. So-called "agent-oriented" and "patient-oriented" circumstantials modify an aspect of the agent's or the patient's relation to the process. Circumstances of place and time typically specify the setting of the whole process-participant nucleus. Some circumstances, however, seem to modify the verb only, such as circumstances of manner (McGregor 1992: 141).

In this section we will concentrate on circumstances of location and instrument, because they are particularly relevant to some of the more intractable middle examples cited in the introduction, like (7) and (8).

Let us begin with circumstances of location, as illustrated by (14) and (15). (Example [15] is a nonmiddle example related to middle Example (7) above, the analysis of which is controversial and will be returned to in Section 1.2 below.)

(14) In the dressing room Mary stowed the shoe chests on the shelves.

(15) On the soft ground the jockeys were riding slower.

As discussed in 1.1.1, (14) construes an action chain with Mary doing the stowing, the shoe chests being stowed, and the shelves being stowed upon. The dressing room, by contrast, is not directly targeted by the process of stowing. Rather, in the dressing room specifies the spatial setting within which the action chain of stowing is realized (Langacker 1987: 308). Likewise, on the soft ground in (15) is not directly involved in the action of horse riding, but specifies an aspect of the spatial setting on which the jockeys were riding faster. The circumstances of location in (14) and (15) thus relate to, and modify, the whole action chain expressed by the clause nucleus.

Turning to instruments, we should first note, following Nishimura (1993: 490f), that in an approach which focuses on the way in which representational roles are construed by structures, circumstances of instrument are distinguished from inanimate agents of an instrumental kind, as illustrated by (16) and (17) respectively.

(16) Bill killed Jane with poison.

(17) Poison killed Jane.

(Nishimura 1993: 490)

As noted by Nishimura (1993: 495), these two roles do not alternate systematically for all examples. For instance, the instrument in a nonmiddle clause corresponding to middle Example (8) The top quality rods from Montegue, H&I, South Bend, etc. [...] fish well, above can only be construed as circumstance:

(18) The members of the local fishing club all fish with the top quality rods from Montegue, H&I, South Bend, etc.

(19) **The top quality rods from Montegue, H&I, South Bend, etc. fish.

The construal of agentive instruments in English requires the possibility of attributing a force of their own to them, e.g., by motion, or physical or chemical impact, so that they can be viewed as setting off the action chain. This is the case for poison in (17), but not for the top quality fishing rods in (19). Circumstances of instrument, by contrast, are not construed as playing an independent role in the action chain, but are dependent on the agent's act (DeLancey 1984: 207). They are not required to have a force of their own. Still, as an extension of the agent's will, circumstances of instrument are indirectly associated with the agentive end of the action chain: instruments wielded by agents relate to the process in an active, not in an undergoing way. Circumstances of instrument can thus be analyzed as modifying the agent-process relation.

In Section 1.1, we have thus outlined our view on the representational organization of the clause. The complementation patterns between verb and nominals define the inherent voice relations of agentive and patientive participants in the process. Circumstances are modifiers, mostly of process-participant nuclei, but in some cases of the process only. With this understanding, we will discuss the main attempts in the literature to capture the semantics of the middle in terms of constraints on predicates or in terms of a generalized semantic role of the subject.

1.2. Constraints on predicates

The predicate selection constraint most often put forward to typify the English middle is the transitive constraint. It has been formulated explicitly by, amongst others, Keyser and Roeper (1984) and Fellbaum and Zribi-Hertz (1989) within Government and Binding: the formation of middles requires transitive and bars intransitive verbs (Keyser and Roeper 1984: 382). A transitive constraint was also implied in the analyses proposed by Halliday (1967), Fawcett (1980), and Davidse (1992) in the Systemic Functional tradition: the middle was viewed as one of the options within the representational system of transitive clauses.

A number of authors specified in addition to the transitive constraint that the patient/subject has to be affected by the action to explain why the transitive verb in (20) cannot occur in the middle (for a survey see Levin 1993: 25-26).

(20) **This metal won't pound.

(Levin 1993: 26)

We have argued elsewhere (Davidse and Heyvaert 2003: 63) against the affectedness constraint in view of counterexamples such as (21) in which the subject entity is not in any way changed or affected by the action of photographing.

(21) She does not photograph well [...].


Other types of middle in which the subject entity is not affected will be discussed below such as those with locations and instruments as subject.

The nonaffected nature of the subject in examples such as (22)-(25) with compound-transitive verbs may be one of the reasons why they are not analyzed as middles by Levin (1993: 82, 173).

(22) That whole wheat flour bakes wonderful bread.

(Levin 1993: 82)

(23) This wood carves beautiful toys.

(Levin 1993: 173)

(24) Their organic, whole wheat flour bakes extraordinary bread!

(Google) (8)

(25) Professional bakers have understood for years that better flour bakes better bread.


However, as will become clear below, these examples have all the criterial features of middle voice that we will define in Section 2. As we will argue, it is important for a correct characterization of the English middle to recognize that not just the form of the predicator, but also the structure of the whole predication--including the possibility of a DO--is that of an active clause. At this stage, we note that some compound-transitive verb uses are possible in middles.

As another correlate of the transitive constraint, Fellbaum (1986: 15) also formulated a constraint on mental predicates in middles: "[v]erbs of perception, doubt, emotion, etc. are all excluded from this [the middle construction, K.D. and L.H.] type". This constraint was not subscribed to by Halliday (1967: 49), who had given the counterexample:

(26) The bass notes don't hear very clearly.

(Halliday 1967: 49)

Corpus examples confirm that the constraint on mental predicates in middles does not hold:

(27) You will not have to worry about drifting of TX or RX signal, it hears well.


(28) It 'hears' well and this well-played and inexpensive CD should win the work (and the composer) new friends.

(Google) (9)

The core of the transitive constraint is, of course, that intransitive verbs are impossible in the English middle. We will argue, however, that some types of English middle examples have to be analyzed as containing intransitive verbs and either circumstances of location or instrument as subject (see also Davidse and Heyvaert 2003; Heyvaert 2003).

Examples with LOCATION subjects and intransitive verbs seem to be found mainly in sports registers, e.g.

(29) You wouldn't give my tip a chance of landing Aintree's Martell Red Rum Chase [... ] but with the ground riding slower, he should improve dramatically.


(30) A middle for diddle draw on Yarmouth's straight course means a fancied runner can switch to either rail depending on which side is riding faster.

(Google) (10)

In these utterances, the speaker assesses the condition of the surface on which the sport in question takes place, and how it affects the game. For instance, middle construction (29) expresses that the ground in question is conducive to slower horse riding, presumably because, as implied in the context, it is rather soft. The aspect of Examples (29)-(30) that we are interested in here is whether their verbs are transitive or intransitive. We hold that in all of them the relation between the verb and the location construed as subject is that of an intransitive process (with implied intransitive agent) and a circumstance of location. This is shown by the impossibility to construe these specific locations either as object in an active clause or as subject in a passive clause, while retaining the representational relations of the middle examples. For instance, corresponding to (29) we cannot get: (11)

(29)' **The jockeys rode the ground slower.

(29)" **The ground was ridden slower (by the jockeys).

There are semantic mechanisms that can sanction the transitivization of the relation between an intransitive verb and a location. These mechanisms include holicity of the relation as in run the whole track, or interaction with the entity as in ride a horse (e.g., Langacker 1991: 303). However, these semantic mechanisms do not apply to ride and the ground in (29) and ride and which side in (30). We conclude that the subjects in (29)-(30) designate the spatial settings of intransitive processes.

Our corpus data also contained middle examples, in which, in contrast with (29)-(30), the verb can be given both intransitive and transitive readings. Example (31) is a case in point.

(31) The top loch is fishing well.


On a transitive reading, the top loch in (31) is what Halliday (1967: 61) has characterized as a pseudo-transitive complement. In this reading, the loch can be construed both as bare postverbal complement in an active clause as in (31)' or as subject in a passive clause as in (31)".

(31)' Local fishermen are fishing the top loch.

(31)" The top loch is being fished by local fishermen.

Even though the semantic component of spatial expanse in the top loch prevents it from being a prototypical patient of the verb fish, a pseudo-transitive relation may be construed between the process of fishing and the loch, akin to that in expressions such as work the field, climb the mountain, etc. Fish a loch construes the loch as a sort of patient in the process, for instance as "to be emptied of its fish". This transitive construction may also imply that the loch is being fished across most or all of its surface; that is, it may have a holicity effect associated with it (Langacker 1991: 303). If we assume a transitive process-participant configuration for middle Example (31), then it conveys that conditions in the loch are such that "fishing it" is not difficult or yields a good catch.

However, the representational layer of (31) can also be interpreted as containing the intransitive verb fish, an (implied) intransitive agent and a circumstance of location. The intransitive clause corresponding to it is then

(31)''' Local fishermen are fishing on the loch.

With this process-participant configuration, middle Example (31) The top loch is fishing well conveys that "in the top loch conditions are such that fishing can take place easily in it".

Finally, our corpus data also contained a number of middle examples that, in our view, have to be analyzed as having subject locations and intransitive verbs, even though these verbs and nominals, given different contextual factors than those associated with the middles in question, can also receive a transitive reading. One such example is

(32) [about a tennis court] It is slightly coarser, so it plays a bit slower.


Play the court (12) does exist as a transitive verb-complement structure, as illustrated by an example such as

(33) Adams played the court with an impressive show of footwork during the first half of the game, dribbling circles around the members of the Kinfolk.

(Google) (13)

In (33) the construal of the court as a transitive object is accompanied by a holicity effect, conveying that Adams used all possible positions on the court well. By contrast, in middle Example (32), the tennis court plays a bit slower, the implied agent is not depicted as exploiting all the possible positions on the court more slowly, or as otherwise interacting with the court as a sort of patient. The representational layer of (32) simply has an implied agent playing more slowly on that court.

On some Dutch verbs of motion and activity, the distinction between intransitive and transitive meaning is morphologically marked by the prefix be- (14). Due to this formal marking, the distinction between middies with intransitive and transitive verbs is incontestable in examples such as (34)-(36) (Peeters 1999: 360-366).

(34) De Matterhorn beklimt/klimt gemakkelijker dan de Mt Everest.

'The Matterhorn climbs (transitive/intransitive) more easily than Mt Everest.'

(Ackema and Schoorlemmer 1994: 75; Peeters 1999: 363)

(35) Dat paard berijdt/rijdt goed.

'That horse rides (transitive/intransitive) well.'

(cf. Peeters 1999: 363)

(36) Die piste rijdt/* berijdt goed.

'That track tides (intransitive/* transitive) well.'

(cf. Peeters 1999: 362)

The use of transitive and intransitive verbs in Dutch Examples (34)-(36) correlates with the same sort of semantic differences as discussed for the transitive/intransitive readings of the English examples. Transitive be-klimt in (34) has a holistic meaning in that it conveys that the Matterhorn has properties conducive to climbing the whole mountain, whereas intransitive klimt merely expresses that the Matterhorn has properties making for easier execution of the act of climbing. The middle with transitive berijdt in (35) construes the horse as a patient of the implied human agent's riding and states that the horse is easy to ride, while the middle with intransitive rijdt expresses that on this horse riding is easy. These transitive/intransitive meaning differences are analogous to the different transitive/intransitive readings we discussed with regard to English examples like (31). In Dutch middle Example (36), only intransitive rijdt is possible, with de piste ("the track") construed as location of the riding. Patient construal of de piste with transitive verb berijdt is excluded here, even though the transitive structure as such is possible in an example like:

(37) De piste is stuk omdat ze te veel bereden is.

'The track is broken because it has been ridden too much.'

In this respect, Dutch middle (36) is analogous to an English middle like (32).

Of course, one always has to be careful adducing evidence from one language as support for a distinction in another. In this case, however, the types of structures and their semantics are fully comparable in Dutch and English. Not only are Dutch and English closely related historically, Dutch and English middles are also structurally analogous. In both languages an active predication is integrated with a nonagentive subject. Moreover, the correspondences between intransitive/transitive pairs such as klimmen/beklimmen and the intransitive/transitive uses of English verbs such as climb are well established (e.g., Dik 1980; Laffut 1998). For these reasons, we view the morphological marking of the intransitive/transitive contrast in Dutch Examples (34)-(36) as further support for the intransitive/transitive analyses we posited for English examples such as (29)-(33) on the basis of their alternation potential and semantics.

Thus, we conclude that the intransitive readings of English examples such as (27)-(32) invalidate the transitivity constraint on English middles. Examples (27)-(32) constitute the sort of data McGregor (2002: 130) refers to as "uncomfortable and exceptional facts" for the alleged transitivity constraint on English middles. They require us to give up the entrenched transitive generalization of English middles and force us to start looking at middles in a different light.

Let us now rum to the discussion of examples with INSTRUMENT subjects such as:

(38) Narrow tyres manoeuvre more easily.


(39) The top quality rods from Montegue, H&I, South Bend etc are good rods, fish well and look good.


In the literature, two analyses of the representational layer in such examples seem to have been considered. Smith (1978) hedges between these two analyses, e.g., (40).

(40) That brush paints well.

(Smith 1978: 103)

In one reading, that brush is treated as the agent of painting, in the other it is treated as a circumstance of instrument dependent on the agent implied by the verb paint. Only in the latter reading is there the relation between a nonagentive subject and active VP characteristic of the middle voice. In Smith's positing of these two analyses for (40), the problems surrounding the notion of instrument discussed in Section 1.3 seem to carry through. As it is not possible to ascribe independent action to the brush in (40), we analyze it as a circumstance of instrument, which presupposes the nucleus of intransitive brush and its implied human agent. However, an example like (41)

(41) Arsenic kills easily.

is ambiguous between a reading in which arsenic is the immediate agent of the killing and a middle reading conveying that with arsenic it is easy for an implied agent to kill. Corpus examples (38)-(39), however, can according to the criteria discussed in Section 1.3 be given a middle interpretation only. Narrow tyres in (38) and the top quality rods from Montegue, H&I, South Bend etc in (39) cannot be conceived of as independent agents in their processes, but have to be seen as instruments conducive to easy maneuvering and successful fishing on the part of the agents using them.

It should also be noted that besides instruments that are handled directly by the agent, there are also circumstances of "means" (Halliday 1985: 139), which are used more indirectly in the execution of the action. This is, for instance, the case with his music in (42) (see also Van Oosten 1986: 84).

(42) That is okay, because I am a Di Sarli fan. His music dances well. (15) We will continue to use the standard term "instrument" with the proviso that it covers both directly handled tools and more indirect and abstract means.

In conclusion, we have established in this section that there exists a subclass of English middles with intransitive verbs whose subjects are circumstances of either location or instrument. Thus the transitive constraint on middles has been invalidated. We have also adduced counterexamples to some constraints that had been formulated as correlates of the transitive constraint, viz. those on compound-transitive and mental verbs. Given that the transitive generalization of middle verbs cannot be maintained, we will investigate in the next section whether any of the proposals ascribing one specific representational role to the middle subject hold.

1.3. The representational role of the subject

The semantic role generalization most often proposed for English middles is that its subject is always PATIENTIVE and that an agent is necessarily implied, even though it cannot be expressed overtly (Sweet 1891; Jespersen 1914-1929, vol. 3; Halliday 1967; Smith 1978; Keyser and Roeper 1984; Fellbaum 1986; Fellbaum and Zribi-Hertz 1989). In support of the nonagentivity of the middle subject it has been pointed out that it is impossible to add expressions such as all by itself and on its own to the construction (Keyser and Roeper 1984: 405; Van Oosten 1986: 34).

(43) The book read easily. * The book read easily all by itself.

Likewise arguing that the subject/patient in an example such as (44) is not agentive, Langacker (1991: 405) aptly remarks that this example does not lead us to picture the ice cream scooping itself out--an agent scooping out the ice cream is necessarily implied.

(44) This ice cream scoops out quite easily.

(Langacker 1991: 405)

We fully agree with these arguments but with the important proviso that the nonagentivity of the subject corresponds to a patient role in middies with transitive verbs only, but to a circumstance of location or means in middies with intransitive verbs, as established in Section 1.2. In other words, the invalidation of the transitive constraint also undercuts the generalization that the subject is always patientive.

Another argument given for the point that an agent is implied in middles is the common occurrence of agent-oriented adverbs such as easily, well, etc. (e.g., Fellbaum 1985). These adverbs can only be made sense of if we state the semantics of the middle in terms of an agent. For instance, in (44), it is the implied agent who reads the book easily. However, Hale and Keyser (1987: 18f) turn this argument around and state that the impression of an agent arises only via an implicature of the agent-oriented adverbs found in many middles. They reject, wrongly in our opinion, the idea that the semantics of the middle contain a nonexpressed agent. In view of this controversy and given that the middle structurally disallows the explicit expression of an agent (Keyser and Roeper 1984: 406), as in (45),

(45) Bureaucrats bribe easily. [??] **Bureaucrats bribe easily by managers.

(Keyser and Roeper 1984: 406)

the precise status of the implied agent needs further clarification.

We propose that the difference between explicit agents and the agent implied in the middle can be elucidated in terms of Langacker's (1987: 183) notions of the "profile" and the "base" of a linguistic predication. Langacker (1987: 183) defines the "base" of a linguistic symbol as the whole conceptual domain invoked by it. Concomitant to this, the "profile" is defined as the substructure of the base actually designated by the linguistic expression.

In middles with transitive verbs, like (44) The ice cream scoops out easily, the active transitive predicate scoops out evokes in its base a two-participant process, with an agent scooping out the ice cream. The transitive lexical verb crucially brings in the concept of a transitive agent in the conceptual base. What is profiled by this middle construction, however, is the relationship between the process and the patientive participant construed as subject (Langacker 1991: 335), i.e., the process of scooping out and the ice cream being scooped out. The patientive entity is the "only participant within the profiled segment of the action chain" (Langacker 1999: 39).

Intransitive middles can be given an analogous semantic analysis. They profile an intransitive process in its relation to a location or instrument, but their base contains the agent who actually executes that intransitive process. For instance, in Example (32) cited above, The tennis court plays a bit slower, an agent is invoked who does the playing on the court. Likewise, in (39) The top quality fishing rods [...] fish well, the conceptual base of the predicator fish contains an agent who does the fishing and handles the rods.

We conclude that the notion of an agent plays a role in the semantics of middles because it is part of the conceptual base of the transitive or intransitive lexical verb. This is also the reason why agent-oriented adverbs, which refer to the agent in the base, can be used in middles.

Interestingly, there is also a tradition in the literature on the middle that proposes a semantic role generalization opposite to that of the "affected" subject, viz. one that claims that middle subjects are in some sense AGENT-LIKE (Lakoff 1977; Hale and Keyser 1987; Van Oosten 1977, 1986). For instance, Van Oosten (1977, 1986) characterizes the middle subject as "energy source of the action" (1986: 85) and as being able to "bring about the action of the predicate independently" (1986: 93). However, this claim is untenable in view of the arguments given above. An agentive reading of the subject results in semantic incongruity, as also reflected by the middle's resistance to continuations bringing out a degree of agentivity in the subject such as all by itself, and on its own. In this context, Keyser and Roeper (1984: 405) have correctly contrasted middle constructions like (46) with ergative intransitives like (47):

(46) **Bureaucrats bribe easily all by themselves.

(47) The boat sank all by itself.

(Keyser and Roeper 1984: 405)

They point out that the subject of an ergative intransitive does have a degree of agentivity and autonomy in its relation to the process (on this point, see also Smith 1978; Davidse 1992; Kemmer 1993; Levin and Rappaport 1995).

In our view (Davidse and Heyvaert 2003: 59-61), approaches attributing agentivity to the middle subject suffer in different ways and to varying degrees from an "ergative fallacy". There are three fundamental problems with analyses that reduce middles in some way to ergative intransitives. All these problems are due, in our opinion, to a confusion of levels and to attempts at making generalizations about the middle at the wrong level, viz. the representational rather than the interpersonal level.

Firstly, middles are claimed to in some sense reconstrue the intransitive, transitive, and compound-transitive verbs in them as ergative intransitive ones. However, as we have argued throughout Section 1, it is crucial to see that the various types of predicates and their complements, and the process-participant relations which they express, are left intact by middle construal. They account for one layer of the middle's semantics, viz. the representational semantics of the different process-participant(circumstance) configurations being depicted. Secondly, by presenting the relation between subject and VP in the English middle as basically one between an agentive role and an active VP, the description becomes that of an active clause and the analysis in fact cancels the notion of a middle voice. As we will argue in Section 2, there is a specifically "middle" component in the relation between subject and finite, but it has to be conceived of in an interpersonal, modal, way. Thirdly, in attempting to capture the special semantic value of the middle subject, the claim is made that properties of the subject are responsible for bringing about the action independently (Van Oosten 1986: 93). As a claim on the representational level of causality and event instigation, this analysis is, for the reasons discussed above, untenable. However, in Section 2, we will redefine the notion of the "responsibility" of the subject in radically interpersonal terms and relate it to the modal statement made by the middle as an interpersonal construction.

2. Interpersonal analysis of the middle

2.1. Interpersonal clause organization

The main claim we make in this article is that the characteristic properties of the middle reside in the specific interpersonal semantics construed by its subject-finite-predication structure. Therefore, we will first briefly outline our general understanding of the interpersonal organization of the clause within the functional-cognitive approach that we assume.

In this approach, the central element in interpersonal clause structure is the FINITE, i.e., the element of the VP expressing primary tense (Halliday 1985) and/or modality (Halliday 1970; Verstraete 2001; Davies 2001). In English, this functional element is realized either by the primary auxiliary or by morphological marking of the lexical verb. The rest of the VP fulfils the function of predicator (Halliday 1985) and is realized by the lexical verb and secondary auxiliaries expressing passive, progressive or perfect (Langacker 1991).

The clausal function of the finite is to "ground" the process-participant-circumstance configuration, i.e., give it a point of reference in the "here and now" of the speech exchange (Langacker 1991; Halliday 1985). This may be done either in terms of tense, which locates the utterance relative to the time of speaking, or in terms of modality, which construes the speaker's or the hearer's attitude with regard to such notions as obligatoriness or likelihood of the proposition. In the interpersonal structure of the clause, the finite links the predication to the subject (Davies 1979: 64). In functional and cognitive frameworks (Halliday 1985: Ch. 4; Dik 1997: Ch. 10; Langacker 1991: Ch. 7) it is stressed that subject and object assignment do not semantically contribute to the representation of states of affairs but to the perspectivizing of the proposition.

In the rest of this section, we will propose our generalizations for the specific value of finite, subject and predication in the English middle, thus also showing that middle voice is an interpersonal construction. We will discuss in order: the modal value of the middle finite (Section 2.2), the subject as modally responsible element (Section 2.3) and the way different types of predication integrate with the subject-finite unit (Section 2.4).

2.2. The modal value of the finite in middle constructions

In this section, we will argue that the finite in middle constructions construes a modal perspective on the representational relations between the process, its participants and possible circumstances. This modal perspective, while being of the "dynamic" type of modality that also includes ability and volition, is specific to the middle and will be characterized in terms of Talmy's (2000) force-dynamic relations of "letting" and "hindering".

2.2.1. Claims in the literature. The point that middles involve a modal meaning is not new (Fellbaum 1985; Fagan 1992; Massam 1992; Iwata 1999). For instance, Massam (1992: 122) posits that all middle constructions contain a null element "which is in essence equivalent to the modal can", expressing possibility/ability. The point we want to make here, however, is that the modal value of the middle finite cannot be reduced to one of the traditional modalities.

The line that has been taken most frequently in the literature is to interpret the middle VP as expressing a "can be V-ed" relation, as in Fagan (1992: 54), who paraphrases the meaning of (48a) as (48b).

(48) a. [about a kind of siding:] It nails easily. It cuts easily.

b. It can be nailed easily. It can be cut easily.

(Fagan 1992: 54)

However, this paraphrase changes the syntactic structure of the middle and does not work for all subtypes. Syntactically speaking, the can + passive predicator paraphrase adds the modal auxiliary can as finite and replaces the rest of the VP, i.e., the active predicator of the middle, with a passive one. For instance, nail in (48a) is turned into be nailed in (48b). Because it has a passive predicator, this paraphrase is restricted to transitive verbs and does not apply to middles with intransitive verbs such as (39) The top quality rods from Montegue, H&I, South Bend etc fish well, cited above:

(39)' *The top quality rods from Montegue, H&I, South Bend etc. can be fished well.

Importantly, the use of a passive predicator after can also changes the structure of the whole predication in comparison with that of middle constructions. Unlike middle predications, passive predications following can are able to take an agentive by-adjunct (49) but cannot take a DO (50) or an adverbial oriented towards the subject/patient (51). (Systematic alternation is indicated by ::, and nonalternation by ::**.)

(49) That siding nails easily :: That siding can be nailed easily by anyone who tries.

(Fagan 1992: 51)

(50) This wood carves beautiful toys :: *This wood can be cut beautiful toys.

(Levin 1993: 51)

(51) Sheila seduces willingly :: *Sheila can be seduced willingly.

(Lakoff 1977: 250)

If a paraphrase builds in a wrong constraint on verb selection and fundamentally changes the syntactic properties of the original, then it cannot be viewed as a systematic alternate that makes the semantics of the original construction more explicit.

A number of authors have also commented on the use of modal auxiliaries won't or wouldn't in English middles with negative polarity (see also Jespersen 1914-1929, vol. 3: 349; Halliday 1967: 47), as illustrated by the following examples:

(52) (...) Mr. Fabius chose to exit by a false door. (...) the scenery began to shake in response to the ex-prime minister's desperate efforts, but still the door would not open.


(53) (...) Mrs. Dambar was having some difficulty with the new dress she had bought (...). For some reason, it wouldn't zip all the way up in the back.


(54) I tried to open the door, but the key wouldn't turn.

(Quirk et al. 1985: 229)

(55) I am at a sentence that will not write.

(Jespersen 1914-1929, vol. 3: 349)

According to Quirk et al. (1985: 229) the negation of will/would in these clauses has "something of the personificatory force of 'refusal'" (see also Ehrman 1966; Huddleston 1969). Quirk et al. thus interpret modal will not as expressing a notion akin to that of the dynamic modality of negative volition: the negation in will not bears on the modality and in an example like (52) expresses something like "the door did not want to open".

Because "volition" in the strict sense requires a human subject and middles have inanimate subjects, Quirk et al. ascribe an element of personification to the use of will not in middles.

In the general argument for a modal analysis of the middle finite, it is significant that one subtype of middle in English does have overt modal marking. Middles with will not assess the "resistance" of an entity towards realization of the process. However, the point that a modal claim is made rather than a statement about a specific state of affairs applies also to the middles without overt modal auxiliary. The most common type of middles in English, viz. those with a positive VP without overt modal, do not designate actually occurring events either but also make a statement about some modal concept--one that is in some way opposite to "resistance".

In the next section, we will develop a more systematic semantic analysis of the modal finite with inanimate subject in middles. We will do this in terms of Talmy's (2000) "force-dynamics", a semantic framework set up to counterbalance the bias in favor of human agency in most existing treatments of causation and modality (16). We will argue that the force dynamic relations of "letting" and "hindering" provide the necessary general concepts that were lacking in the traditional approach to modality, showing that they can be adapted to capture the specific semantic value of the finite in its relation to subject and predication in middle constructions.

2.2.2. The letting modality in middles. Talmy (1985, 2000) has introduced "force dynamics" as a hitherto neglected but fundamental semantic category that "most uniquely characterizes the grammatical category of modals as a whole" (Talmy 2000: 409). Force dynamics refers to the opposition of two forces with different roles and different strengths. One force-exerting entity has the role of agonist; linguistic construals focus on whether or not this entity is able to manifest its tendency toward action (or inaction). The second force-exerting entity has the role of antagonist: it is considered for the effect it has on the inclination to action of the first, overcoming it or not (Talmy 2000: 413). "According to their relative strengths, the opposing force entities yield a resultant" (Talmy 2000: 414), which, Talmy claims, is always assessed in respect of the agonist in linguistic construals.

Turning to the issue of the modal meaning of the middle finite, we find that Talmy's (2000: 441) characterizations of the traditional notions of volition and ability further clarify why they cannot be used to capture the semantics of the middle.

According to Talmy, the negation of volition in an example such as

(56) John won't tell the truth.

"indicate[s] refusal by the subject to yield to external pressure to perform the expressed action" (Talmy 2000: 441). The human subject John is the agonist while the antagonist force is constituted by the psychosocial pressure "to tell the truth" that is implied in the utterance. By contrast, in middles with will + not, the agonist is the agent implied by the lexical verb and the inanimate subject is the antagonist. The subject-finite unit expresses the "hindering" or "blocking" (as defined by Talmy 2000: 426-430) of the action towards which the agonist is inclined. For instance, in (54) the key wouldn't turn, the key offers resistance to the agonist's trying to turn it.

The different agonist-antagonist roles in clauses expressing negative volition and in middles with will+not also correlate with different basic "force interaction" patterns. With volition, the relation between the subject/agonist and the action expressed by the lexical verb is to do with "causing", which Talmy (2000: 419) defines as "beginning" to exert force resulting in action. In middles, the relation between the subject/ antagonist and the action expressed by the lexical verb hinges on "letting", that is, "ceasing" to exert a force blocking action (Talmy 2000: 419).

In English, the modal notions of hindrance or blocking expressed by the auxiliaries won't and wouldn't can also be realized by a lexical verb such as refuse:

(57) (...) although the key turns, the door will refuse to lock (or to unlock). (CB--ukmags)

(58) (...) when the matches refused to strike. (Jespersen 1914-1929, vol. 3: 348)

This ties in with Talmy's (2000: 443) point that the "greater modal system" in languages also includes expression by lexical verbs. Apart from the realization of the modal notion of hindrance by an open-class lexical verb, these examples share all the formal and semantic properties of middles with will + not.

Middles with negative polarity and without overt modal auxiliary construe the same force dynamic pattern. They also assess how the properties of the subject entity hinder the agonist's action, as illustrated by

(59) His earlier short stories don't read so well. (CB--times)

This modal judgment of the subject's hindering action is not overtly expressed by a modal auxiliary, but is the result of the subjectified meaning of the VP. The subjectification at stake here is that from a nonmodal VP to a VP conveying DYNAMIC MODALITY. According to Traugott's (1989: 36) gradient view of subjectification in the emergence of modal meanings (17), this shift is an instance of the second degree of subjectification, that from main verb towards deontic modal meaning (Traugott 1989: 43). (In Traugott's [1989: 46] continuum, the subjectification leading from deontic to epistemic modal meaning constitutes the third or highest degree of subjectification.) Declerck (1991: 183) discusses a comparable case in which modal meaning is conveyed by a bare VP, in examples such as Chris understands/speaks Japanese, which attribute ability to the subject. In middles without overt modal auxiliary like (59) the VP has in a similar way acquired the dynamic modal value of "not letting". In Traugott's (1989) general terms, this modal meaning of the VP in a middle like (59) represents a shift from representational to interpersonal meaning, from designation of a verifiable state of affairs to expression of a subjective speaker judgment (Traugott 1989: 31). (18) Thus, the VP in a clause such as I didn't read the newspaper yesterday locates the nonoccurrence of my reading the newspaper at a specific moment in the past. By contrast, the negative VP in middle Example (59) His earlier short stories don't read so well, expresses the speaker's judgment that the properties of the short stories are such that they hinder reading enjoyment. Middles such as (59) thus express the same dynamic modal meaning of "hindrance" as middles with will + not, but through subjectification of the VP rather than by an overt modal.

The semantics of middles with positive polarity and without overt marking of modality have, as mentioned above, often been analyzed as expressing that the subject "can be V-ed". By applying Talmy's semantic framework to them, we can analyze the precise "force dynamics" expressed by a clause such as (48) with can + passive predicator.

(48)' The siding can be nailed easily (by anyone who tries). (cf. Fagan 1992: 54)

The agonist tending towards action is the implied agent (who, in this type of construction, can be made explicit in a by-adjunct). The modal can profiles the agonist's ability to overcome resistance (whatever its nature) to realization of that action. In the traditional approaches to modality, can is said to be intrinsically agent-oriented (e.g., Halliday 1970: 339). Indeed, can of ability ascribes the ability to execute the process designated by the lexical verb to the agent irrespective of whether the agent is construed as subject in an active clause or as by-adjunct in a passive clause, or is merely implied in a passive clause. However, Ehrman (1966: 14) has also pointed out that passive clauses with can often highlight that "there are certain positive qualities of the subject such that the way is cleared for the predication", as in:

(60) These engines can be removed from a boat with relative ease (Ehrman 1966: 14)

Ehrman's analysis is fully consistent with Talmy's "force dynamic" principles. In (60), the subject is construed as a force that has an effect on the agonist's inclination to remove the engines. More precisely, the subject these engines is the antagonist that "lets" the agonist execute his intended action by not putting any barriers in the way.

The corresponding middle has the same agonist and antagonist roles and realizes the same general force dynamic pattern of "letting":

(60)' These engines remove from a boat with relative ease.

The subject these engines is the antagonist that does not offer any resistance to the implied agent's removing them. Purely in terms of agonist-antagonist roles and "letting" force dynamics, clauses with can + passive predicator and middles with positive polarity are analogous. This is, no doubt, the reason why the former have so often been used as a semantic paraphrase of the latter. However, the analogy is only partial, as the two types of construction make statements about different types of modality. Passive clauses with can, even if they foreground the inherent properties of the patientive entity in subject position, judge the ABILITY of the agonist to carry out the action. By contrast, the middle does not assess the agonist's ability but the subject's LETTING modality. As in middles with negative polarity and without overt modal, this modality arises via subjectification of the VP: in middle examples like (49) and (60)' the bare VP has the dynamic modality of "letting".

We thus see how Talmy's concepts of letting and hindering allow us to elucidate the modal meaning of the middle finite in a way that the concepts from the traditional treatments of modality did not. However, "letting" and "hindering" are still fairly general notions applicable to other types of modality as well. In the following paragraphs we will identify the semantic components that are specific to the letting/hindering relation in middles. We will also spell out how these semantic components account for the specifically "middle" semantics of the construction, that is, in what way the letting modality involves a mix of active and passive.

Firstly, the "letting" relation has a PASSIVE aspect in that the subject/ antagonist is always lower on the semantic role hierarchy than the agentive agonist: it is either the patient targeted by the agent, the location on which the agent moves or is positioned, or the instrument used by the agent. Therefore, the letting role of this subject entity can be understood as one of "lending itself to" the agonist's action. This adds what is in a broad sense a semantically "passive" feature to the relation between subject and predication.

This passive element in the modal layer of the VP is foregrounded in a type of construction that shares many functional and formal parameters with middles, viz. the "let itself"-construction found, amongst others, in Dutch and German (see also Fagan 1992: 211). This construction type that makes use of the general reflexive verb "let itself", zich laten in Dutch and sich lassen in German, can be illustrated by the following Dutch and German examples.

(61) GSM-antwoordapparaat laat zich gemakkelijk mobile phone-answering machine lets itself easily kraken (19) hack 'mobile phone-answering machine is easy to hack'

(62) Der Wagen lasst sich angenehm fahren. the car lets itself pleasantly drive 'the car is pleasant to drive' (Fagan 1992: 211)

The "let itself" construction has the same basic force dynamic set-up as the middle: the subject entity is the antagonist, with a letting relation to the agonist's action designated by the complement verb (kraken in [61], fahren in [62]). Moreover, like the middle and unlike constructions with can + passive predicator, its main claim is about the subject's letting relation to the agonist's action, which is lexicalized by the reflexive verb zich laten/sich lassen. Syntactically, the VP in this construction has an active predicator (kraken, fahren), but its modal layer is passive due to the use of reflexive verb zich laten/sich lassen, as shown by the fact that the agent can be expressed by a door/durch 'by'-adjunct.

(63) Dat boek laatzich door elk geinteresseerd lezer lezen. that book lets itself by any interested reader read 'that book can be read by any interested reader'

As a result, a transitive selection restriction attaches to the predicator. This restriction is illustrated by (64), which, in contrast with its middle counterpart Die berg klimt/beklimt goed, ('that mountain climbs well', see Section 1.2) cannot take the intransitive verb.

(64) Die berg laat zich goed *klimmen/beklimmen. that mountain lets itself well climb 'that mountain climbs well'

The English middle construction differs from the "let itself" construction in that the semantically passive relation between the letting subject and the agentive agonist does not receive overt realization. The predication of the English middle is that of an active clause and does not have any passive features: it does not impose transitive selection restrictions on the predicator and it bars agentive by-adjuncts. Still, to the extent that the middle assesses whether and to what extent the subject "lends itself to" the agonist's action, the traditional idea that middle voice involves a passive feature in the relation between subject and predicator is valid.

In the second place, there is also an ACTIVE aspect to the letting relation between the subject-entity and the agonist's action expressed by the predication. This active aspect stems from the fact that contextually invoked properties (20) of the subject entity are presented as positively conducive or as actively barring the action envisaged. It is for this reason that Langacker (1991: 334, p.c.) characterizes middle subjects as 'conducive' subjects (see also Heyvaert 2003: 142; Davidse and Heyvaert 2003). In for instance an example like (30) cited above, this side is riding faster, the speaker claims that certain properties of this specific side of the track, such as its harder surface, are conducive to riding faster on it. Likewise, an utterance such as The ladder fixes quickly and easily under the window ledge conveys that the properties of the ladder are such that they are conducive to fixing it quickly and easily under the window ledge. In other words, the relation of "letting" in middles conveys not simply that the subject entity does not oppose barriers, but also that features of that entity actively help the action along. Middles with positive polarity state that the subject, because of contextually invoked properties, is positively conducive to the action as specified by the predication. In middles with negative polarity, the subject is claimed to have properties that are not conducive to, i.e., actively hinder, the action.

Finally, because we propose that the middle finite is always intrinsically modal, we briefly discuss the reality status of the action described by the predicator. It seems to us important to point out that this action may remain purely virtual in some contexts of use, but may also be actualized in other contexts. (21) For instance, the middle in Example (65) makes a general statement about the conduciveness to cleaning of latex paint but does not necessarily imply any specific actualization of that action.

(65) We have latex paint on the walls in the kitchen, which cleans up quickly and easily.

By contrast, the proposition in Example (66) applies to one specific, but nonactualized, situation, and (67) does imply actualization of a specific situation:

(66) ... the matches refused to strike.

(Jespersen 1914-1925, vol. 3: 348)

(67) The window opened only with great difficulty.

(Langacker 1991: 334)

Importantly, this does not cancel the point that the propositions made in (66) and (67) are modal assessments of the entities' conduciveness to the acts described. As stressed by Declerck (1991) and Declerck and Reed (2002), modals typically have uses covering the whole range between nonimplied actualization, nonactualization and actualization of the situation described, as illustrated, in the same order, by the following examples with can of ability.

(68) As a life-saver, she can save people about to drown.

(69) He couldn't swim to the other side.

(70) He could just swim to the other side.

Summing up: in this section we have, with extensive reference to Talmy's (2000) force dynamics proposed a specific modal analysis of the finite in English middle constructions. We have first shown that the meaning of the middle cannot be captured in terms of the traditional modal concepts of ability or volition. The modalities of ability and volition assess how human agonist forces relate to the execution of their acts in the presence of implied antagonist pressures (Talmy 2000: 441). However, middles express a judgment of how the antagonist force exerted by the inanimate subject entity impacts on the agonist's action. Thanks to the fact of Talmy's having shown the general significance of the "letting" category to the analysis of modals, we have been able to characterize the relation between the middle subject and the finite as a modal letting relation. We have elucidated the semantics of this letting relation in such a way that it embodies the active-passive mix traditionally associated with the middle, viz. as both lending itself to and being positively conducive to the act designated by the lexical verb.

2.3. The conducive subject of middles

In the previous section, we argued that the modal value of the finite can best be characterized as a "letting" force dynamic relation. In this section, we will spell out what the analysis of the middle subject as antagonist of the modal "letting" relation contributes in comparison with earlier characterizations of the middle subject in the literature.

While Talmy (2000: 422) stresses that all the interrelated elements of a force dynamic pattern realized by a linguistic construction are invoked by that construction, he also points out that constructions can pick out different elements for explicit construal and leave the other elements implicit. This corresponds roughly to Langacker's (1987) distinction between the profile and the base of a linguistic predication. In our discussion of the various transitivity statuses of verbs occurring in English middles, we made the point that the agent of these processes is always implied in the base of the predicates (Section 1.3). Now that we have also discussed the modal force dynamics construed by the middle finite, we can formulate the generalization that, in leaving the agentive agonist implicit, middles are a construction characterized by "agonist demotion", a characteristic feature of modals with inanimate subjects (Talmy 2000: 442). Talmy also notes that agonist demotion often goes together with foregrounding of the antagonist, which may be "expressed earlier in the sentence or higher in a case hierarchy" (Talmy 2000: 422). This raises the question to what extent and in what way the antagonist subject in middles is foregrounded.

From our earlier discussion in Section 1, we conclude that the antagonist in middles is not foregrounded in terms of case hierarchy. As argued in Section 2.2.2, the semantic role of the subject in English middles is always lower on the case hierarchy than the agentive agonist. The middle subject constitutes what in functional frameworks (Halliday 1985; Dik 1997) is referred to as a marked mapping of semantic roles to the subject function, the unmarked mapping being between subject and agent role. The very fact that middle subjects in English are either patients or circumstances of location or instrument draws attention to the special factors motivating this choice of subject.

Our approach to the subject is based on Halliday's (1985) INTERPERSONAL characterization of its function. According to Halliday (1985: 76), the subject is the element that is made rhetorically responsible by the speaker for the truth or the persuasiveness of the proposition. It is the element singled out by the speaker as "the one on which the validity of the proposition is made to rest" (Halliday 1985: 76).

Langacker (1991: 306-307, p.c.) further elucidates this interpersonal characterization of the subject function by pointing out that its semantic value is inherently subjectively construed. This is not to deny that the objective situation described may offer various degrees of motivation for subject selection (Langacker 1991: 308-309, p.c.). Because agents are objectively responsible for the action described, rhetorical responsibility for the validity of the proposition can be naturally rested on them. With non-agentive entities, subject selection may be motivated by the speaker's subjective assessment of that entity, or it may simply be a matter of speaker choice to focus on that entity in representing the situation (Langacker p.c.). Whether objectively motivated or not, the subject function involves selection of an entity for purposes of ascertaining the truth of the proposition, and concomitant conferral of focal prominence on that entity (Langacker 1991: 306-307). In this respect, the subject function is inherently subjectively construed (Langacker 1991: 312, p.c.).

Homing back in on middle subjects, we will specify in the following paragraphs how, as nonagentive entities, they instantiate this subjective construal. More specifically, we will discuss the relation of the middle subject to the modal finite, and the implied speaker assessment of properties of the subject entity.

First, we will look at the relevance of the letting modal finite to the semantic value of the subject, a question that involves the ORIENTATION of the modality in middles. It is fairly generally accepted that, in English, modal finites may have three types of orientation: towards the agent, towards the subject, or towards speaker or hearer (Halliday 1970; Declerck 1991). In Talmy's (2000) terms, the modal's orientation is towards its source force, irrespective of whether that source is agonist or antagonist. A typical example of agent orientation is formed by can of ability (Halliday 1970). The source force in ability is the one able to perform the action, i.e., the agent. If the modal VP is positive, this source force is also the stronger one in the force opposition. The agent-orientation persists, irrespective of whether the agent is construed as subject or by-adjunct: John can fly that jet. That jet can be flown by John. The source force of volition designated by will, by contrast, is the one willing to act or undergo the action. Volitional will is subject oriented (Declerck 1991: 363), as illustrated by He WILL revenge his mother.--He WILL be killed by his enemies. The subject always designates the volitional being inclined to the realization of the action, irrespective of whether s/he is involved in it as agent or patient. With volition, unlike with ability, the force designated by the subject is the stronger one also when the modality is negated, as in He will not accept help--He will not be helped. Finally, the source of the modality may also be either speaker or hearer, as with may of permission. The permission for the agonist to carry out an action depends on either the speaker, e.g., You may go now ("I allow you"), or the hearer, as in May I go now? ("do you allow me?") (Declerck 1991: 365). [22]

What, then, is the orientation of the letting modality construed by English middles, both those without and those with overt realization of the modal meaning, as in:

(71) These lines read well.

(72) These lines just won't read.

(73) These lines refuse to read.

Clearly, the modality in middles is subject-oriented. It is always the subject entity that is construed as exerting the modal "letting" force. Moreover, whether the modality is affirmed or negated, the letting source force is always the stronger one in the force opposition. It is thus the middle subject's relation to the modal finite that allows one to formulate a coherent semantic generalization: the entity designated by the middle subject is construed as being CONDUCIVE, or as not being conducive, to the agonist's action. Whether in the representational layer it is the entity targeted by the action, the instrument used in carrying out the action, or the location on which action takes place, in the modal layer of the proposition it is always construed as the conducive force that determines the resultant of the force opposition.

By locating the letting relation and the conduciveness of the subject in the interpersonal, modal, layer of the utterance, we avoid the problems incurred by a characterization of the middle such as Van Oosten's (1986). According to Van Oosten (1986: 93) "the purpose of the patient-subject construction is precisely to enable the speaker to assert that [...] properties of the patient bear the responsibility for the occurrence of the action of the verb". While this formulation could be read as merely very vague regarding the level of clausal organization on which the subject's responsibility is situated, other formulations of Van Oosten's in which the subject is for instance claimed to "bring about the action of the predicate independently" (Van Oosten 1986: 85) can only be understood as attributing a causal force to the subject entity in the objective situation being described. As alternative to this untenable characterization (see above Section 1.3), we have proposed the semantic generalization that the middle is a modal assessment of the subject's conduciveness to realization of the action described by the predicator.

The second subjective element to be pointed out about the middle subject is that it implies a value judgment of properties of the entity designated. The modal claim made by the middle always includes reference to inherent properties of the conducive subject. The speaker asserts of the subject that it is conducive or not conducive to the agonist's action in virtue of its specific "letting" or "blocking" properties. For instance, an example such as

(74) Tube tent. Sets up in minutes! (Google) [23]

refers to the design of the tent and evaluates how this affects its installation.

Occasionally, these conducive properties are explicitly named in the subject itself, as in

(75) Compact lipstick size design, slips easily into your handbag or pocket.


or in the surrounding discourse, as in

(76) Her face is her best asset because she has very good bone structure which will photograph well.


Often, however, the properties that make the subject conducive to the predicated action remain implicit. The conduciveness of the subject as such is part of the linguistic semantics of the construction, but what these conducive characteristics are has to be inferred from the context or from general knowledge and belongs to the area of pragmatics. In this respect, the middle is a paradigm case of the tenet (see e.g., Horn 1992) that the linguistic semantics of a construction are more restricted than the actual message that is exchanged between speaker and hearer, which involves a lot of inferences on the basis of contextual and cultural factors. For instance, the intransitive middles with location subject found in sports registers (see Examples [29]-[30]) above evaluate how properties of pitches, greens and tracks influence the sports activities taking place on them, but they are fully interpretable only by specialists in these sports who are in on all the technicalities.

It is important to note that in evaluating the conduciveness of implied properties of the subject entity, the speaker also assesses the entity itself. Pragmatically, middles often amount to a quality judgment of the subject entity. We will return to this feature in the next section in which we discuss how specific claims made in the predication may link up with qualities imputed by implication to the subject entity, as in:

(77) The new Holden Berlina handles like a junior sports sedan.


The point to be made in this section is that subject selection in middles is clearly motivated by subjective speaker assessment of the entity being designated.

In sum, the subject in middles is strongly foregrounded, not in terms of case hierarchy, but in purely subjective terms. Firstly, its very construal as a conducive entity in relation to the "letting" modal involves subjective speaker-assessment. Secondly, the subject entity is further subjectively evaluated by the speaker in terms of its contextually implied features.

2.4. The predicator and its complements and adjuncts

In this section we will discuss the last basic element of the interpersonal structure of the middle, the predication. We will look at the various ways in which the predication can be elaborated and relate these to the analysis of the subject-finite unit that we presented in Sections 2.2 and 2.3. In our view, the conduciveness of the nonagentive subject towards the realization of the process, expressed by the relation of subject and finite to the predicator, accounts for the semantic core shared by all middles. However, middle constructions can focus on various facets of this relationship between the subject entity and the process whose realization is being envisaged. Which facet is being focused on is typically indicated by the adjuncts or complements that the predicator is elaborated by. By sub-classifying middle predications in this way, one also arrives at a typology that corrects the bias towards 'facility-oriented' middles generally found in the literature.

A first main type of middle is that in which the predicator is NOT elaborated by any complements or adjuncts. This type simply focuses on whether the properties of the entity construed as subject are conducive to the action as such. For instance,

(78) [from a children's book] The laces of these shoes won't tie.--Yes, they TIE! (Davidse 1991: 44)

We can refer to this type as the PROCESS-Oriented middle. The example in (78) illustrates that process-oriented middles with positive polarity typically have intonational prominence on the VP. This means that the VP forms the "information focus" (Halliday 1985: 275), or the most salient information of the utterance: the speaker marks the predicator and its modal finite value as the "point" (Martin 1992: 448) of his or her message.

The type that has been discussed most in the literature is the FACILITY-oriented (Fawcett 1980: 148; Fellbaum 1985) middle. It focuses on whether the properties of the subject-entity are conducive to carrying out the process easily or with difficulty:

(79) The window opened only with great difficulty.

(Langacker 1991: 334)

(80) [about conditioning milk:] (...) rinses easily away and really works!


This facility aspect is typically specified by adverbs such as the very frequent easily and well, as well as other adverbials such as with great difficulty in (80). Note that in this type of middle it is the adverbials that typically carry the information focus and are thus presented as the point of the proposition made by the speaker.

A type of middle that is closely related to facility-oriented middles is that which provides a more general indication of the way in which the process can be carried out. Often, the specifications added to the predicator, such as the adverb in (81) and the adjunct of comparison in (82), express a quality judgment related to the qualities of the subject entity (Kemmer 1993: 147).

(81) That is easily done because the car handles superbly.


(82) Her life story reads like the Hollywood movies she's starred in. (CB--today)

Still other middle constructions comment on the (inferable) properties of a particular entity by indicating how these properties influence the TIME it takes to carry out a certain process. For instance,

(83) [about a cosy car seat protector:] Quickly attaches / removes with elastic straps and velcro tabs.


Some middle constructions specify the typical LOCATION of the subject entity (Fellbaum 1985: 28; Fagan 1992: 80).

(84) [about a children's coat:] Outer flap wraps around little hands and secures with Velcro (...).


(85) Playset folds up into a storage ease with handle for easy carrying.


Such middles may specify various things such as where the subject entity is meant to be placed to make it function (Example [84]) or where it is stored when it is not being used (Example [85]). Often, it is "deliberately designed properties" (Lemmens 1998: 80) of these entities that are conducive to placing them in the specified locations. The designed properties of the entity may also link up with a specification of the MEANS to realize the action, as in (84) (with Velero).

Finally, there exist middle constructions that specify the result of carrying out a certain process on the entity construed as subject, which, in accordance with Fellbaum (1986: 10) and Heyvaert (1997: 438), we will refer to as RESULT-oriented middles. An example of a result-oriented middle is (86), in which the woman functioning as subject is characterized as not photographing well. This does not mean that it is difficult to photograph her (facility-oriented meaning), but that the features of this lady's face are not conducive to successful pictures.

(86) She does not photograph well, and the portrait of her pinched features snatched outside the High Court only added to the picture of a shrew. (CB--times)

The conduciveness of the subject entity to a qualitatively good result may also be specified by evaluative adjectives in the complement of the predicator (as illustrated in [87]).

(87) Their organic, whole wheat flour bakes extraordinary bread! (Google, see [24])

In summary, the subject-finite unit in middles expresses that an entity is conducive to the carrying out of an action as specified in the predication. In this section, we have seen that the middle predication may contain various sorts of specifications: of the action as such, of the way in which it can be carried out, and of the facility, means, speed or result with which the action can be executed. All of these specifications focus on one aspect of the action that the properties and qualities of the subject entity are conducive to. For this reason, a number of middles pragmatically come down to being quality judgments of the subject entity. As observed by Fellbaum (1985), elements of middle predications may refer to inherent as well as noninherent, surprising, properties of the subject entity. For instance, Example (78) falls more into the inherent category as laces are normally meant to be tied, but an example such as (77) refers to a surprising property, as a Holden Berlina is not a sports car.

3. The interpersonal analysis of the middle voice and subjectification: a synthesis

We are now in a position to pull together all the threads and synthesize our view on the middle voice and the English middle. In Section 1, we have seen that in the literature so far middle voice in English has been approached mainly as a specific REPRESENTATIONAL relation. Two general semantic schemata have been proposed. The most traditionally entrenched schema (e.g., Jespersen 1914-1929; Halliday 1967; Fagan 1992) is that according to which the middle subject is always a patient and the lexical verb necessarily transitive, a claim that we showed to be untenable in view of the occurrence of English middles with intransitive verbs. In this schema, the middle VP is said to be formally active but semantically passive and to be paraphrasable as "can be V-ed". The specific import of the modal element that thus seems to be introduced is not spelt out and the middle generally continues to be treated as a sort of transitivity option, besides transitive, intransitive and ergative clauses. According to the second schema (e.g., Van Oosten 1986; Hale and Keyser 1987), middle clauses impose an agentive element on the patient/subject, which is represented as responsible for the action, and even as bringing it about independently. In this semantic schema, middle voice is thus interpreted as involving a specific, mixed agentive-patientive, way in which the subject entity participates in the process. We have argued against this semantic account, which in some ways equates the middle with the ergative intransitive. We then concluded that the specific semantic value of the subject-VP relation in middles does not correspond to one specific process-participant relation but is situated at a different level of clausal organization, viz. the interpersonal level.

In Section 2, we proposed that this INTERPERSONAL schema characterizing the English middle centers on the specific modal value of the middle finite. Against existing treatments in the literature in terms of the traditional notions of volition and ability, we developed an alternative characterization, with reference to Talmy's (2000) force-dynamic relations of letting and hindering. We proposed that the middle finite expresses a subject-oriented type of letting modality in which the subject is the antagonist conducive (or not conducive) to the carrying out of the action by the implied agentive agonist. The active-passive mix traditionally associated with the middle voice could then be located in the semantics of the modal relation between subject and finite. Because the subject entity in English middles is always lower on the case hierarchy than the agentive agonist, it is, in having its conduciveness assessed, evaluated as "lending itself to" the action envisaged. By the same token, the subject is construed as actively conducive or not conducive to the agonist's action in view of its contextually implied properties.

Thus in this discussion have emerged two distinct types of "active", "passive" and "activopassive" relations, which have not always been properly distinguished, particularly not in the discussion of middle voice. On the one hand, there are the inherently agentive and patientive relations between the lexical verb and its nominal complements. These representational relations between verb and nominals remain constant across active and passive clausal voice, as in:

(88) John marched the prisoners.

(89) The prisoners were marched by John.

The in-between option is formed by the participant in ergative intransitive clauses (see e.g., Keyser and Roeper 1984; Davidse 1992; Levin 1993; Kemmer 1993), e.g.,

(90) The door opened.

(91) The hat blew off.

in which agentivity and patientivity are mixed in the relation of the single nominal to the process. The events expressed by these clauses involve, as Kemmer (1993: 94) puts it, "conceptual fusion between initiating and affected aspects of an entity".

On the other hand, there are the options of clausal voice construed by the subject-finite-predication structure of the clause. In the active clausal voice an active predication is integrated with the subject by the finite, e.g.,

(92) Yesterday grandmother baked delicious bread with that organic flour.

while in the passive clause a passive predication is integrated with a patientive subject.

(93) Yesterday delicious bread was baked with that organic flour.

The middle voice constitutes the in-between option, in which the finite integrates an active predication with a nonagentive subject.

(94) That organic flour bakes delicious bread.

However, this is only possible by virtue of the finite construing letting modality between subject and predication.

Having established the ergative intransitive as in-between option in the system of participant voice and the middle as in-between option in the system of clausal voice, we can shed more light on a contentious issue in the literature, viz. the relation of middle constructions to ergative intransitive clauses. With regard to English, the hypothesis has often been formulated that diachronically the ergative intransitive is the source from which the middle developed (Jespersen 1914-1929; Curme 1931; Fellbaum and Zribi-Hertz 1989). The diachronic data analysis required to check this hypothesis is beyond the scope of this article. However, the analytical question underlying the diachronic one is directly related to the concerns of this article. In the literature, the lexical and grammatico-semantic relations between ergative intransitive and middle have been envisaged roughly as follows.

Firstly, it has been observed that the class of ergative verbs as a rule allows middle formation while there are at least partial restrictions on predicates of all other transitivity types occurring in the middle (Hale and Keyser 1987; Fellbaum and Zribi-Hertz 1989). Ergative verb uses are, across linguistic schools, identified on the basis of their allowing the ergative alternation (e.g., Lyons 1969; Cruse 1972, 1973; Keyser and Roeper 1984; Sinclair et al. 1987; Davidse 1992; Levin 1993): they are lexical causatives that have (acquired) an intransitive use (Levin and Rappaport 1995), as in The wind blew his hat off:: His hat blew off. The unresolved point in this argument is why, if the ergative verb class is posited as a link between ergative intransitive and middle clauses, middles do take verbs of other linguistic types besides ergative verbs.

Secondly, formal and semantic analogies between the two construction types have been investigated extensively by Kemmer (1993). She notes that crosslinguistically ergative intransitives and middles tend to have the same formal marking such as the reflexive clitic in French, Italian, Spanish, and German. Semantically, ergative intransitives depict spontaneous events, in which the initiating and affected aspects of the entity are conceptually fused (Kemmer 1993: 94). Kemmer links the semantics of these spontaneous events to the "facilitative" meaning of the middle, by invoking a comparable conceptual fusion of initiation and affectedness: "The Initiator status of the Patient, unlike in the case of the spontaneous events, derives from the fact that the event is conceived of as proceeding from the Patient by virtue of an inherent characteristic of that entity which enables the event to take place" (Kemmer 1993: 148). While enlightening, this formulation is not fully explicit about the mechanism linking the ergative intransitive and the middle, which we claim is one of subjectification. In the following paragraphs, we describe this subjectification mechanism in more detail, stressing that subjectification is invoked as a general analytical concept rather than to formulate a strictly diachronic hypothesis.

As observed by Fellbaum and Zribi-Hertz (1989: 35), decontextualized syntagms with ergative verbs tend to be ambiguous between two possible readings. Take Example (90) The door opened. If this clause is used to profile a state of affairs, depicting the door's opening as a semiautonomous event but hinting also at some external cause of the opening, then the ergative intransitive reading is activated. In this ergative intransitive reading the clause depicts the actual occurrence of a specific event, in which the door is profiled in an agentive-patientive relation to the process of opening. However, the same clause can also be used to assess the door's properties at some past moment in respect of its conduciveness to an agent's opening the door. This is the middle reading, which would make sense in a context such as "After having been oiled, the door opened easily again". This meaning includes the notion of a nonspecific agent for whom it is easy to open the door.

These two readings correspond to different constructional analyses of a syntagm such as (95). On the basis of the descriptive characterizations put forward in this article, we propose that the middle construction is linked to the ergative intransitive by a process of SUBJECTIFICATION, viz. a shift from the description of a verifiable state of affairs to a subjective statement of dynamic modality. This is the second degree of subjectification recognized by Traugott (1989) in the gradual development of modal meanings, over a continuum from nonmodal/premodal to deontic to epistemic. In Section 2.2.2, we already invoked this type of subjectification to elucidate the meaning of VPs without overt modal marking in middles. However, this shift also operates on the relation between subject and VP. More specifically, the agentive-patientive complementation of the lexical verb by the one participant designated by the subject in the ergative intransitive shifts to the activo-passive letting modality between subject and finite in the middle. As is often the case with subjectification mechanisms (Traugott 1989), this shift involves a "forward" move from the lexical verb to the finite in the VP, viz. from agentive-patientive elaboration site in the lexical verb to activo-passive letting modality construed by the finite. (24) Together with this shift in the VP, the semantic relations between subject and VP undergo a fundamental reanalysis. The subject's agentive-patientive relation to the lexical verb in the ergative intransitive is replaced by the subject's conducive relation to the modal finite in the middle. Subject selection in the middle is thus detached from any objective motivation such as agentivity--a process that is supported by the fact that the middle subject is foregrounded by the speaker on the basis of subjective evaluation of its properties.

With the subjectification of the relation between subject and VP, the subject is freed from the strict agentive-patientive participant relation that it had to the verb in the ergative intransitive. What is more, the letting force dynamics impose different agonist-antagonist relations. The subject is no longer agonist of the action but is construed as antagonist, and the letting force dynamics require the predicate to invoke an agonist in its conceptual base. The conduciveness of an antagonist entity to an action can only be assessed with reference to an agonist carrying out that action. The necessary incorporation of an agent in the predicate's base thus turns the subject into a straightforward patient of the transitively used ergative verb.

How such a shift can come about in actual language usage can be illustrated by a typical example such as (95)

(95) That door opens easily.

As argued by Evans and Wilkins (2000), emergent polysemy and semantic change requires that one form can plausibly be used in the same context with two different meanings, a phenomenon that Evans and Wilkins (2000) refer to as a "bridging context". Bridging contexts allow language users to select different contextual factors in support of either of the two readings. The two readings share some significant semantic features but also depart from one another by the loss and addition of features. With regard to Example (95) we can imagine a bridging context in which a door whose lock has just been oiled is being discussed. For one conversationalist, That door opens easily could be interpreted as ascribing an inclination to open to the door, implying for instance that it now opens too easily or too quickly, whenever there is a draft. In this reading, the process-participant relation between that door and open is still ergative intransitive. For another conversationalist, however, That door opens easily could convey a middle reading. Having been oiled, the door lends itself to easy opening by an implied agent. In this reading, the VP has a modal letting meaning, and the process-participant relations have the door as patient and open as transitive verb with implied agent. Numerous similar bridging contexts can be imagined in which middle readings plausibly co-occur with ergative intransitive readings (see also Langacker 1991: 335).

Interestingly, if we define the subjectification mechanism in the way just described, then the extension of the middle category to predicates that are not ergative can be explained. In the process that subjectifies an agentive-patientive process-participant relation into an activo-passive letting modality, the subject acquires a purely patientive relation to the lexical verb and an agent comes to inhere in the predicator. This explains, in the first place, why some transitive verbs that do not detransitivize, such as cut and handle, can be used in the middle analogously to the transitive uses of ergative verbs such as open. Just as in After having been oiled, the door opened easily again, the subject the door is patient of the "opening" by an implied agent, the subject wood in

(96) Wood cuts easily without producing crumbling. (Google) (25)

is patient of the "cutting" by an implied agent.

By defining the subjectification mechanism as we have, we also see why a further extension of transitivity types is possible and intransitive and compound-transitive verbs can be used in English middles, as established in Section 1. In principle, the force interaction between antagonist and agonist construed by the letting modality can also apply to other representational relations than those between patient and agent of a transitive verb--provided this relation can be interpreted as one of the subject entity both lending itself and actively contributing to the action described by the predicate. Such extended middle construals may be further motivated by the strong foregrounding of the subject entity on the basis of subjective speaker assessment, as is the case with the middles with intransitive and compound-transitive verbs found in sports and advertisements registers. These middles construe the conduciveness of a location or instrument to an agent's action (e.g., Examples [29], [30], [38], and [39]), or the conduciveness of a primary patient to an agent's compound-transitive action resulting in a secondary patient (as in Examples [24] and [25]).

Comparison with other languages such as French and Dutch shows that this extension of transitivity types in the English middle is also dependent on the nonreflexive realization of the conducive relation.

In French, as in Spanish and Italian, both ergative intransitive and middle clauses are marked by reflexive clitics (Kemmer 1993). In ergative intransitive clauses such as

(97) La porte s'ouvrait.

(98) La verre se cassait.

the reflexive form expresses the coalescence of agentive and patientive participation by the entity in the process (Kemmer 1993). In middle clauses, this relation is subjectified into a modal assessment of "conduciveness", whose active and passive elements are equally marked by the reflexive form. This subjectification process is analogous to the first stage of the subjectification process in English, which involves the use of a transitive predicate in the middle: the semantic role of the subject becomes a straightforward patient and the predicator invokes an inherent agent. We see that in French also, besides all the ergative verbs in their transitive uses, as in Example (99), many ordinary transitive verbs can be used in the middle (Fellbaum and Zribi-Hertz 1989: 28), as illustrated by (100) and (101).

(99) Apres avoir ete graissee, la porte s'ouvrait facilement.

(100) Ce papier se recycle.

(Kemmer 1993: 2)

(101) Le livre se vend bien.

(Kemmer 1993: 147)

However, the reflexive form of the middle in French imposes a selection restriction on the predicates that prevents further extension: lexical verbs in French middles have got to be transitive and cannot be intransitive (see [101] and [102]) or compound-transitive (as in [103]), as shown by the impossibility of these French middles, which correspond roughly to English Examples (29), (8) and (25) respectively.

(102) **Cette piste se court plus rapidement. 'this track runs faster'

(103) **Ces cannes a peche haut de gamme se pechent bien. 'These top-quality fishing rods fish well'

(104) **Cette farine bio se cuit du pain delicieux. 'This wild flour bakes delicious bread'

By contrast, in middle constructions in Dutch, the letting modality is not marked reflexively either, but arises via subjectification of the VP. Therefore, in Dutch middles the predicates need not be strictly transitive but can also be intransitive (105) or compound-transitive (106) (Davidse and Heyvaert 2003: 65):

(105) Ja, ja bomen genoeg, maar sommige bomen zitten nu eenmaal lekkerder dan andere. (INL) yes, yes, trees enough, but some trees just sit more comfortably than others

'Yes, yes, there are enough trees, but some trees are just more comfortable to sit in than others.'

(106) Die cake snijdt mooie sneetjes. 'That cake cuts nice slices.'

Middles with intransitive verbs are currently a more common and extended phenomenon in Dutch than in English and have long been recognized in the literature about Dutch middles as a significant subtype (e.g., De Vries 1910; Holierhoek 1980; Van den Toorn 1982; Hoekstra and Roberts 1993; Ackema and Schoorlemmer 1994; Peeters 1999).

The typological difference of reflexive versus nonovert marking of ergative intransitive and middle has thus led to a different distribution of types of middle predicates in a language such as French in comparison with languages such as Dutch and English, even though the conceptual factors underlying the subjectification process seem in origin to be the same in these languages.

4. Conclusion

In our analysis of the English middle, the theoretical distinction between representational and interpersonal layers of clause organization has proved central. We have put forward the idea that middle construal is defined by categories of interpersonal clause organization (as defined in general by Halliday 1985). We have argued that the relation between nonagentive subject and active VP characteristic of the middle is a modal letting relation. In English, this modality is realized overtly only in some negative VPs with will + not or lexical verbs such as refuse. Mostly the letting meaning of the VP results from subjectification of the VP without overt marking. In terms of Talmy's (2000) force-dynamic approach, middies assess the conduciveness of an antagonist entity (designated by the subject) to execution of the action (denoted by the predicate) by an implied agentive agonist. This conduciveness is due to contextually implied properties of the subject entity. In what way these contribute to the envisaged action is specified by the elements of the predication. The properties of the subject entity may be conducive to the designated action as such, to the ease or result of execution of the action, to a way of acting on or with the entity, or to transfer to a spatial destination of that entity. Pragmatically, the specifications given in the predication thus give further information about and evaluate the properties of the subject entity.

We then proposed that this interpersonal organization of the middle can be interpreted as the result of subjectification in the sense of Traugott (1989) of the ergative intransitive. In the ergative intransitive, the lexical verb and its sole nominal complement express an agentive-patientive form of participation in the process. In middle construals with ergative verbs, this agentive-patientive participant relation subjectifies into the activo-passive relation of the letting modality between subject and finite. We have suggested how, with this understanding of the subjectification mechanism, it can be explained why middles in English, Dutch and French do not only take transitive uses of ergative verbs, but can also take many ordinary transitive verbs. In the existing approaches that recognized the relation between ergative intransitive and middle but did not explicate the subjectification involved, this extension of verb types occurring in the middle could not be explained satisfactorily. By contrasting English and Dutch middles with the reflexively marked French middles, we could also begin to see why the former, unhampered by the transitivity restriction incumbent on the latter, can accommodate compound transitive and intransitive verbs, albeit still marginally in current-day English. However, as it stands, this subjectification hypothesis remains an analytical form of explanation, which future research will have to subject to extensive diachronic verification.

Received 2 July 2002 Revised version received 13 December 2004

University of Leuven


Ackema, Peter and Schoorlemmer, Maaike (1994). The middle construction and the syntaxsemantics interface. Lingua 93, 59-90.

Cruse, Donald A. (1972). A note on English causatives. Linguistic Inquiry 3, 520-528.--(1973). Some thoughts on agentivity. Journal of Linguistics 9, 11-23.

Curme, George O. (1931). A Grammar of the English Language, vol. 3: Syntax. Boston: Heath.

Davidse, Kristin (1991). Categories of experiential grammar. Unpublished doctoral dissertation, University of Leuven.

--(1992). Transitivity/ergativity: the Janus-headed grammar of actions and events. In Advances in Systemic Linguistics." Recent Theory and Practice, M. Davies and L. Ravelli (eds.), 105-135. London: Pinter.

-- and Geyskens, Sara (1998). Have you walked the dog yet? The ergative causativization of intransitives. WORD 49(2), 155-180.

-- and Heyvaert, Liesbet (2003). On the so-called 'middle' construction in English and Dutch. In Empirical Approaches to Contrastive Linguistics and Translation Studies, Sylviane Granger et al. (eds.), 57-73. Amsterdam: Rodopi.

Davies, Eirian (1979). On the Semantics of Syntax: Mood and Condition in English. London: Croom Helm.

--(2001). Propositional attitudes: a semantic model of epistemic mood and modality in English. Functions of Language 8(2), 217-251.

Declerck, Renaat (1991). A Comprehensive Descriptive Grammar of English. Tokyo: Kaitakusha.

-- and Reed, Susan (2002). Conditionals." a Comprehensive Empirical Analysis. Berlin and New York: Mouton de Gruyter.

DeLancey, Scott (1984). Notes on agentivity and causation. Studies in Language 8(2), 181-213.

De Vries, Wobbe (1910). Opmerkingen over Nederlandsche syntaxis: I. Usurpaties. Tijdschrift voor Nederlandsche Taal-en Letterkunde 29, 122-165.

Dik, Simon C. (1980). Studies in Functional Grammar. London: Academic Press.

--(1991). The Theory of Functional Grammar, part 1: The Structure of the Clause. Berlin and New York: Mouton de Gruyter.

--(1997). The Theory of Functional Grammar, part 2: Complex and Derived Constructions, K. Hengeveld (ed.). Berlin and New York: Mouton de Gruyter.

Ehrman, Madeline E. (1966). The Meanings of the Modals in Present-Day American English. The Hague: Mouton.

Evans, Nicholas and Wilkins, David (2000). In the mind's ear: the semantic extensions of perception verbs in Australian languages. Language 76, 546-592.

Fagan, Sarah (1992). The Syntax and Semantics of Middle Constructions. Cambridge: Cambridge University Press.

Fawcett, Robin P. (1980). Cognitive Linguistics and Social Interaction: Towards an Integrated Model of a Systemic Functional Grammar and the Other Components of a Communicating Mind Exeter and Heidelberg: Julius Groos Verlag and Exeter University Press.

Fellbaum, Christiane (1985). Adverbs in agentless actives and passives. In Papers from the Parasession on Causatives and Agentivity at the Twenty-First Regional Meeting of the Chicago Linguistic Society, William H. Eikfont, Paul D. Kroeber, and Karen L. Peterson (eds.), 21-31. Chicago: University of Chicago.

--(1986). On the Middle Construction in English. Bloomington, IN: Indiana University Linguistics Club.

-- and Zribi-Hertz, Anne (1989). The Middle Construction in French and English: a Comparative Study of its Syntax and Semantics. Bloomington, IN: Indiana University Linguistics Club.

Hale, Kenneth and Keyser, Jay (1987). A View from the Middle. Lexicon Project Working Papers 10. Cambridge: MIT Press.

Halliday, M. A. K. (1967). Notes on transitivity and theme in English. Part 1. Journal of Linguistics 3(1), 37-81.

--(1968). Notes on transitivity and theme in English: Part 2. Journal of Linguistics 4(2), 179-215.

--(1970). Functional diversity in language as seen from a consideration of modality and mood in English. Foundations of Language 6(3), 322-361.

--(1985). An Introduction to Functional Grammar. London: Arnold.

Hengeveld, Kees (1989). Layers and operators in Functional Grammar. Journal of Linguistics 25, 127-157.

Heyvaert, Liesbet (1997). Patientive -er nominals. Leuven Contributions in Linguistics and Philology 86(4), 433-456.

--(2003). A Cognitive-Functional Approach to Nominalization in English. Berlin and New York: Mouton de Gruyter.

Hoekstra, Teun and Roberts, Ian (1993). Middle constructions in Dutch and English. In Knowledge and Language, vol. 2: Lexical and Conceptual Structure, Eric Reuland and Werner Abraham (eds.), 183-220. Dordrecht: Kluwer.

Holierhoek, C. M. G. (1980). Werkwoorden van waarneming. Aspekten van hun systematiek en ontwikkeling. The Hague: Leiden University Press.

Horn, Lawrence (1992). Pragmatics, implicature, and presupposition. In International Encyclopedia of Linguistics, William Bright (ed.), 260-266. Oxford: Oxford University Press.

Huddleston, Rodney (1969). Review article of Madeline Ehrman, The Meanings of the Modais in Present-Day American English. Lingua 23, 165-176.

Iwata, Seizi (1999). On the status of an implicit argument in middles. Journal of Linguistics 35, 527-553.

Jespersen, Otto (1914-1929). A Modern English Grammar on Historical Principles, 7 vols. London: George Allen and Unwin.

Kemmer, Suzanne (1993). The Middle Voice. Amsterdam and Philadelphia: John Benjamins.

Keyser, Samuel J. and Roeper, Thomas (1984). On the middle and ergative constructions in English. Linguistic Inquiry 15, 381-416.

Laffut, An (1998). The locative alternation. A contrastive study of Dutch versus English. Languages in Contrast 1, 127-160.

--(2006). Three-Participant Constructions in English. A Functional-Cognitive Approach to Caused Relations. Studies in Language Companion Series 79.

-- and Davidse, Kristin (2002). English locative constructions: an exercise in neo-Firthian description and dialogue with other schools. Functions of Language 9(2), 169-207.

Lakoff, George (1977). Linguistic gestalts. In Papers from the 13th Annual Regional Meeting of the Chicago Linguistics Society, Woodford Beach, Samuel Fox, and Shulamith Philosoph (eds.), 236-287. Chicago: University of Chicago.

Langacker, Ronald W. (1985). Observations and speculations on subjectivity. In Iconicity in Syntax, John Haiman (ed.), 109-150. Amsterdam: John Benjamins.

--(1987). Foundations of Cognitive Grammar: Theoretical Prerequisites. Stanford: Stanford University Press.

--(1991). Foundations of Cognitive Grammar. Descriptive Application. Stanford: Stanford University Press.

--(1999). Losing control: grammaticization, subjectification, and transparency. In Historical Semantics and Cognition, Andreas Blank and Peter Koch (eds.), 147-175. Berhn and New York: Mouton de Gruyter.

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

Levin, Beth (1993). English Verb classes and Alternations. Chicago: University of Chicago Press.

-- and Rappaport Hovav, Malka (1995). Unaccusativity: at the Syntax-Lexical Semantics Interface. Cambridge: MIT Press.

Lyons, John (1969). Introduction to Theoretical Linguistics. Cambridge: Cambridge University Press.

Massam, Diane (1992). Null objects and non-thematic subjects. Journal of Linguistics 28, 115-137.

Martin, James R. (1992). English Text: System and Structure. Amsterdam and Philadelphia: John Benjamins.

McGregor, William B. (1992). The place of circumstantials in Systemic-Functional Grammar. In Advances in Systemic Linguistics. Recent Theory and Practice, Martin Davies and Louise Ravelli (eds.), 136-149. London: Pinter.

--(1997). Semiotic Grammar. Oxford: Clarendon.--(2002). Introduction. Functions of Language 9(2), 125-131.

Nishimura, Yoshiki (1993). Agentivity in cognitive grammar. In Conceptualizations and Mental Processing in Language, Richard Geiger and Brygida Rudzka-Ostyn (eds.), 487530. Berlin and New York: Mouton de Gruyter.

Peeters, Roger J. (1999). The adjunct middle construction in Dutch. Leuven Contributions in Linguistics and Philology 88(3-4), 355-401.

Quirk, Randolph; Greenbaum, Sidney; Leech, Geoffrey; and Svartvik, Jan (1985). A Comprehensive Grammar of the English Language. London: Longman.

Roberts, Ian G. (1987). The Representation of Implicit and Dethematized Subjects. Dordrecht: Foris.

Sinclair, John; Fox, Gwyneth; Moon, Rosamund; and Stock, Penny (1987). Collins COBUILD English Language Dictionary. London: Collins.

Smith, Carlota (1978). Jespersen's 'move and change' class and causative verbs in English. In Linguistic and Literary Studies in Honor of Archibald A. Hill, M. Jazayery et al. (eds.), 101-109. The Hague: Mouton.

Sweet, Henry (1891). A New English Grammar: Logical and Historical. Oxford: Clarendon Press.

Talmy, Leonard (1985). Force dynamics in language and thought. In Papers from the 21st Regional Meeting of the Chicago Linguistic Society, William Eilfort, Paul Kroeber, and Karen Peterson (eds.), 293-337. Chicago: University of Chicago.

--(2000). Toward a Cognitive Semantics, vol. 1: Concept Structuring Systems. Cambridge, MA: MIT Press.

Traugott, Elizabeth Closs (1982). From propositional to textual and expressive meanings: some semantic-pragmatic aspects of grammaticalization. In Perspectives on Historical Linguistics. Current Issues in Linguistic Theory, Winfred Lehmann and Malkiel Yakov (eds.), 245-271. Amsterdam and Philadelphia: Benjamins.

--(1989). On the rise of epistemic meanings in English: an example of subjectification in semantic change. Language 65, 31-55.

--(1995). Subjectification in grammaticalisation. In Subjectivity and Subjectivisation: Linguistic Perspectives, Dieter Stein and Susan Wright (eds.), 31-54. Cambridge: Cambridge University Press.

Van den Toorn, M. C. (1982). Nederlandse Grammatica. Groningen: Wolters Noordhoff.

Van Oosten, Jeanne (1977). Subjects and agenthood in English. In Papers from the 13th Annual Regional Meeting of the Chicago Linguistics Society, Woodford Beach, Samuel Fox, and Shulamith Philosoph (eds.), 451-471. Chicago: University of Chicago.

--(1986). The Nature of Subjects, Topics and Agents: a Cognitive Explanation. Bloomington, IN: Indiana University Linguistics Club.

Verstraete, Jean-Christophe (2001). Subjective and objective modality: interpersonal and ideational functions in the English modal auxiliary system. Journal of Pragmatics 33, 1505-1528.


* We would like to thank the two anonymous referees for their very helpful feedback on the first version of this article, which forced us to make the relation between our descriptive proposals and theoretical assumptions much more explicit. We are also grateful to all the other people who over the years have impacted on our thinking about the English middle in much appreciated written or oral discussion: Hubert Cuyckens, Renaat Declerck, Michael Halliday, Peter Lauwers, Maarten Lemmens, William McGregor, Nele Olivier, Jan Rijkhoff, Stig Johansson, John Taylor, Willy Van Langendonck, Jean-Christophe Verstraete. In arriving at the position assumed in this article, we owe a special debt of gratitude to Ronald Langacker for sharing his views on the middle subject with us and to Roger Peeters for extended discussion, particularly about the question of intransitives in middles. Correspondence address: Kristin Davidse, Department of Linguistics, University of Leuven, Bhjde-Inkomststraat 21, 3000 Leuven, Belgium. E-mail:

(1.) Halliday (1970) put forward the idea that the clause has an interpersonal layer of organization distinct from the ideational, or representational, layer of organization. Hengeveld (1989: 128) imported this model into Dik's functional grammar: at its representational level the clause describes a state of affairs, while at its interpersonal level it relates this situation to the speech exchange.

(2.) Examples followed by (CB) were extracted by remote login from the COBUILD corpus and are reproduced here with the kind permission of HarperCollins. For each COBUILD example, the specific subcorpus in which it occurs is also indicated. The abbreviations used are: oznews for Australian newspapers, times for UK Times newspaper, usephem for US ephemera (leaflets, adverts, etc.), ukephem for UK ephemera (leaflets, adverts, etc.), ukmags for UK Magazines, sunnow for UK Sun newspaper, today for UK Today newspaper. Examples marked by (Google) were found on the internet; their Web addresses are provided in the endnotes.

(3.) Other names used for this type of structure in the literature include: activo-passives (Jespersen 1914-1929, vol. 3: 347), medio-passives (e.g., Declerck 1991: 203), processoriented passives (Halliday 1967),facility-oriented passives (Fawcett 1980: 148), patientsubject constructions (Van Oosten 1977: 459) and pseudo-intransitives (Lyons 1969: 366; Smith 1978: 103).

(4.) The Web site address of Example (6) is:

(5.) The Web site address of this example is:

(6.) The distinction between a representational and interpersonal layer is found also in Langacker's cognitive grammar analysis of the clause. The representational layer corresponds to the conception of a situation as coded by the complementation relations of the verb and modification of the clausal nucleus. Speech event-related elements are, together with additional cognitive mechanisms such as attention and figure/ground organization (Langacker 1991: 138) included in the notion of the CONSTRUAL relationship between the speaker and "the scene so structured" (Langacker 1991: 128).

(7.) Langacker (1987: 300-306) points out a similar change of semantic profile imposed on the nominal by the preposition in a PrepP such as in the park. In designates a spatial relation and the park a thing, but the whole structure in the park designates a spatial relation.

(8.) The Web site addresses of Examples (24) and (25) are and latestnews/2004archive/0904multipack.html respectively.

(9.) The Web site addresses of Examples (27) and (28) are: fopinions.files/bopinions.files/Reviews1115htm.

(10.) The Web site address of Example (30) is:

(11.) As we are concentrating here on the representational layer, we work with nonmiddle alternates to establish process-participant and circumstance relations and we make abstraction of the special modal value of the middle finite, which will be discussed in Section 2.

(12.) We thank one of the anonymous referees for bringing the existence of expressions such as play the pitch to our attention.

(13.) The Web address of (33) is archive/articles/bbchampionship.html.

(14.) As pointed out by Laffut (2006: 23), exactly the same "transitivizing" prefix was found in older stages of English with verbs such as bedaub, besow, bespatter, and is still productive in the formation of adjectives such as bewinged, bewired, etc.

(15.) The Web site address of (42) is:

(16.) Apart from its suitability to capture the modal relation between the inanimate subject and finite in the middle, Talmy's approach to modality fits in naturally with our functional-cognitive approach, as reflected, amongst others, by Langacker's (1991: 273-275) adoption of Talmy's force dynamics as a starting point of his analysis of modal auxiliaries. Moreover, Talmy (2000) also makes a principled distinction between the application of his force-dynamic semantic framework to expressions of causality and event structure and to construals of modality, i.e., the basic representational and interpersonal semantic layers in the clause.

(17.) Traugott (1989: 43) posits, in accordance with Langacker's (1985) gradient view of subjectivity, a cline of gradually increasing subjectification in the development of modals, from main verb to premodal to deontic to epistemic. She questions whether truly objective modality exists (1989: 36).

(18.) At the outset of her work on subjectification, Traugott (1982: 247-248) referred to Halliday and Hasan (1976) for the distinction between ideational, or representational, meaning and interpersonal meaning, which is so basic to the meaning shifts characterized as subjectification by her. In Traugott (1995: 47), she dissociated herself from Halliday and Hasan's classification of some specific linguistic resources as either textual or interpersonal. However, she maintained the general idea that "material for purposes of creating text and indicating attitudes in discourse situations" tend to be recruited from "lexical (prepositional) material".

(19.) This example is extracted from the INL corpus of the Institute of Dutch Lexicology.

(20.) That properties of the subject entity are important to the meaning of the middle has often been pointed out in the literature (e.g., Fellbaum 1985; Fellbaum and Zribi-Hertz 1989; Van Oosten 1986). Our concern here is how these properties interact with the modal meaning of the finite.

(21.) We follow Declerck (1991) in talking about the "actualization", rather than for instance about the "factuality", or the "eventiveness", of the situations grounded by modals.

(22.) If, with Traugott (1989), we view the modal meaning of VPs as emerging via a process of subjectification from VPs depicting verifiable occurrences of SoAs, then agentorientation, subject-orientation, and speaker/hearer-orientation can be viewed as degrees of increasing subjectification.

(23.) The Web site address of Example (74) is:

(24.) This description of the subjectification mechanism presupposes an analysis of the VP such as proposed by Halliday (1985) and Langacker (1991), which distinguishes between finite, secondary auxiliaries and lexical verb (see Section 2.1).

(25.) The Web site address of Example (96) is:
COPYRIGHT 2007 Walter de Gruyter GmbH & Co. KG
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2007 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Davidse, Kristin; Heyvaert, Liesbet
Publication:Linguistics: an interdisciplinary journal of the language sciences
Geographic Code:1USA
Date:Jan 1, 2007
Previous Article:Developing register differentiation: the Latinate-Germanic divide in English *.
Next Article:Dutch collective nouns and conceptual profiling *.

Related Articles
Negative-initial sentences in Old and Middle English.
The Scots -- Northern English continuum of marking noun plurality.
Composite Predicates and idiomatisation in Middle English: a Corpus-based approach (1).
From Domesday Book to Lay Subsidy Rolls: place-names as informants of linguistic change.
Is reanimation of voices possible? Pragmatics of reported speech in selected Middle English texts.
Rethinking the coordinate-subordinate dichotomy; interpersonal grammar and the analysis of adverbial clauses in English.
Beiersdorf launches nivea global brand campaign in the middle east.
Beiersdorf launches nivea global brand campaign in the middle east.
A Companion to the Middle English Lyric.
Beiersdorf in beauty expo tieup.

Terms of use | Privacy policy | Copyright © 2019 Farlex, Inc. | Feedback | For webmasters