Publishing Partner: Cambridge University Press CUP Extra Publisher Login
amazon logo
More Info

New from Oxford University Press!


Linguistic Diversity and Social Justice

By Ingrid Piller

Linguistic Diversity and Social Justice "prompts thinking about linguistic disadvantage as a form of structural disadvantage that needs to be recognized and taken seriously."

New from Cambridge University Press!


Language Evolution: The Windows Approach

By Rudolf Botha

Language Evolution: The Windows Approach addresses the question: "How can we unravel the evolution of language, given that there is no direct evidence about it?"

The LINGUIST List is dedicated to providing information on language and language analysis, and to providing the discipline of linguistics with the infrastructure necessary to function in the digital world. LINGUIST is a free resource, run by linguistics students and faculty, and supported primarily by your donations. Please support LINGUIST List during the 2016 Fund Drive.

Review of  Middle Voice

Reviewer: Andrew McIntyre
Book Title: Middle Voice
Book Author: Markus Steinbach
Publisher: John Benjamins
Linguistic Field(s): General Linguistics
Subject Language(s): German
Language Family(ies): Germanic
Issue Number: 14.260

Discuss this Review
Help on Posting

Date: Wed, 15 Jan 2003 18:18:31 +0100
From: Andrew McIntyre
Subject: Review: Syntax/Semantics: Steinbach (2002), Middle Voice

Steinbach, Markus (2002). Middle Voice: A comparative study in
the syntax-semantics interface in German. Amsterdam: John Benjamins.

Andrew McIntyre, University of Leipzig

Imagine a language designed with the express purpose of rotting the self
esteem of linguists. In such a language constructions like (1) and (2)
would arguably have a rightful place:
(1) Das Buch liest sich gut
the book reads itself well
'The book reads well'
(2) Es pennt sich gut im Büro
it sleeps itself well in.the office
'The office is a good place to sleep (in)'
Among the properties of such German constructions -middle constructions-
are the following. In (1) the normal objects of 'lesen/read' become
subjects. In German, but not in English, a reflexive pronoun appears. In
(2), an intransitive verb trades its usual human subject for an
expletive subject and reflexive object. (1) and (2) show the notorious
'adverbial effect': they are well-nigh unacceptable without the
modifiers 'gut/well' and 'im Büro'. We are left wondering how to handle
the oxymoronic notion 'obligatory modifier'. Much ink has been spilled
in attempting to explain these phenenoma, and the extant accounts -it
seems to me- fail to satisfy.
Markus Steinbach's book makes what I see as a worthwhile
contribution to the study of middle constructions and German reflexive
constructions in general. I now summarise the descriptive points in
chapter 1. Alongside middles like (1-2), he also analyses the following
reflexive constructions:

(3) a. Fritz rasierte sich [reflexive reading]
Fritz shaved
b. Gabi kritisierte sich
Gabi criticised herself
(4) Die Tür öffnete sich [anticausative reading]
the door opened itself
'the door opened'
(5) Karl schämt sich [inherent reflexive reading]
Karl shames himself
' Karl is ashamed'

The genuine reflexive constructions in (3) are the also termed 'argument
reflexives'; the reflexive is a nomal verbal direct object, witness its
ability to be coordinated with normal object NP's: 'Gabi criticised
herself and Dave'. In anticausatives like (4), the reflexive
construction alternates with a transitive one ('I opened the door').
English lacks this type of reflexive. In 'the door opened itself', the
door opens by itself, not through another agent. This intuition is
absent in (4). Anticausative reflexives thus look like valency reduction
markers. Inherent reflexives like (5) (cf. English 'perjure/ disgrace/
betake/ vaunt oneself') look like semantically vacuous arguments
licensed by some stipulated c-selection requirement. Here the (relevant
sense of) the verb is obligatorily reflexive. Steinbach calls all the above
reflexives apart from (3) 'non-argument reflexives' because the
reflexive does not obviously map onto a semantic argument of the verb.
The constructions noted above are a crosslinguistic natural class, being
distinguished by the same morphological markers in language after
language. The term 'middle voice' is used for this natural class (the
'voice' part is needed to distinguish middle voice in general from the
narrower use referring to constructions like (1) and (2). A reflexive
pronoun is the typical middle voice marker in Indo European languages.
Chapter 2 gives more empirical information about middle voice. Steinbach
discusses the verb classes acting as input to middle in German. We find
middles with transitives, unergatives, unaccusatives, ditransitives and
resultative constructions, but not with weather verbs. Middle acts on
verbs with at least one semantic argument. The subject of a middle is
either expletive 'es', cf. (2), or corresponds to the normal accusative
object of the verb. It is impossible to promote dative objects in German
middles. Steinbach reasons that promotion in middles is sensitive to case, not
thematic roles. This seems to be at variance with Steinbach's observation that
German marginally allows adjunct middles like (6a,b). (I found these
under; S:296 attests another real example.) Here the
subject corresponds to the NP complement in an adjunct PP; cf. the
near-synonymous impersonal middles in (6c).
(6) a. Die Schuhe laufen sich hervorragend
The shoes walk themselves excellently
'These shoes make for excellent walking'
b. Die weiße Wolle strickt sich viel leichter als die blaue Wolle
the white wool knits itself much easier than the blue wool
'Knitting is easier with the white than the blue wool'
c. Es läuft/strickt sich hervorragend {in diesen Schuhen/mit dieser
It walks/knits itself excellently {in these shoes/with this wool}
After giving more details on anticausative and inherent reflexives, Steinbach
(46) turns to the typology of middle voice constructions. Crucial is the
existence of two-form languages (Kemmer 1993), where we find strong and
weak reflexive forms (Russian 'sebja' vs. '-sja', Dutch 'zichself' vs.
'zich'). The weak reflexive is formally more grammaticalised and is
further to the right on the cline 'word-clitic-affix-zero'. It is
confined to special verb classes, e.g. verbs of grooming and posture
change, quintessential 'reflexive' processes. Middle markers must be
weak forms. English is a two form language: its weak form is zero. Thus,
the English glosses of the German reflexives in (1) and (3a) (but not
(3b)) are intransitive rather than reflexive. Steinbach's claim is that German
lacks a strong reflexive. (Note for Germanists: S:156ff argues that
'sich selbst' (=emphatic 'oneself') is not gramaticalised as a strong
reflexive, but is a regular use of the focus particle 'selbst', which
can focus any kind of NP ('Maria selbst'). I add that Steinbach's claim is also
supported by the synonymous focus particle 'selber'. Disagreeing with
Steinbach would force one to posit two German strong reflexives, 'sich selbst'
and 'sich selber', straining credulity given that both can focus any type of
Chapter 3 is a detailed literature survey. Syntactic analyses
(actually: existing syntactic analyses) are argued to be problematic
(These include English-oriented analyses emaphasising parallels with
passive, analyses using a pro subject, and analyses assuming the
reflexive to bind the subject argument.) Lexicalist analyses fare no
better, since they merely stipulate facts like obligatory adverbials,
reflexives etc., rather than explaining them. This moves Steinbach to
adumbrate a 'postsyntactic' theory where middles are derived 'at the
syntax-semantics interface'.
Chapter 4 proposes that argument and non-argument reflexives have an
identical syntax: both are direct objects. Steinbach assumes an early
Minimalist non-Kaynean syntax for German in which direct objects are
case-licensed in an agreement projection just above a head-final VP. (Steinbach
(204) cites Chomsky 1995 without page numbers in defense of the agreement
projections; this is unfortunate given Chomsky's abandonment of them in
chapter 4.) The claim that non-argument reflexives are syntactically
identical to argument reflexives forces Steinbach to show why the former
differ from the latter in being unable to be coordinated, focussed (using
stress or focus particles) or fronted. I think Steinbach succeeds in this.
Chapter 5 is about binding theory. Steinbach notes empirical problems with
Chomsky's (1981) binding theory. Steinbach's binding theory modifies Reinhart
& Reuland (1993), notably in order to capture non-argument reflexives,
ignored by other binding theorists. It is hard to do justice to the
complexities of the theory in a short space. Here is my attempt. Steinbach
stipulates a General Condition on A-Chains (GCC) which says that an
argument chain must (a) be headed by a case-marked link with the feature
[+R], and (b) contain exactly one [+R] element. ([+R] roughly stands for
'referentially independent', but see below.) Unlike other theories where
reflexives are specified as [-R], Steinbach leaves German reflexives unspecified
w.r.t. [R]. The [-R] specification gives the nonargument reading, while
[+R] gives the argument reading. With [-R], the GCC forces the reflexive
to be construed as part of an argument chain, so that it ends up
mediating the linking between the subject and its base position
(=complement of V). With argument reflexives, there are two distinct
argument chains. That they are coindexed is due to the semantic part of
the binding theory, which uses an adaption of Pollard & Sag's (1994)
Principle A: a [+R] must be bound by a less oblique argument of the same
predicate. Steinbach argues that oblicity should be defined in terms of case
rather than in terms of thematic roles as in Pollard & Sag.
I pause to consider Steinbach's choice to characterise argument reflexives as
[+R]. This may seem an odd use of the [R] feature, which normally stands
for 'referentially independent'. Note 10, p. 219 comments that Reinhart
& Reuland use of [R] as a purely morphosyntactic feature not directly
associated with reference. If Steinbach understands [R] in this sense (contrary
to what the equation of [-R] with 'referentially deficient' on p. 183
suggests but in conformity with the possible specification of reflexives
as [-R]), then we would want some kind of independent semantic
characterisation of what the essence of [R] actually is. Without this,
[R] be seen as a mere trick serving merely to reformulate a descriptive
problem more elegantly. Perhaps future work can rectify this problem.
Chapter 6 is really two chapters. 6.1 explores the differences between
middles, anticausatives and unaccusatives. Two things can happen to
argument variables:
-Saturation (=binding by some semantic operator)
-Reduction (=removal from the semantic representation)
The middle reading arises from saturation by a generic operator (see on
ch. 7). Reduction is responsible for anticausative and inherent
reflexive structures. There is a discussion of the factors licensing
anticausative formation. Steinbach (231) assumes that it is licensed when the verb
does not require its subject to have any particular mental state and
when the event can be perceived as happening by itself; there is
strangely no mention of the 'internal causation' of Levin & Rappoport
(1995), although that book is cited. See Härtl (2003) for more on the
German data. There ensues a brief and inconclusive discussion of whether
unaccusativity is syntactically encoded. No attempt is made to determine
when a change of state verb will be reflexive or non-reflexive in
German. See Labelle (1992) on the parallel French problem.
It is hard to assess whether Steinbach is right in applying reduction (cf.
section 6.1 above) to inherent reflexives. S:232 asserts without
evidence that they 'seem to be derived from an underlying two-place
representation'. Steinbach (233) leaves open whether the inherent reflexive
follows (a) from the verb meaning or (b) lexical stipulation. Let us
take this further. If there are verbs where (b) holds, then I see two
possibilities. One is that the reflexive is synchronically just a dummy
argument idiosyncratically c-selected by the verb. If so, then reduction
cannot apply. The second possibility is that what is stipulated is not
the reflexive itself, but that reduction has to apply. This should not
shock us given that similar stipulations are needed for obligatorily
passive verbs

(7) Mervyn got reincarnated (but *'God reincarnated Mervyn'),
He is rumoured to be an alien (but *'Bild-Zeitung rumored him to be an

and even obligatory middles:

(8) a. She scrubs up well/nicely 'She dresses up effectively, looks nice
when dressed up'
b. *She scrubs up (* unless taken in sense of (a), as can be brought out
with an emphatic 'really')
c. *She scrubbed (herself, her daughter) up for the party
Thus, at least in my variety, 'scrub up' has a sense 'dress up' which is
licensed only in the middle construction, although I see no earthly
reason why this should be so. Thus, we seem to have independent evidence
that for the stipulation of obligatory operations of the type envisioned
by Steinbach. On the other hand, it seems hard to believe that every dummy
object is due to an obligatory reduction or saturation operation,
witness cases like 'live it up, leg it, wing it' listed in Postal &
Pullum (1988).
Section 6.2 discusses the status of dative in German. The theory
predicts that nonargument reflexives should have structural case, so Steinbach
must give independent arguments for the non-structural status of German
datives. Some of the material is based on Vogel & Steinbach (1998), who
analyse dative NP's as adjuncts. Even though I argue against structural
dative in German in work in preparation, I should alert non-Germanist
readers to some problems. Steinbach notes that nominative and accusative
determiners are formally identical outside the masculine singular. This
is used in arguing that dative (unlike accusative) needs special
morphological marking. Steinbach does not mention that many German dialects
syncretise accusative and dative yielding the "Akkudativ". Some other
arguments in 6.2 indeed show that nominative and accusative pattern as a
natural class, but do not show that dative is not structural. For
instance, Steinbach notes that the first item in a synthetic compound can
correspond to an accusative but not a dative, but the same holds for
English ('gift-giving' but *'children-giving' (on beneficiary reading),
where there is no morphological dative and where the indirect object is
structural ('the children were given gifts'). I advise caution with the
multiple datives in (43) on page 253, since some informants I consulted
dislike them. This need not refute Steinbach's datives-qua-adjuncts-position,
since processing difficulties may affect an otherwise grammatical
Chapter 7 deals with three topics specific to middles. 7.1 is about
genericity. It is well known that middles 'lack specific time reference'
(Levin 1993), being generic events, witness the simple present in
English cases like (1a). Remember that section 6.1 suggests that middle
involves not reduction (argument deletion) but saturation (binding by an
operator). Steinbach assumes that the generic operator that binds the event
variable also binds the variable of the unexpressed argument. In an
illuminating discussion, Steinbach derives the following intuitions/facts, among
-that middles attribute a property to the promoted subject and hold it
responsible for the situation,
-that middle constructions are somehow 'modal'
-that individual-level predicates cannot input to middles
-that the unrealised argument receives an arbitrary interpretation.
Steinbach also gives a fairly satisfying explanation for the 'adverbial effect'
(the tendency whereby middles are often bad without adverbials: 'the
book reads *(well)'), which relies on the interaction between
genericity, information structure and general informativeness
considerations. (Goldberg & Ackerman 2001 independently reach similar
conclusions.). The theory extends to cases where adverbials are not
needed, like the following example I heard:
(9) The thread won't pick up. It's part of the carpet.
Section 7.3 is a likewise illuminating discussion of adjunct middles
like (6a,b).
Chapter 8 summarises the book and notes some conclusions and open

My overall evaluation is positive. The book reads easily. (An exception
is the use of endnotes rather than footnotes. I implore all publishers
to eliminate this annoying and unnecessary nuisance to the reader.) The
book is necessary reading for everyone interested in reflexivity, middle
voice, middle constructions and German grammar. The data take Steinbach
into many diverse issues (information structure, case, binding,
genericity...). It is perhaps utopic to expect a more adequate coverage
of the literature and data by one author in a book of the same length
(337 pages with references and index). Inevitably, some relevant
literature will be missed (e.g. Pesetzsky 1995 and Bouchard 1995 on
Romance reflexives and binding) and some topics will be given too short
a shrift. On the latter score, I should note the absence of a serious
theory of argument realisation (beyond reflexives). We read (p. 205)
that German has a 'very simple' linking system where spec,VP is for the
first argument of the verb and complement of V is its second argument.
This cannot accommodate three-place verbs. The theory is only 'very
simple' because it lacks empirical content: there is no semantic
definition of 'first' and 'second' argument, nor are we told why the
respective arguments should be linked as they are. This can be read as
an invitation to combine one's favourite linking theory with Steinbach's ideas
on reflexives, but not every linguist is at liberty to do so. For
instance, many theories assuming some variant of Uniformity of Theta
Assignment, such as the generative linguists following Hale & Keyser
(1993) in assuming that agents/causers enter the syntax as specifiers of
performative/causative light verbs, will have to jettison either their
own theory or Steinbach's idea that the non-reflexive NP occupies the position
normally reseved for the subject.
To the criticisms I made while summarising the book, I now add couple of
other minor points.
Steinbach makes no recommendations concerning the use of middle as a test; it
has been variously claimed to diagnose affectedness (Hale & Keyser
1993:82 and Levin 1993:26, Jackendoff 1996:312, fn. 7), unaccusativity
(Hale & Keyser ibid.) and the status of the direct object in a resultive
construction as argument of the verb (Carrier & Randall 1992:188ff).
Linguists using middles to diagnose something are bedevilled by the
delicacy of the judgements. Hence, Steinbach gives inconsistent judgements
for the same sentence ((20a) p.29 vs. (16b) p. 84), fortunately without
detriment to his argumentation. English middles are also fraught with
uncertainty. Should we reject (10a) but countenance (10b), as do Carrier
& Randall (1992: 190)? How should we judge the other the sentences in
(10)? To me (10d) sounds better than (e), and the German translation of
(d) is better than the English, but I would not bet my house (or my
theory) on this.
(10) a. Competition Nikes run threadbare easily
b. Teary-eyed witnesses believe easily
c. The mountain climbs easy
d. The wine drinks easily and well [attested once in internet]
e. Only small hamburgers eat easily
There being no clear answer to these questions, it seems that middles
should be left well alone as diagnostics for anything. Either there are
too many as yet unknown pragmatic factors affecting their
unacceptability, or perhaps the Enlish and German middle are not fully
productive. (This possibility is neither assumed nor denied by Steinbach.)
Some readers may lament Steinbach's failure to mention the reflexive found
in resultative constructions ('er trank sich zu Tode/he drank himself to
death'), but this is not a deficit. Coordination tells us that it is a
special case of the argument reflexive ('he shouted himself and the
audience into a frenzy'). See Levin & Rappaport (1995, 2001) and
McIntyre (2001) for different accounts.
In supporting his binding theory, Steinbach (200) is forced to argue that
exceptionally case-marked (ECM) subjects are semantic arguments of
matrix verbs as well as being case-marked by them. Thus 'John' is a
semantic argument of 'hear' in 'I heard John sing'. Steinbach's reasoning
is that 'hear John sing' entails hearing John, but 'hear that John is
singing' does not entail hearing him. I query this. The entailments
follow from the fact that the perception of an event normally involves
the perception of the participants. I say 'normally' advisedly, because
sometimes it is possible to perceive a situation without perceiving the
referrent of ECM NP. Various examples of this type disconfirm the idea
that ECM subjects are semantic arguments of the perception verbs:
(12) Ich sah den unsichtbaren Mann eintreten, weil er gerade eine
Zigarre rauchte.
'I saw the invisible man enter because he was smoking a cigar.'
(11) I heard Fred being shouted at [I need not hear Fred]
I saw the wind blow her hat off.
Suddenly, I felt the fever leave me [reports if anything cessation of
It is purely for expository reasons that I have gone into more detail
criticising the book than praising it. I do not think that any of the
points I have criticised devalues the contribution of the book as a

-Bouchard, D. 1995. The Semantics of Syntax. Chicago: University of
Chicago Press.
-Carrier, J., & Randall, J. (1992). The argument structure and syntactic
structure of resultatives. Linguistic Inquiry, 23, 173-234.
-Chomsky, N., 1981. Lectures on Government and Binding. Cambridge
(Mass.): MIT Press.
-Chomsky, Noam, 1995 The Minimalist Program. Cambridge, MA: MIT Press.
-Goldberg, A. & Ackerman, F. 2001. The Pragmatics of Obligatory
Adjuncts. Language. 77 4. 798-814.
-Hale, K., & Keyser, S., 1993. On Argument Structure and the Lexical
Expression of Syntactic Relations. In: Hale, K. et al. (eds). The View
from Building 20. 53-110. Cambridge: MIT Press.
-Härtl, H. 2003. Conceptual and grammatical characteristics of argument
alternations. Linguistics 41, 5.
-Jackendoff, R., 1996. The proper treatment of measuring out... Natural
Language and Linguistic Theory 14: 305-54.
-Kemmer, S. (1993) The middle voice. Amsterdam: Benjamins.
-Labelle, M. (1992) "Change of State and Valency'', Journal of
Linguistics 28, 375-414.
-Levin, B. (1993). English verb classes and alternations. University of
Chicago Press.
-Levin, B., & Rappaport Hovav, M. (1995). Unaccusativity. MIT Press.
-McIntyre, A. 2002. Event paths, conflation, argument structure and VP
shells. Under revision for Linguistics. Ms. Leipzig
-Pesetsky, D. 1995. Zero syntax: Experiencers and cascades. Cambridge,
MA: MIT Press.
-Pollard, C. & Sag, I. (1994) Head-driven phrase structure grammar.
Chicago: University of Chicago Press.
-Postal, Paul M., and Geoffrey K. Pullum. 1988. "Expletive noun phrases
in subcategorized positions." Linguistic Inquiry 19:635--670.
-Rappaport Hovav, M. and B. Levin (2001) "An Event Structure Account of
English Resultatives'', Language 77, 766-797.
-Reinhart, T. & Reuland (1993) Reflexivity. Linguistic Inquiry 24,
-Vogel, R. & Steinbach, M. (1998). The Dative - An Oblique Case.
Linguistische Berichte 173: 65-90.

Andrew McIntyre ( has a postdoctoral research and teaching position at the University of Leipzig, Germany. He is mainly interested in the syntax-semantics mapping in the VP, e.g. argument linking, complex verb formation, possessive constructions, syntactic lexical decomposition. He works mainly on German and English.Œ