1 Contextual Licensing of Marked OVS Word Order in German

1
Contextual Licensing of Marked OVS Word Order in German
Thomas Weskott,
Robin Hörnig,
Gisbert Fanselow,
and
Reinhold Kliegl
University of Potsdam, Germany, SFB 632 ”Information Structure”
Address correspondence to:
Thomas Weskott
Department Linguistik
Universität Potsdam
Karl-Liebknecht-Str. 24-25
14476 Golm
Email: [email protected]
2
Abstract
Numerous studies have shown markedness effects of German OVS word order, that is:
decrease in acceptability and higher reading times. Some of these studies have shown these
effects can be ameliorated by the presentation of context. These studies found that in certain
contexts the acceptability and processing disadvantage of German marked word orders can be
leveled out. We report on acceptability ratings and a self-paced reading study which
addressed the question whether there are contexts in which German marked OVS word order
is not only judged as being equally acceptable and as easy to process as its unmarked SVO
counterpart, but where OVS structures are strongly licensed in the sense that they are actually
judged as more acceptable and are easier to process than the unmarked SVO order. Our
results show that whole-part contexts are strongly licensing in this sense, and that, an
adversative relation between the marked word order sentence and the context does not play an
indispensable role in the contextual licensing effect.
3
1. Background
Word order is one of the most well-studied aspects of German grammar. Both the possibility
of topicalisation, i.e. the movement of elements to the so-called Vorfeld/prefield-position, as
well as Scrambling, the possibility of a re-arranging elements in the syntactic domain of the
Mittelfeld/middle field, have attracted much attention from researchers interested in the
interplay between syntax and information structure (see Lenerz 1976, Hoberg 1981, Höhle
1982, Uszkoreit 1986, Haider and Rosengren, 1998/2003, Müller, 1999, Fanselow 2001, and
Frey, 2004, to name only a few). Common to most of these approaches is the assumption that
an unmarked order exists, and that the alternative order(s) show, compared to the unmarked
order, some or all of the empirical correlates of markedness. These correlates of the marked
word order, compared to the unmarked one, are (a) degraded acceptability (as e.g.measured in
controlled rating experiments); (b) processing difficulty (as measured in self-paced reading,
EEG, or eye-tracking studies); (c) lower frequency in corpora; and (d) later age of acquisition.
All of these empirical correlates seem to point into the direction of a general dispreference for
the marked form compared to the unmarked one: it is less acceptable, harder to process, less
frequent, etcetera. Moreover, most syntactic approaches to word order in German assume that
the choice for a marked variant, or the syntactic operation deriving the marked variant from
the underlying unmarked order, is optional – under certain circumstances, the marked form
might be chosen, but it does not have to be. Taken together, these two uncontroversial
assumptions—dispreference and optionality of marked word orders—raise an interesting
question: why would the grammar of German provide a serialisation option which apparently
is dispreferred? Why should the processing mechanism of Germans be confronted with the
possibility of having to process a serial input that counters its preferences?
Most probably, there is no uniform answer to this question. Word order can be shown to be
influenced by relative processing ease (as a function of constituent length, see Hawkins 1995)
4
and this factor can be shown to be at least independent of the impacts of givenness-newness
(Arnold et al 2000) on the choice of marked word order. The latter aspect leads us to the
general idea that the marked orders can reflect a certain information structural partition of the
sentence they are part of, and that this information structural partition indicates the way how
the sentence should be related to its preceding context. This has been particularly argued for
in the cartographic approach introduced by Rizzi (1997), according to which the functional
syntactic structure of a sentence directly reflects its partitioning into informational structural
units such as Topic and Focus, but also in models which hesitate to express this basic claim in
terms of functional categories (see., e.g., Neeleman & de Koot 2008).
We will borrow from Krifka (2008) the idea that the information structure of an utterance is
shaped by the way the sentence relates to the common ground established by the interlocutors
at the time of the utterance. Thus, we may in general hypothesize that a marked form
frequently serves the function of indicating a certain state of the common ground (leaving the
exact nature of that state open for the time being). This is in accordance with Höhle’s (1982)
observation that markedness of word order correlates with contextual restrictions: while the
unmarked form may appear in the maximal set of contexts, sentences (or utterances) with a
marked word order tend to felicitously appear in a comparatively restricted set of contexts.
But wherein does this pragmatic function consist? What, for example, are pragmatic functions
of the marked object-verb-subject (OVS) order that we are concerned with here in a German
sentence with a transitive verb? It is a remarkable fact that it is the left periphery of languages
such as Italian or Hungarian, rather than German, for which the idea of a direct encoding of
the pragmatic notions topic and focus has been developed, and it is also noteworthy that the
focus-background and topic-comment articulation in the middle field (rather than the prefield)
figured as the empirical basis of the theory of Neeleman & de Koot (2008) developed for
Dutch. In other words, that notions such as topic or focus should matter in the analysis of the
German prefield does not seem to suggest itself. Indeed, the proposals concerned with the
5
movement to the German prefield position rarely refer to these concepts. Weskott (2003)
proposes that left peripheral direct objects are presuppositional, while Frey (2006) argues for
a necessarily contrastive interpretation (Frey, 2006), a view later extended to the notion of
'salience' (Frey (2010). Fanselow & Lenertová (2010) identify cases of left peripheral objects
without an information structural role (in a narrower sense). Certain instances of the
construction they focus on may be due to the fact that the preposing of an object facilitates a
topic shift in the following text (Hörnig et al., ms.)
Speyer's (2007) corpus analysis suggests that topics (or, rather, the centers in the sense of
Grosz et al., 1995) do appear in the German prefield, but when they compete with elements
bearing a Poset-relation (Prince 1998, Birner & Ward, 1998) to preceding discourse, they
usually lose. Poset relations are transitive, reflexive and antisymmetric, for example "part-of".
Only frame-setting elements are better candidates for occupying the prefield than expressions
bearing a poset-relation to an argument in the preceding sentence.
In the present study, we will investigate whether the corpus effect described by Speyer (2007)
is mirrored in acceptability data and processing ease. Our experiment builds on previous
insights coming from the comprehension of spatial relational assertions (see Hörnig,
Oberauer, & Weidenfeld, 2006; Hörnig, Weskott, Kliegl, & Fanselow, 2006), namely
contexts establishing a Poset relation between the referent of the object of the critical
sentence, and a referent from the preceding context.
2. Experimental studies on word order variation in German
Psycholinguistic studies on German word order variation have mostly concentrated on the
apparent preference of orders in which the subject precedes the object, as compared to the
reverse orders where the object precedes the subject, as indicated for example in shorter
reading times for sentences with unmarked word order. This processing advantage for
unmarked word order has been first uncovered by Krems (1984) and labelled “subject first
6
preference” by Hemforth (1993). It has been empirically supported by numerous experimental
studies with different methodologies (see Gorrell, 2000, for an overview; but also see
Burckhardt, Fanselow & Schlesewsky 2007 for the absence of a subject first preference in
certain contexts). These findings are in accordance with the intuitive assessment of the
subject-first order as the “more natural” or “unmarked” one—naïve speakers of German, if
presented with, for example, an OVS structure out of context, sometimes even go so far as to
claim it to be “ungrammatical”. At the same time, however, the notion of markedness raises
the question why a structure which is dispreferred should be provided by the grammar of
German in the first place. As mentioned in the previous section already, one possible answer
to this question alludes to the connection between the notion of markedness and the notion of
linguistic context: it is a well-established fact in many linguistic domains (e.g., phonology,
morphology, lexical semantics) that marked forms have a more restricted context of
appearance (hence their lower frequency), but that they do not seem marked at all when
appearing in a context fulfilling the restriction the marked form imposes on the context, a
phenomenon dubbed “markedness reversal” by Battistella (1996). For example, the
dimensional adjectives ‘long’ and ‘short’ show a markedness asymmetry: ‘long’ is the
unmarked member of that pair, cf. ‘How long is the paper?’ vs. ‘How short is the paper?’. In
the (restricted) context of writing a short article and having to shorten it to a maximum
number of pages or words, however, the marked question ‘How short is the paper?’ loses its
air of markedness. In these cases, the context may be said to license the marked form.
The notion of contextual licensing has also been applied to the markedness asymmetry
between German subject-object word order variants (see Höhle, 1982, for an early proposal).
The leading idea behind psycholinguistic studies investigating marked word orders of German
in context was that there must be contexts in which the marked structures would not only
“lose their air of markedness” (and hence would be judged as equally acceptable, for example,
in a rating study), but would also be as easy to process as the unmarked structure. We will call
7
this annihilation of the markedness difference in a licensing context weak licensing. A natural
follow-up question is whether there are contexts in which the marked form is judged as more
acceptable and/or is easier to process than its unmarked counterpart. Such contexts, if they
exist, would exert what we will refer to as strong licensing.
To the best of our knowledge, so far there is no experimental evidence for strong
contextual licensing of German marked word order, with the exception of Bornkessel &
Schlesewsky (2006). A number of acceptability rating studies report weak licensing effects
for marked structures. These include Bader (1999) who found weak licensing of OSV
structures with ambiguous and unambiguous full NPs in question contexts which asked for
the subject (see Keller, 2000, Exp. 10, for a similar finding); Weskott, Stolterfoht, Bornkessel,
and Schlesewsky (2004), who found weak licensing for both OSV and OVS structures with
unambiguous full DPs in question contexts, and Fanselow, Lenertová & Weskott (2008) for
object fronting in wide and narrow focus contexts. Weak licensing has also been found in a
number of studies using on-line measures such as self-paced reading times, among them
Marslen-Wilson & Bayer (1992), for unambiguous and ambiguous OVS structures in
discourse contexts (see Weskott, 2003, for related findings), and in a reading-cum-judgment
ERP study by Bornkessel & Schlesewsky (2006) on unambiguous OSV structures with full
DPs. The latter study reports a strong licensing effect for OSV structures, that is, that they are
judged as more acceptable than their SOV counterparts. However, given that in this study the
target sentences were preceded by structurally parallel ones (i.e., an OSV target sentence was
preceded by another OSV sentence), it is not entirely clear to us whether this effect is
independent of the effect of structural parallelism of the target sentence with the sentence
preceding it.
With the aforementioned exception, none of these studies, however, reported strong
licensing. We attribute this to two properties of these studies: to the employment of the wrong
kind of contexts, which may simply have lacked the restricting force to strongly license
8
OSV/OVS, and to the employment of the wrong kind of structures in the critical sentences,
which always used full DPs for both arguments (with the exception of Keller (2000), who
used object pronouns), which runs counter the fact that in corpora, OVS structures are quite
frequent with pronominalized subjects (see Weber and Müller, 2004). Moreover, the use of
two referential NPs makes the experimental sentences susceptible to the possibly damaging
effect of similarity-based interference (see e.g. Gordon, Hendrick, Johnson, & Lee, 2006).
We set out to remedy the first problem with a context type which we hypothesized to be
strongly licensing on the basis of the corpus study of Speyer (2007) and our own previous
findings coming from the comprehension of spatial relational assertions (see Hörnig,
Oberauer, & Weidenfeld, 2005; Hörnig, Weskott, Kliegl, & Fanselow, 2006), namely
contexts establishing a Poset relation between the objects of the critical sentences (see Prince,
1998, and Birner & Ward, 1998, for a detailed account of Poset licensing of marked word
order in English). Poset relations like whole-part relations are transitive, irreflexive and
antisymmetric. In (1), a whole-part relation between two referents [the car in (a) and the side
mirror in (b)] is given as an example of an instance of a poset relation:
(1) a.
Peter hat
den
Wagen
Peter has
theACC car
gewaschen.
washed.
‘Peter has washed the car’
b.
Er
hat
He NOM has
b´.
Den
den
Außenspiegel ausgelassen.
theACC side mirror
Außenspiegel hat
TheACC side mirror
has
‘The side mirror, he left out.’
er
left-out.
ausgelassen.
heNOM left-out.
9
The second problematic aspect of previous studies was solved by employing
pronominalized subjects instead of structures realising both arguments as full DPs. Possibly,
this move is not innocuous, as will be discussed below.
Note that, in addition to the whole-part relation between the referents of ‘the car’ and ‘the
side mirror’ in (1.a and .b/b´), there is also an adversative relation between the propositions
expressed by the sentences: for the car it holds that it was washed by Peter, while the same
does not hold for the side mirror. In order to test whether the adversative relation between a
referent in the preceding context and the referent of the direct object of the critical sentence
had an influence on the hypothesized licensing effect, we also tested sentences/sentence pairs
as the following:1
(2) a.
Peter hat
den
Peter has
Wagen
gewaschen.
theACC car
washed.
‘Peter has washed the car’
b.
Er
hat
HeNOM has
b´.
Den
den
Außenspiegel besonders
theACC side mirror
Außenspiegel hat
TheACC side mirror
has
er
gründlich
gewienert
particularly
diligently
polished.
besonders
gründlich
gewienert
diligently
polished.
heNOM particularly
‘The side mirror, he polished with particular diligence.’
In the two experiments to be described below, we asked participants to judge (Experiment 1)
and read (Experiment 2) SVO sentences like (1.b) and (2.b) and compared their judgments
and reading times to those of sentences of the OVS type exemplified in (1.b’) and (2.b’). One
group saw the sentences in isolation (null context group), while the other saw sentence pairs
like (1.a, b) (the whole-part context group). Furthermore, we tested whether the adversative
relation would have an influence on the observed effects; thus, we tested one group with the
10
adversative sentences/sentence pairs exemplified in (1) and compared the results to those of a
group which we tested with the nonadversative ones in (2). Our general prediction was that
the subject-first preference should be present in the null context group (better judgments and
shorter reading times for SVO than for OVS), while, if the whole-part context is strongly
licensing, the reverse pattern should be found in the whole-part context group: here, OVS
should be judged as more acceptable and be read faster than SVO. Statistically this means we
predicted an interaction of the factors context and word order. Concerning the adversativity of
the whole-part relation, we hypothesized that the contextual licensing effect should be
stronger in the adversative context-sentence pairs than in the nonadversative ones. In other
words, while we predicted strong licensing of adversative contexts, we expected weak
licensing of nonadversative contexts.
3. Experiment 1
In order to first assess the acceptability of German SVO and OVS structures with and
without context and with and without adversativity, we conducted a paper-and-pencil
acceptability rating study with four groups of participants (each N=36): one being tested on
the structures in the null context on the contrastive items, one in which we tested the
adversative structures in a whole-part context; one where the nonadversative items were
tested in the null context condition; and finally one in which the nonadversative items were
tested in the whole-part context.
Method
Participants
We tested 144 persons, 36 per context group, 93 female and 51 male, between 20 and 26
years old (M = 22.6 years). They were all students at the University of Potsdam and were paid
4 € or received course credit for their participation.
11
Materials
We created 16 adversative context-sentence pairs which exhibited the whole-part relation
described in (1) above holding between a referent in the context sentence and a referent in the
target sentence (direct object). Then we created a nonadversative condition by changing the
verbs of the target sentences to yield an elaborative rather than an adversative reading. For the
null context group, the target sentences were stripped of their contexts. All subject noun
phrases were pronominals of masculine gender (er (he)); all object noun phrases were full
NPs with either case-unambiguous masculine gender (denACC; nine of 16 items), or with caseambiguous feminine gender (dieNOM/ACC; seven items). In addition to the SVO/OVS
structures2, the material contained 80 filler sentences testing the influence of length of
intervening adverbials on the acceptability of German subject- vs. object-extracted relative
clauses.
For all four groups, eight experimental lists were created by randomly combining the 16
experimental items with the fillers; eight further lists were created by inverting the order of
the eight original lists. Each pair of consecutive experimental items was separated by at least
one filler item. In the first version, four of the experimental items were presented in the SVO
condition, while the other four were presented in the OVS condition (the other eight items
being realized in the two remaining conditions; s. Fn 2). For the second version, the
assignment of items to word order condition was reversed. The two context groups were
tested on two different occasions, as were the two contrast groups.
Procedure
Participants were handed a 10-page booklet containing the experimental materials,
preceded by a short instruction asking them to read the sentences/sentence pairs carefully, not
to rely on normative standards in their assessment of the acceptability, and to use the full
range of the scale. Additionally, they were given two sample judgments, one of a perfectly
grammatical sentence judged with a ‘7’, and an ungrammatical one with a gender mismatch
12
([DenMASC AutoNEUTR] wäscht [dasNEUTR MannMASC], the car cleans the man.). Below each
experimental item, the scale from ‘1’ to ‘7’ was printed out, with endpoints being labelled as
‘totally unacceptable’ and ‘perfectly acceptable’. Booklets were handed to participants
together with the admonition to work independently of each other. The session lasted
approximately 15 minutes.
Design and Predictions
The design was 2 (WORD ORDER; SVO VS. OVS) × 2 (CONTEXT; null vs. whole-part) × 2
(ADVERSE; adversative vs. nonadversative relation between context and target sentence), the
latter two factors being tested between participants. Apart from a general acceptability
advantage of sentences presented in context (main effect of CONTEXT), we predicted that in
the null context, SVO sentences should be judged as more acceptable than their OVS
counterparts, while in the whole-part contexts, the reverse should be true, i.e., an interaction
of WORD ORDER and CONTEXT. Moreover, we predicted this strong licensing effect to hold for
the contrastive group only, i.e. an interaction of ADVERSE with the two other factors.
Data Analysis
Raw judgment scores were subjected to repeated measures ANOVAs for participants and
items with the factors WORD ORDER, CONTEXT, and ADVERSE.
Results
The mean judgments scores for the four conditions of Experiment 1 are summarized in
Table 1.
Table 1: Overall mean ratings (standard deviations)
adversative condition
null context
whole-part context
SVO
5.78 (1.44)
5.93 (1.59)
OVS
5.61 (1.46)
6.33 (1.18)
SVO
OVS
Nonadversative condition
null context
whole-part context
6.32 (1.24)
6.28 (1.41)
5.89 (1.56)
6.44 (1.12)
13
First, and most importantly, our results showed the predicted interaction of WORD
ORDER
and CONTEXT (F1(1,140) = 12.87, p < .001; F2(1,60) = 14.49, p < .001): while there
was no significant effect of CONTEXT on the SVO condition (both ts < 1), there was a
significant effect of CONTEXT on the OVS condition, t1(71) = 3.85, p < .001; t1(31) = 4.16, p
< .001. There was, accordingly, no overall main effect of WORD ORDER (SVO: 6.08 VS. OVS:
6.07), both Fs < 1, but, as predicted by the strong licensing hypothesis, an inverse effect of
word order in the whole-part context condition, t1(71) = 2.22, p < .05; t1(31) = 2.56, p < .05.
The other two main effects were significant, CONTEXT having, unsurprisingly, an overall
positive effect on ratings (no context: 5.09 vs. whole-part context: 6.25), F1(1,140) = 6.91,
p=.01; F2(1,60)=4.61, p=.04, as well as ADVERSE (adversative: 5.91 vs. nonadversative 6.23),
F1(1,140) = 5.97, p = .02; F2(1,60) = 3.89, p = .05. All other effects and interactions were not
statistically reliable; in particular, ADVERSE did not enter into any interaction with the other
factors, contrary to our prediction.
Discussion
We interpret our results as an instance of strong contextual licensing (in the sense
introduced above) of German marked OVS word order. While the unmarked SVO sentences
were unaffected by the contextual manipulation, the OVS sentences in the licensing wholepart context were rated significantly better than in the null context condition. Moreover, OVS
sentences were rated better than their unmarked SVO counterparts in the whole-part contexts.
That is, the marked word order OVS appears to be strongly licensed by a context in which the
referent of the object stands in a whole-part relation to a referent in the preceding context, and
in which the two referents are contrasted to each other with respect to a predicate established
in the discourse (e.g., to wash the whole car vs. to leave out one of its parts, the side mirror).
Note that this holds despite the fact that the target sentences in the marked condition exhibited
14
non-parallel syntactic structure to the preceding context sentence, different from the target
sentences of Bornkessel & Schlesewsky (2006).
Despite the strong licensing of the OVS order, the sentences are by all means acceptable
also with SVO word order; both word orders have an average acceptability value in the
vicinity of 6 which signals high acceptability on a 7 point scale. Our results therefore also
reflect the overall optionality of word order variation in German.
In addition, our results did not show an effect of the adversativity of the context-targetpair on the licensing effect. Apart from a general preference for the nonadversative items,
which could either be a general preference for sentences with more lexical material in them,
or simply a sampling error, we did not find any interaction of the ADVERSE factor with the
other factors. This can be taken as a first piece of evidence that adversativity as an instance of
a contrastive discourse relation between the context and the target sentence is not a necessary
ingredient of strong licensing.
Since the acceptability rating task can be argued to involve a fair deal of off-line
metalinguistic processes, we wanted to know whether a strong licensing effect of whole-part
contexts is also obtained in an on-line measure such as self-paced reading times.
4. EXPERIMENT 2
In this experiment, we tested whether the strong licensing effect found for acceptability
ratings in Experiment 1 can also be established for reading time (RT). That is, we
hypothesized that, while we should observe a processing difficulty for the marked OVS
structures in the null context (as compared to the unmarked SVO), RTs for OVS in the wholepart context should be shorter than those for SVO. Concerning the factor ADVERSE, we
expected the same main effect on reading times, as it was previously found on acceptability
judgments in Experiment 1.
Method
Participants
15
144 students from Potsdam University, 36 for each group, took part in Experiment 2 (95
female, 49 male; age range: 19-36 years, M = 23.2 years). They were paid 4 € or received a
course credit for their participation.
Materials
The same set of materials as in Experiment 1 was used. For the null context group,
critical sentences were stripped of the context sentences. Comprehension questions were
constructed which asked for factuality of the event described in the critical sentence. For
example, if the critical sentence was ‘He left out the side mirror’ (s. (1) above), the yesquestion would read ‘Did he forget to wash something?’, while the no-question would read
‘Did he wash the side mirror?’. For the nonadversative conditions, the comprehension
questions were adjusted accordingly. All other details of the preparation of the experimental
items and fillers were identical to Experiment 1. The overall experiment consisted of 96 trials.
Procedure
Participants were tested individually at a PC using the LINGER experimental software
(version 2.94, http://tedlab.mit.edu/~dr/Linger/) developed by Doug Rohde. The experiment
started with a training phase familiarizing the participants with the self-paced reading
procedure. The experimental trials worked as follows: First, a start screen told participants to
press the space bar when they were ready. Then, the sentences (for the null context group) or
the sentence pairs (for the whole-part context group, one line per sentence) appeared leftaligned and masked on the screen by underscores matching the length of the masked word. By
pressing the space bar, the first word appeared. A further press of the space bar revealed the
second word, while the first word disappeared (non-cumulative moving window). By pressing
the space bar after having read the last word of the sentence or sentence pair, respectively, the
comprehension question appeared on the screen. Answers were indicated by a ‘yes’ area on
the lower left of the screen, and a ‘no’ area on the lower right. Participants were instructed to
respond by using the ‘1’ key of the numerical keyboard for ‘yes’, and ‘3’ for ‘no’. After the
16
response to the comprehension question, the start screen reappeared and the next trial would
start. Overall, the experiment lasted approximately 35 minutes.
Design and Predictions
The design was the same as in Experiment 1, i.e. 2 (WORD ORDER) × 2 (CONTEXT) × 2
(ADVERSE), with the latter two again realized between participants. As in Experiment 1, we
predicted a main effect of CONTEXT (RTs for the context group to be shorter than for the null
context group). More importantly, we predicted an interaction of CONTEXT and WORD ORDER:
in the null context, SVO RTs should be shorter than those for the marked OVS order, while
OVS should show shorter RTs than SVO in the whole-part context.
Data Analysis
We chose the reading times of the sentence final participle (lexical verb) as the
dependent variable. First, the position of this constituent was the only one being identical
across all conditions. Second, the lexical verb has been shown to be the carrier of word order
effects in previous studies (see Meng and Bader, 2000; Gorrell, 2000, among others). Raw
RTs for the participles were screened for outliers (50ms < RT > 5000ms), and only RTs for
correctly answered trials were entered into the analysis (8.3% of the data were excluded by
this step). Following common usage, we log-transformed reading times and submitted log RT
as well as raw RTs to separate repeated measure ANOVAs for participants and items. These
analyses yielded the same results for both measures.
Results and Discusssion
Untransformed reading times for the sentence final participle for all conditions are
summarized in Table 2.
Table 2: Raw reading times of the sentence final participle in ms
(standard deviations), adversative condition
null context
whole-part context
SVO
656 (99)
652 (171)
OVS
647 (132)
599 (86)
17
SVO
OVS
nonadversative condition
null context
whole-part context
521 (73)
547 (72)
538 (74)
582 (41)
The first thing to note is that we found no evidence for the predicted main effect of
CONTEXT
(F1< 1; F2(1,60)=1.32, p >.10). Furthermore, the predicted interaction of word order
and context turned out to be significant, though only marginally by items (F1(1,140)=5.48,
p=.02; F2(1,60)=3.44, p=.07). Given the finding of the interaction in the rating experiment,
we were interested in the effect of WORD ORDER separately for the two context groups, and
performed a single comparisons for the two CONTEXT conditions. For the null context
condition, we found no effect of word order (both ts < 1). Sentence final participles of
sentences presented in isolation were read equally fast for SVO and OVS word order. In the
whole-part context condition, however, we found significantly faster reading times in the
OVS than in the SVO condition (t1(71)=2.85, p<.05; t1(31)=3.16, p<.01). Finally, ADVERSE
had a significant main effect: participles of critical sentences in the nonadversative condition
were read faster than those in the adversative condition (639 vs. 547 ms). However, ADVERSE
did not enter into any interaction with the other factors.
We interpret the result from the self-paced reading study in terms of a strong licensing
effect of the whole-part context: while reading times between unmarked SVO and marked
OVS structures did not differ in the null context, marked OVS structures were processed
significantly faster than their unmarked SVO counterparts in the context where the object
referent stood in a whole-part relation to a referent mentioned in the context sentence. We
conclude from this that strong contextual licensing of marked OVS word order in German can
be obtained if the local discourse context provides a whole-part relation in which the object,
but not the subject of the sentence participates. Furthermore, it seems that, apart from the fact
that the nonadversative items in our materials were generally processed faster than the
18
adversative ones, the adversativity of the whole-part relation does not play a role for the
processing of marked word order in context.
General Discussion
The most important aspect of the findings described above is that marked OVS word
order in German can indeed by strongly licensed by whole-part contexts of the type we have
employed. Not only are OVS structures judged as more acceptable if they are embedded in
such contexts, they are also processed faster, as indicated by the reading times for the lexical
verb in Experiment 2. As mentioned above, this strong licensing effect does not mean,
however, that the unmarked word order is not acceptable in the licensing context. Rather, the
difference in acceptability is small.
We attribute the failure of previous studies to find strong licensing to the fact that in
these studies, participants were presented sentences with both arguments realised as full NPs,
as well as to their employment of context types which apparently induced a licensing effect
not strong enough to reverse the subject-first preference found in the null context.
Pairs of sentences which both involve two lexical NPs are disadvantageous from the
perspective of textual coherence, because they often imply either a repeated name penalty
(Gordon, Grosz, & Gilliom, 1993) or a very complex textual relation between the two
sentences. The use of a pronominal subject in the second sentence of our experimental items
avoids these difficulties. However, the decision may appear problematic in other respects.
Frey (2006) assumes an operation of formal movement for German that allows to front
material across an unstressed pronoun in the absence of pragmatic licensing. Likewise,
Fanselow & Lenertová (2010) also show that objects can move across unstressed subjects
without pragmatic motivation. These observations may be in line with the idea that the
leftmost placement of the strongest accent is prosodically optimal in German (Féry 2008).
OVS structures with a pronominal subject thus come with a number of syntactic and
19
phonological properties that need to be controlled for before more far-reaching conclusions
can be drawn.
To our knowledge, the type of whole-part relation we have employed in our experiment
was not tested so far in connection with German OVS structures, although they have been
tested in the context of locative PP fronting in German (see Hörnig et al., 2006). Given that
poset-relations have been claimed to play an important role in the licensing of marked word
order constructions in English (e.g. Prince, 1998 for topicalisation), it does not come as a
surprise that a language with a more flexible word order than English, as German is, can be
shown to be sensitive to this kind of contextual relation. As remarked above, this is also in
line with the corpus study of Speyer (2007).
This being said, we want to point out that the present findings have to be taken with
caution for one further reason. Neither of our experiments showed clear evidence for a subject
first preference in the null context condition—in Experiment 1, SVO was judged better than
OVS, but this effect did not reach significance; and in Experiment 2, we found no difference
in processing difficulty between SVO and OVS in the null context condition. We ascribe this
finding to the fact that the object NPs of the critical sentences made the accommodation of a
context (in fact, a whole-part context), in which the sentence might not be marked, highly
salient. Accordingly, participants in the null context group might have been successful in
accommodating such a context on-line, which would explain the lack of a clear markedness
effect even in the null context.
These limitations notwithstanding, we take the current results to show clearly that
German marked word order is subject to contextual licensing effects, and that these licensing
effects can reverse the relation between marked and unmarked forms with respect to
acceptability and processing difficulty. We did not attempt to integrate these findings into a
larger theoretical picture of how contextual information and sentence-level preferences should
be taken to interact both in theoretical and in processing terms. But we are confident that the
20
evidence presented here can be taken as a starting point for further investigations of the
sensitivity of marked word order to local discourse structure.
21
References
Arnold, J. E., T. Wasow, T. Losongco, & R. Ginstrom (2000). Heaviness vs. Newness: the
effects of structural complexity and discourse status on constituent ordering. Language, 76(1),
28-55
Bader, M. (1999). Die Verarbeitung von Subjekt-Objekt Ambiguitäten im Kontext. In:
Proceedings der 4. Fachtagung der Gesellschaft für Kognitionswissenschaft. St. Augustin:
Infix, 31–36
Bader, M., & M. Meng (1999). Subject-object ambiguities in German embedded clauses:
An across-the-board comparison. Journal of Psycholinguistic Research, 28, 121-144.
Battistella, E. (1996): The Logic of Markedness. OUP.
Birner, B. & G. Ward (1998): Information Status and Noncanonical Word Order in
English. Amsterdam/Philadelphia: John Benjamins.
Bornkessel, I., & M. Schlesewsky (2006). The Role of Contrast in the Local Licensing of
Scrambling in German: Evidence from Online Comprehension. Journal of Germanic
Linguistics, 18.1,1–43.
Burkhardt, P., G. Fanselow, & M. Schlesewsky. (2007). Effects of (In)transitivity On
Structure Building and Agreement. Brain Research, 1163, 100-110.
Fanselow, G. (2001). Features, theta-roles, and free constituent order. Linguistic Inquiry,
32:405–437.
Fanselow, G., D. Lenertová, & T. Weskott (2008): Studies on the Acceptability of Object
Movement to Spec,CP. In: A. Steube (ed): The Discourse Potential of Underspecified
Structures. Berlin: Mouton de Gruyter. 413-437.
Fanselow, G., & D. Lenertová (2010): Left Peripheral Focus: Mismatches between Syntax
and Information Structure. To appear in Natural Language and Linguistic Theory.
22
Féry, C. (2008): Syntax, Information Structure, Embedded Prosodic Phrasing and the
Relational Scaling of Pitch Accents. To appear in N. Erteschik-Shir & L. Rochman (eds.): The
Sound Patterns of Syntax. OUP.
Frey, W. (2004). The grammar-pragmatics interface and the German prefield. Sprache
und Pragmatik 52. 1-39
Frey, W. (2006): Contrast and movement to the German prefield. In: V. Molnár & S.
Winkler (eds.): The architecture of focus. Berlin: Mouton de Gruyter, , 235-264
Gordon, P.C., R. Hendrick, M. Johnson, & Y. Lee (2006): Similarity-Based Interference
During Language Comprehension: Evidence From Eyetracking During Reading. Journal of
Experimental Psychology: Learning, Memory, and Cognition, 32(6), 1304-1321.
Gordon, Peter C., Barbara J. Grosz & Laura A. Gilliom (1993): Pronouns, Names, and the
Centering of Attention in Discourse. Cognitive Science 17, 311-347.
Gorrell, P. (2000): The subject-before-object preference in German clauses. In: B.
Hemforth & L. Konieczny (eds.): German Sentence Processing. Dordrecht: Kluwer, 25–63.
Grosz, Barbara J., Aravind K. Joshi & Scott Weinstein (1995): Centering: A Framework
for Modeling the Local Coherence of Discourse. Computational Linguistics 21, 203-225.
Haider, H. & I. Rosengren (1998): Scrambling. Sprache and Pragmatik, 49, Lund.
Haider, H. & I. Rosengren (2003): Scrambling: Nontriggered Chain Formation in OV
Languages. Journal of Germanic Linguistics 15.3, 203 - 267
Hawkins, J.A. (1995): Argument-predicate structure in grammar and performance: A
comparison of English and German. In: I. Rauch & G.F. Carr (eds.): Insights in Germanic
Linguistics 1, Berlin: de Gruyter, 127-144.
Hemforth, B. (1993): Kognitives Parsing: Repräsentation und Verarbeitung sprachlichen
Wissens. Sankt Augustin: Infix.
Hoberg, U. (1981): Die Wortstellung in der geschriebenen deutschen Gegenwartssprache.
München: Hueber.
23
Höhle, T. (1982): Explikation für ‘normale Betonung’ und ‘normale Wortstellung’. In: W.
Abraham (ed.): Satzglieder im Deutschen. Tübingen: Narr, 75-153.
Hörnig, R., K. Oberauer, & A. Weidenfeld (2005): Two principles of premise integration
in spatial reasoning. Memory & Cognition, 33(1):131–139.
Hörnig, R., T. Weskott, , R. Kliegl, & G. Fanselow (2006): Word order variation in spatial
descriptions with adverbs. Memory and Cognition, 34(5), 1183-1192.
Hörnig, R., T. Weskott, , G. Fanselow & R.Kliegl (ms.): Linguistic Markers of TopicShifts.
Keller, F. (2000): Gradience in Grammar: Experimental and Computational Aspects of
Degrees of Grammaticality. PhD Thesis, University of Edinburgh.
Krifka, M. (2007): Basic notions of information structure. In C. Fery and M. Krifka (eds.):
Interdisciplinary Studies of Information Structure 6, Potsdam: Universitätsverlag Potsdam.
Marslen-Wilson, W., & J. Bayer (1992): Configurationality in the Light of Language
Comprehension: The Order of Arguments in German. Manuscript. Nijmegen: MPI for
Psycholinguistics
Müller, G. (1999): Optimality, Markedness, and Word Order in German.
Linguistics,37:777–818.
Neeleman, A., & H. Van De Koot (2008): Dutch Scrambling and the Nature of Discourse
Templates. The Journal of Comparative Germanic Linguistics, 11.2, 137-189
Prince, E. (1998): On the limits of syntax, with reference to Left-Dislocation and
Topicalization. In: P. Culicover & L. McNally (eds): Syntax and semantics. Vol. 29. The
limits of syntax. NY: Academic Press, 281-302.
Rizzi, L. (1997): The Fine Structure of the Left Periphery. in L. Haegeman (ed.): Elements
of Grammar. Dordrecht: Kluwer.
Speyer, A. (2007): Die Bedeutung der Centering Theory für Fragen der Vorfeldbesetzung
im Deutschen. Zeitschrift für Sprachwissenschaft 26, 83-115
24
Uszkoreit, H. (1986): Constraints on order. Linguistics, 24, 883-906.
Weber, A., & K. Mueller (2004): Word order variation in German main clauses: A corpus
analysis. In: Proceedings of the 20th International Conference on Computational Linguistics.
Weskott, T. (2003). Information structure as a processing guide. PhD dissertation,
University of Leipzig.
Weskott, T., B. Stolterfoht, I. Bornkessel, & M. Schlesewsky (2004): The TaskDependency of Acceptability Judgements: Processing Scrambling and Topicalization in
German. Paper presented at the 26th DGfS, Johannes-Gutenberg-Universität, Mainz.
25
Acknowledgements
This research was supported by the Deutsche Forschungsgemeinschaft (SFB 632, Project
C1). We want to thank Chuck Clifton, Lyn Frazier, Claudia Maienborn, and Matthias
Schlesewsky for helpful discussion. We thank Petra Schienmann and her team, as well as
Lena Benz, Jutta Boethke, Pavel Logacev, Anselm Metzger, Cornelia Neumann, and Nikolaus
Werner for their assistance in conducting the experiments.
Footnotes
1) The idea to test the contribution of adversativity to the licensing effect was brought to
our attention by Claudia Maienborn.
2) The experiments contained two further conditions realized within items and subjects:
one with a target sentence exhibiting SVO order and the adversative conjunction
“aber” (but; “Er hat aber den Außenspiegel ausgelassen.”, But he omitted the rearview mirror.); and one with SVO order and a pronominal epithet instead of the subject
pronoun (“Der Trottel hat den Außenspiegel ausgelassen.”). Since the two conditions
are not of interest for the issue of contextual licensing, we will not report the results
for them here.
Robin Hörnig 18.11.10 16:49
Gelöscht: contrastivity