and psycholinguistic investigations of linguistic ... - Semantic Scholar

Sandra Pappert (Leipzig), Johannes Schließer (Leipzig), Dirk P. Janssen (Kent) & Thomas Pechmann (Leipzig)

Corpus- and psycholinguistic investigations of linguistic constraints on German object order 1 Introduction The aim of this paper is to evaluate several linguistic constraints as predictors of the frequency with which word order variants of double object sentences occur, as well as of the processing ease associated with these structures. We will provide corpus data and behavioural data from completion questionnaires and self-paced reading experiments that address the validity of the Dative constraint, the Animacy constraint, and the Definiteness constraint. Case is found to have an effect on the order of the objects in double object structures. However, this effect cannot be disentangled from an Animacy effect because dative and animacy on the one hand and accusative and inanimacy on the other hand turn out to be strongly associated. Finally, evidence for the validity of the Definiteness constraint is only found in the corpus data extracted from its context but not in the behavioural data gathered without context.1 German is generally assumed to be a language with a relatively free word order. The acceptability of a given structure is said to depend on the interaction of several linear precedence rules that refer to morphological, semantic, and information structural properties, such as Case, Animacy, Definiteness, Pronominality, and Focus (e.g., Lenerz 1977; Uszkoreit 1986, 1987; Jacobs 1988) (PR: Precedence rule, read < as 'precedes'):

1

The studies presented here were pursued in the project Predictive Sentence Processing in German (http://www.uni-leipzig.de/~parsing/) supported by the German Research Foundation. The authors would like to thank Gerard Kempen as well as an anonymous reviewer for helpful comments on an earlier version of this paper.

300

Sandra Pappert, Johannes Schließer, Dirk P. Janssen & Thomas Pechmann

(PR 1) (PR 2) (PR 3) (PR 4) (PR 5) (PR 6)

+ NOM < – NOM DAT < ACC + ANIM < – ANIM + DEF < – DEF + PRONOUN < – PRONOUN – FOCUS < + FOCUS

Within an Optimality Theoretic framework, a ranking of these constraints has been proposed (e.g., Müller 1999; Büring 2001) which is intended to capture the relative acceptability of word order variants. Corpus data may inform us about the predictive power of the linguistic factors with respect to the frequency of occurrence of the structures (cf. Gries 2003; Bresnan, Cueni, Nikitina & Baayen in press). Recently, corpus studies have been reported that address the issue of word order variation in German (Kempen & Harbusch 2003a, 2003b, 2004, 2005; Weber & Müller 2004). As an alternative to corpus studies, behavioural studies may indicate whether processing is influenced by the hypothesized differences in acceptability of these structures. Numerous studies found evidence for the subject-first preference during processing of German sentences (e.g., Bader 1994; Hemforth, Konieczny & Strube 1993; Pechmann, Uszkoreit, Engelkamp & Zerbst 1996). Moreover, there is some evidence that participants prefer pronouns to precede non-pronominal constituents (Keller 2000). Effects of other constraints have been also show to exist. The data will be discussed below (e.g., Featherston no year; Keller 2000; Pechmann et al. 1996; Rösler, Pechmann, Streb, Röder & Hennighausen 1998). The current paper reports a series of corpus queries as well as psycholinguistic experiments, namely completion questionnaires and selfpaced reading experiments. Crucially, materials consist of structures with the subcategorizing verb in final position, such that information carried by the verb cannot influence expectations towards upcoming materials. Contrary to materials from most studies cited above, the materials used in the current study consist of declarative main clauses with a nominal constituent in the so-called Vorfeld position, i.e., the initial position of a sentence with an inflected auxiliary in second position – cf. (1). (1)

Der Mann wird dem Kind den Brief geben. theNOM man will theDAT child theACC letter give The man will give the letter to the child.

Corpus counts indicate that such sentences are more numerous in German (see Corpus Study 1) than sentences with all constituents in the

Investigations of linguistic constraints on German object order

301

Mittelfeld, e.g., the region between an initial complementizer and the sentence-final verb, cf. Kempen & Harbusch 2003, 2005; cf. (2). (2)

... dass der Mann dem Kind den Brief geben wird. ... that theNOM man theDAT child theACC letter give will ... that the man will give the letter to the child.

The data presented here relates to the relative impact of the Dative constraint (PR 2), the Animacy constraint (PR 3), and the Definiteness constraint (PR 4). The processing of German nominative-first sentences with two objects will be investigated.

2 Testing the Dative constraint As the Nominative constraint (PR 1), which states that nominatives tend to precede non-nominatives, is relatively well established, this factor will be ignored here and the focus will lie on ditransitive sentences that have the subject in first position. Accordingly, the Dative constraint (PR 2), stating that datives tend to precede accusatives (cf. Büring 2001; Müller 1999), will be addressed and its validity for corpus- and psycholinguistic data will be explored. Whereas several authors formulate precedence rules by reference to the case of the involved constituents (e.g., Büring 2001, Müller 1999, Uszkoreit 1987), others refer to syntactic functions (e.g., Lenerz 1977), and still others emphasize the importance of thematic roles for word order constraints (Uszkoreit 1986). As we wanted to dissociate semantic effects from morphosyntactic ones and as thematic roles become explicit late in sentences with the subcategorizing verb in final position, we decided to discard Thematic Role as a factor and to treat Animacy as a separate factor. In this section, Case effects will be explored. The reference to Case (and not to syntactic function) reflects the annotation criteria of the used syntactic corpora as well as the experimental manipulation. 2.1 Corpus Study 1 Corpus counts on German subordinated sentences with varying word order (Kempen & Harbusch 2003a, 2003b, 2004, 2005) show that syntactic functions have a strong impact on the frequency of occurrence of the respective structures. The following findings hold for full nominal constituents: Firstly, subject-first sentences are much more frequent

302


than sentences with an object before the subject. Secondly, indirect (i.e., dative) objects tend to precede direct (i.e., accusative) objects in double object sentences. Even though data on non-pronominal double object structures is sparse, they indicate that the use of sentences with non-canonical word order is fairly restricted indeed. The goal of the present study was to re-address the issue of word order variation in German for double object sentences with a nominal constituent in the Vorfeld. Consequently, it adds to the sample provided by Kempen and Harbusch. 2.1.1 Method Syntactic frequencies were extracted from two German newspaper corpora, Negra2 and Tiger. These corpora include about 60.000 syntactically annotated sentences that are already marked for Case. As the syntactic status (argument vs. adjunct) of the constituents is not specified in the annotations, the following reports will refer to Case instead of syntactic function. Criteria for the extraction of structures were as follows: Sentences including a control or a raising verb were excluded. Complex sentences were split into clauses. The critical clauses had to consist of three nominal constituents marked for nominative, dative, and accusative. Clauses with pronouns were excluded, all constituents had to be represented by full noun phrases instead. The nominative constituent had to be positioned in the Vorfeld. Finally, only clauses with a finite auxiliary in the second position and the subcategorizing verb at the end were considered.2 2.1.2 Results In sum, 196 clauses were extracted that satisfied the constraints specified above. There were 176 sentences with the order dative before accusative and 20 sentences with the order accusative before dative.

2

The last criterion ensured that the clauses extracted from the corpus matched the sentences presented in the experiments.


303

2.1.3 Discussion The word order pattern found for double object nominative-first structures agrees with the Dative constraint. As such, it may be taken as an indication that dative before accusative is in fact the unmarked word order and that the alternative linearization of the accusative before the dative is indeed the marked word order. Note that the Dative constraint is not uncontroversial. Thus, some authors claim that both object orders are unmarked for at least some verbs (Höhle 1982; Haider 1993). The findings from Corpus Study 1 add to results from Kempen and Harbusch (2003a, 2003b, 2004, 2005). Both investigations come to the conclusion that a linearization with a dative preceding an accusative is preferred. However, one should bear in mind that the data from Kempen and Harbusch, as well as that presented here, relies on the same sources, namely Negra2 and Tiger. The use of a common sample may involve confounds, as for example identical text types or the same source of annotation. The fact that Kempen and Harbusch (2005) also supply data from the Verbmobil corpus does not remedy this problem because in this corpus of spoken dialogs structures without pronouns were rarely produced. Consequently, objections could be raised. One pertains to the syntactic status of the constituents. As Lenerz (1977) points out, free datives are more restricted in their position than subcategorized datives. But the syntactic annotation of the used corpora does not differentiate between subcategorized and free datives. Thus, the inclusion of both dative types might have produced a bias towards the satisfaction of the Dative constraint. Nonetheless, there is an obvious reply to such objections. The experiments presented below focus on processes involved in the integration of datives and accusatives in sentences with the subcategorizing verb in final position. Accordingly, the subcategorizing verbs are not yet explicitly available to the processing system and the syntactic status of a constituent cannot be determined. 2.2 Completion Questionnaire 1 Several questionnaire studies have dealt with word order preferences in double object structures. Keller (2000) and Featherston (no year) presented ditransitive sentences and varied the order of the nominal constituents. Subjects rated sentences with datives preceding accusatives as more acceptable than sentences with accusatives preceding datives.

304


However, Kempen and Harbusch (2005) cite the study by Keller (ibid.) and show that structures with the accusative before the dative yielded better acceptability ratings than the corpus data would predict. We decided to re-address the issue of word order preferences in double object structures in a completion task. Completion data should satisfy similar processing constraints to corpus data (cf. the production threshold proposed by Kempen & Harbusch 2005). 2.2.1 Method The completion task was administered as a paper and pencil questionnaire to thirty-two native speakers of German. Before completing the questionnaire, participants were not informed about the aims of the study. Materials included 32 experimental items and 32 filler items. The experimental items obeyed the following constraints: They consisted of two full noun phrases. The first one was a nominative constituent occupying the Vorfeld, followed by an auxiliary followed by a second noun phrase. The auxiliary wird 'will' was used in all fragments because it does not restrict the type of the subcategorizing verb. Both nouns were masculine and unambiguously marked for case. The nouns referred to animate entities and were headed by definite determiners. This helped to control potential effects of Animacy or Definiteness (cf. Completion Questionnaires 2 and 3). Case of the second noun phrase was manipulated. It was either marked for accusative or for dative. Examples are given in (3) and (4). (3) (4)

Der Doktor wird dem Krankenpfleger .... theNOM doctor will theDAT nurse ... Der Doktor wird den Krankenpfleger .... theNOM doctor will theACC nurse ...

Sentence fragments were distributed over two lists according to a Latin square design such that each list included each item once and had an equal number of items per condition. Filler items were added and the lists were pseudorandomized. They were assigned to an equal number of subjects. 2.2.2 Results Completions of the experimental items were hand-coded. The completions were categorized as to whether they resulted in a double object


305

structure, an error (i.e., an ungrammatical structure), another structure, or whether they were missing. The respective proportion of completions per condition is given in Figure 1.

100 90 80 70 60 50 40 30 20 10 0

86,3

double object error other missing

53,7 40,4

4,9

nom-dat …

1

4,5

7,4 0,8

nom-acc …

Figure 1: Proportion of completions per Case condition from Completion Questionnaire 1 (in %).

The proportion of missing completions was negligible in both conditions. Subjects did not commit many errors, but there were slightly more errors after an accusative than after a dative. The majority of the completions in the dative condition resulted in double object structures (53.7 %), whereas double object completions were rare in the accusative condition (4.5 %, most "other" completions were single object completions). 2.2.3 Discussion In accordance with the corpus data on double object structures, there were more linearizations with the dative preceding the accusative than the inverse pattern. Object order in completions does not seem to be less constrained than object order in the corpus sample. Thus, corpus and completion data can be interpreted as joint evidence in support of the Dative constraint.

306


2.3 Reading Experiment 1 In the preceding sections, corpus data and off-line completion data were presented that address the issue of word order preferences in double object structures. The aim of the experiment reported here was to ascertain whether the dative-before-accusative preference found in the off-line data also holds for on-line data. This preference has already been attested in a series of production experiments (Pechmann, Uszkoreit, Engelkamp & Zerbst 1996) as well as in a study measuring Event-Related Potentials (ERPs) during reading of sentences with the subcategorizing verb in final position (Rösler, Pechmann, Streb, Röder & Hennighausen 1998). Moreover, it has been shown that it applies to full noun phrases but not to pronouns (Schlesewsky, Bornkessel & Frisch 2003). However, Rösler et al. (ibid.) report a result that is somehow surprising in the light of the data reported above. They found an early effect on the accusative determiner when the accusative preceded the dative. But at this moment, the structure could either be completed as a double object structure or as another structure. Taking the frequency and completion data reported above into consideration, one would predict that subjects prefer a sentence with a single object (cf. Gibson, 1998), and therefore, no evidence for processing difficulties should arise on the accusative. We suggest that the early effect attested by Rösler et al. was an artefact. Due to the absence of fillers, double object structures were predictable. Thus, a reading experiment with materials including single and double object sentences was conducted to substantiate this objection. 2.3.1 Method Forty-four subjects participated in this experiment. All had German as their native language. None had participated in Completion Questionnaire 1 and all were naive with respect to the purpose of the study. The 36 experimental sentences were double object structures with an auxiliary in second and the main verb in final position. All nominal constituents included masculine nouns and they were introduced by definite determiners that were unambiguously marked for case. The nominative occupied the Vorfeld, while the two objects were placed in the Mittelfeld, one marked for dative, the other marked for accusative. All referents were animate. As ditransitive verbs allowing for three animate referents are rare, each verb appeared in two sentences. In contrast, there were no noun repetitions across sentences. The critical conditions


307

differed in the Case of the first object (dative or accusative). Examples are given in (5) and (6). (5)

(6)

Der Doktor wird dem Krankenpfleger den theACC theNOM doctor will theDAT (male) nurse Rollstuhlfahrer zeigen. wheel-chair-person point out to The doctor will point the person in the wheel-chair out to the nurse. Der Doktor wird den Krankenpfleger dem theDAT theNOM doctor will theACC (male) nurse Rollstuhlfahrer zeigen. wheel-chair-person point out to The doctor will point the nurse out to the person in the wheel-chair.

Four more permutations were presented that are not discussed here for reasons of space, the data is reported in Pappert, Schließer, Janssen & Pechmann (2005). Thirty-six single object distractor sentences had an auxiliary in second and the subcategorizing verb in final position. They were inserted to prevent subjects from predicting sentences' length. Moreover, 72 filler sentences of different syntactic structure were added. For each experimental sentence, either a matching or a non-matching sentence differing in case of the first object was constructed. For example, (6) functioned as a non-matching sentence for (5). Experimental items in six conditions were distributed over six lists according to a Latin square, such that each list contained each item once and six items in each condition. Six pseudo-randomised versions of each list were generated.Sentences were presented word-wise in a moving window for self-paced reading without masking signs. Subjects were instructed to push a button as soon as they were ready to read the next word. When a sentence was completely read, a control sentence appeared, and the subjects pushed a yes or a no button to indicate whether the critical and the control sentence were identical. 2.3.2 Results The data from eight subjects who committed 39 % or more errors in the matching task was excluded from further analysis. The data from the remaining 36 subjects (mean error rate = 20.8 %) entered the data treatment. Outliers that exceeded more than two standard deviations of the mean reaction times per condition were excluded.

308


In Figure 2, the first determiner (Det1), the first noun (N1), the auxiliary (Aux), the second determiner (Det2), the second noun (N2), the third determiner (Det3), the third noun (N3), and the sentence-final verb (V) are plotted along the x-Axis, mean reading times in milliseconds per condition are plotted along the y-axis. T-tests for pairs were calculated. There were no significant differences between conditions from the first determiner up to the second noun. On the third determiner, there were longer reading times if the first object was marked for accusative than if it was marked for dative. This difference was significant in the analysis by subjects only (t1(35) = 2.37, p < .05; t2 (35) = 1.19, p = .24). Reading times on the third noun or on the subcategorizing verb did not differ significantly. However, there was a difference in the accuracy with which subjects performed the matching task. Subjects committed more errors in the Accusative condition (mean error rate = 22.2 %) than in the Dative condition (mean error rate = 12.0 %). This effect turned out to be significant in both analyses (t1(35) = 3.74, p < .01; t2 (35) = 2.90, p < .01). 1000 900 800 700 600 nom-dat-acc nom-acc-dat

500 400 Det1

N1

Aux

Det2

N2

Det3

N3

V

Figure 2: Mean reading times per Case condition from Reading Experiment 1 (in ms).

2.3.3 Discussion Reading times for the third determiner as well as error rates in the matching task yielded a significant Case effect (for the reading data at


309

least in the subject analysis) with longer reading times and higher error rates in the accusative condition than in the dative condition. This effect indicates processing difficulties during reading of sentences with an accusative preceding a dative as compared to reading of sentences with a dative before an accusative. Such a finding is in line with the Dative constraint. As single object distractors were included, the effect on the third determiner can be explained as follows. After an accusative object, subjects expected a single object structure. But this possibility was ruled out when the third determiner appeared instead of a verb. The integration of the second object induced processing costs that did not arise to a similar extent after a dative object. Crucially, there was no processing disadvantage on the second determiner in the accusative condition as compared to the dative condition. We propose that the effect reported by Rösler et al. (1998) in this position was due to the absence of single object distractors in their study. These findings are in line with expectations formulated by Gibson (1998) within the framework of Syntactic Prediction Locality Theory, even though one should bear in mind that there is no evidence to generalize them irrespective of Animacy (see below). 2.4 Evidence for the Dative constraint In the preceding sections, empirical data from both corpus counts and psycholinguistic experiments was presented that provides consistent evidence that double object structures with the subject in the Vorfeld are preferred when the dative precedes the accusative as compared to the inverse order. This finding matches the data on double object structures with the subject and the two objects in the Mittelfeld (e.g., Featherston no year; Keller 2000; Kempen & Harbusch 2003a, 2003b, 2004, 2005; Rösler et al. 1998). Thus, several corpus- and psycholinguistic studies yield results that accord with the Dative constraint as one ingredient of an optimality theoretic account of word order variation in German (cf. Büring 2001; Müller 1999). However, the Dative constraint might interact with other constraints on German word order preferences. A potential confound in the studies presented here is Animacy. In the next sections, we will try to disentangle the influence of Case from that of Animacy on German word order preferences.

310


3 Testing the Animacy constraint The Animacy constraint (PR 3), which states that constituents referring to animate entities precede those referring to inanimate ones is another constraint that is assumed to have a strong impact on word order variation in German. Müller (1999) and Büring (2001) even assign a higher ranking to it than to the Dative constraint. In the following, the interaction of Animacy and Case with object order preferences will be investigated. 3.1 Corpus Study 2 In their corpus study based on Negra2, Kempen and Harbusch (2004) show that Animacy has an independent impact on the incidence of word order patterns in German. However, their data is relatively sparse, especially with respect to full accusative objects. Accordingly, they do not address linearization issues in double object structures with full noun phrases only. We decided to reconsider the sample from Corpus Study 1 to reveal potential effects of Animacy on syntactic preferences. 3.1.1 Method The sample was the same as for Corpus Study 1. It consisted of 196 double object clauses that were extracted from Negra2 and Tiger, including full nominal constituents only, the nominative one in the Vorfeld and the subcategorizing verb at the end. All nominal constituents were manually annotated for Animacy. They were assigned to one of three categories: animate, inanimate, or equivocal (i.e., nouns referring to collective or abstract entities, e.g., Vorstand 'management'). 3.1.2 Results Two analyses were performed, one with the equivocal items counted as animate and one excluding them. The overall pattern was very similar but the data was sparse after the exclusion of clauses with equivocal referents. Only results from the analyses where they were counted as animate entities will be reported here. First, the relative association of Case and Animacy of the two objects in nominative-first double object structures was determined. Ac-


311

cording to Table 1, in all but one item from the accusative condition the noun referred to an inanimate entity. In contrast, there were more animate than inanimate noun referents in the dative condition. animate inanimate

dative 151 45

accusative 1 195

Table 1: Number of nouns per Animacy by Case Condition from Corpus Study 2 (double object structures with the nominative in the Vorfeld).

Second, the number of structures per Animacy by Case Condition was calculated. Table 2 shows that in most sentences dative constituents with animate referents preceded accusative constituents with inanimate referents. animate before animate animate before inanimate inanimate before animate inanimate before inanimate

dative before accusative 1 134 0 41

accusative before dative 0 0 16 4

Table 2: Number of nominative-first double object clauses per Animacy by Case Condition from Corpus Study 2.

3.1.3 Discussion Corpus counts on objects annotated for Animacy reveal a strong association of Case and Animacy. Animate entities tend to be represented by datives whereas inanimate entities are preferentially represented by accusatives. An explanation might recur to thematic roles. In typical events, animate non-agents function as recipients whereas inanimate non-agents are the patients. Recipients are realized as indirect objects with dative Case and patients as direct objects with accusative Case (cf. Primus 1999). However, there are some structures that do not fit the thematic role generalizations because both objects refer to inanimate entities. An inspection of the used verb types reveals a high incidence of transfer verbs in non-literal use, e.g., Die Hochheimer haben ihrem Fund jedoch eine neue, durchaus angemessene Funktion verliehen 'The people from Hochheim attributed a new, by all means appropriate function to their find.' Since in the majority of these instances the dative object precedes the accusative object, independent evidence for the Dative constraint is found.

312


With respect to structures where the Dative and the Animacy constraint might interact, the data indicates that there is in fact no competition between these constraints but mere correspondence. At least in the sample considered here, there were no structures that include an animate accusative referent and an inanimate dative referent. Therefore, the rare violation of the Dative constraint is inseparable from the violation of the Animacy constraint. Consequently, there is no evidence to rank one constraint over the other. Moreover, these findings indicate that a systematic variation of Case and Animacy of both objects might not be practicable because of the lack of plausible verbs. 3.2 Completion Questionnaire 2 A completion questionnaire was designed so that Case and Animacy are independently manipulated. However, it only permits the manipulation of the subject and the first object, whereas the second object may or may not be added by the subjects. There is one study which tried to manipulate the Animacy of the second object as well. Featherston (no year) presented double object sentences in a grammaticality judgement task and included Animacy as a factor. But the claim to have manipulated Animacy does not stand a closer inspection of the materials. The datives assumed to refer to inanimate entities referred to institutions that may be interpreted as groups of persons instead (cf. the equivocal category in Corpus Study 2). 3.2.1 Method As in Completion Questionnaire 1, the completion task was administered as a paper and pencil questionnaire. Forty-nine native speakers of German participated in the experiment. They were naive with respect to the purpose of the study. Materials were composed of 36 experimental items and 36 filler items. The experimental items resembled those from Completion Questionnaire 1. The first full noun phrase was definite, marked for nominative and located in the Vorfeld. It referred to an animate entity. The second one was preceded by the auxiliary wird 'will'. Both nouns were of masculine gender, case marking was unambiguous. Animacy (animate vs. inanimate) and Case (dative vs. accusative) of the second noun phrase were manipulated. Examples are given in (7) and (8).

313


(7)

Der Doktor wird dem/den Krankenpfleger .... theNOM doctor will theDAT/ACC nurse ... Der Doktor wird dem/den Kieferknochen .... theNOM doctor will theDAT/ACC jaw bone ...

(8)

The questionnaire included two more conditions not reported here. In addition to Case and Animacy of the object, Animacy of the subject was manipulated. Sentence fragments were distributed over six lists. Each item occurred once in each list, in different conditions across lists. Lists included an equal number of items per condition. Filler items were added. The six lists were pseudorandomized, and each subject completed fragments of one list. 3.2.2 Results The data from one subject who made 44 % errors was excluded. The remaining completions were categorized as to whether they resulted in a double object structure, a grammatical error, another syntactic structure or whether they were missing (cf. Completion Questionnaire 1). Figure 3: Proportion of completions per Animacy by Case Condition from Completion Questionnaire 2 (in %). 98,6

100 91,7

90 80 70 60

60,1


52,1

50 40

34,1

31,9

30 20 10 0

6,6

5,2

9,4

nom-dat, anim-anim

5,9 1,7

0,7

nom-dat, animinanim

0,7

nom-acc, anim-anim

0,7 0,4

0,4

nom-acc, animinanim

Figure 3 specifies the relative proportion of these completions in the conditions considered here. The Case effect found in Completion Questionnaire 1 persisted: Double object completions were most frequent in

314


the dative condition and other completions were most frequent in the accusative condition. Other completions increased and errors decreased after an accusative referring to an inanimate entity. In contrast, double object completions decreased after a dative referring to an inanimate entity whereas missing completions increased.Double object completions were coded as to whether the inserted noun referred to an animate or inanimate entity. Table 3 reports the number of double object completions per Animacy by Case condition. animate before animate animate before inanimate inanimate before animate inanimate before inanimate



Table 3: Number of double object completions per Animacy by Case Condition from Completion Questionnaire 2.

The data on double object structures with the accusative before the dative was too sparse to determine any differences. In contrast, there was enough data on double object structures with the dative preceding the accusative. This data showed a strong prevalence of structures with the accusative referring to an inanimate entity. 3.2.3 Discussion In Completion Questionnaire 2, the interaction between Case and Number of Objects found in Completion Questionnaire 1 was replicated. After a dative, double object structures were more frequently produced whereas after an accusative, other structures were more frequently produced. Animacy of the first object tended to have an impact on this pattern with a more pronounced prevalence of other structures after inanimate accusative referents and a more pronounced double object prevalence after animate dative referents. Moreover, the data was once again too sparse to allow for a dissociation of Case and Animacy effects on the linear order of the two objects in double object sentences. The inserted accusatives mostly referred to inanimate entities. This phenomenon is attributed to the higher typicality of direct objects referring to inanimate entities as compared to such referring to animate entities. As discussed before, this may be mediated by thematic proto-roles, namely by the patient role which is typically realized as direct object (cf. Dowty 1991; Primus 1999). The slightly higher incidence of gram-


315

matical errors in the accusative-animate condition, i.e. completions of another accusative object, indicates that subjects tended to misread accusatives as datives if they referred to animate entities. After a dative referring to an inanimate entity, subjects experienced problems in conceiving an action involving a second object. The inanimate dative condition produced a higher incidence of missing responses than the animate dative condition. Moreover, double object structures were slightly less frequent in the former condition. However, there were relatively more double object completions than one would expect after consideration of the corpus data. The inserted objects referred to an inanimate entity, resulting in an inanimate-inanimate pattern. This pattern was predominant over an inanimate-animate pattern in the corpus data as well. As double object structures typically represent transfer actions which involve an animate recipient realized as dative (cf. Grimshaw 1990) the difference between animate and inanimate datives might again be mediated by thematic proto-roles. An inspection of the most frequently inserted verb types showed that completions after datives in both Animacy conditions mostly described transfer actions. The transfer verbs after inanimate datives were again used non-literally (cf. Corpus Study 2). Finally, we have to conclude that the experimental manipulation of Case and Animacy of the first object did not help to disentangle word order effects because the factors are too strongly associated. 3.3 Evidence for the Animacy constraint Corpus and completion data reveal a strong association of Case and Animacy in double object structures. Accusatives typically refer to inanimate entities whereas datives typically refer to animate entities. Due to the lack of ditransitive verbs that are plausible with accusatives referring to animate entities and datives referring to inanimate entities, a reading experiment is not feasible. The data show that both the Dative and the Animacy constraint are rarely violated. However, the strong association of Case and Animacy does not allow to dissociate the constraints and to rank one above the other. Finally, it must be stressed that this conclusion holds for the linearization of objects in double object structures only. As for single object structures, there is evidence from corpus and elicited production data (Kempen & Harbusch 2004; vanNice & Dietrich 2003) showing that constituents with animate referents tend to precede constituents with inanimate referents.

316


4 Testing the Definiteness constraint Another factor that is assumed to have an impact on word order variation is definiteness. The Definiteness constraint (PR 4) states that definite constituents tend to precede indefinite ones (cf. Lenerz 1977). Predictions based on the Definiteness constraint resemble those based on the Focus constraint according to which background information tends to precede focused information (cf. Abraham 1994; Müller 1999). The relatedness of the two constraints gives a hint to the involvement of contextual factors. But formulations of the Definiteness constraint do not explicitly specify contextual conditions of its application. We decided to investigate the validity of the Definiteness constraint ignoring contextual restrictions simply because as such, the Definiteness constraint is easier to operationalize than the Focus constraint. Even though the authors agree on the relevance of the Definiteness constraint, accounts differ with respect to the relative importance they attribute to the constraint. For example, Müller (1999) ranks Definiteness above Animacy and Dative whereas Büring (2001), quoting Müller (ibid.), ranks Animacy and Dative above Definiteness. In the following sections, corpus- and psycholinguistic studies will be presented in which the interaction of the Definiteness and the Dative constraint with object order in double object sentences was examined. 4.1 Corpus Study 3 Weber and Müller (2004) looked at Definiteness effects on word order in Negra2 sentences as well. However, they only considered single object sentences. Their data revealed a strong association of Definiteness and syntactic function, namely a prevalence of definite subjects and indefinite objects. But there was no evidence for the Definiteness constraint as a valuable predictor of the frequency of occurrence of word order variants. Kempen and Harbusch (2004) also considered Definiteness as a potential factor in word order variation. In their sample of Negra2, they differentiated between definite and indefinite constituents. However, they included pronouns in the definite category and their data without pronouns were sparse. As the studies cited here do not address the issue of object order variation in double object sentences, we decided to test for potential Definiteness effects in our corpus sample.


317

4.1.1 Method The sample was the same as for Corpus Study 1 and 2. It consisted of 196 double object clauses extracted from Negra2 and Tiger that included full noun phrases only, with the subject in the Vorfeld and the subcategorizing verb in final position. All nominal constituents were hand-coded for Definiteness. Constituents were categorized as definite if they were introduced by a definite or demonstrative determiner (e.g., der 'the', dieser 'that') or if they were proper nouns. In contrast, constituents were categorized as indefinite if they were introduced by an indefinite determiner (e.g., ein 'a' or a zero determiner in plural) or if they were quantified (e.g., alle 'all', viele 'many'). As quantified constituents may be interpreted as generic or as specific, the latter classification is not without problems (cf. Löbner 1985). However, an exemplary inspection of the Negra2 double object structures revealed no single instance of alle in a dative or accusative constituent. 4.1.2 Results First, the relative association of Definiteness and Case of the objects in the double object structures was determined. As shown in Table 4, the pattern is more obvious for the dative conditions as compared to the accusative conditions. definite indefinite

dative 170 26

accusative 81 115

Table 4: Number of objects per Definiteness by Case Condition from Corpus Study 3 (double object structures with a nominative in the Vorfeld).

Dative constituents were most frequently introduced by a definite determiner. In contrast, accusative constituents occurred more frequently with an indefinite determiner than with a definite one. Second, the incidence of word order variants per Definiteness by Case Condition was determined. Table 5 indicates that sentences with the accusative before the dative were sparse. In the dative before accusative condition, definite constituents mostly preceded indefinite ones.

318


definite before definite definite before indefinite indefinite before definite indefinite before indefinite



Table 5: Number of double object clauses per Definiteness by Case Condition from Corpus Study 3 (double object structures with a nominative in the Vorfeld).

4.1.3 Discussion Counts on objects reveal an association of Definiteness and Case. In double object structures, accusative constituents tended to be introduced by an indefinite determiner whereas dative constituents were more frequently introduced by a definite determiner. This finding adds to the data reported by Weber & Müller (2004). Moreover, we found evidence that (at least for the order DAT < ACC) definite objects mostly preceded indefinite objects. This finding supports the Definiteness constraint. 4.2 Completion Questionnaire 3 A completion questionnaire was designed to investigate the influence of Case and Definiteness on the availability of syntactic structures. The systematic variation of the factors was expected to result in more data on double object structures. In his study, Featherston (no year) presented double object structures, manipulating Case and Definiteness of the objects. Subjects preferred definite constituents to precede indefinite ones, irrespective of Case. In Corpus Study 3, we found Case and Definiteness to be correlated, resulting in a prevalence of definite datives preceding indefinite accusatives. 4.2.1 Method Completion Questionnaire 3 was again presented as a paper and pencil questionnaire. Forty-eight native speakers of German participated in the experiment. They were not informed about the aims of the study before they had completed the questionnaire. Materials consisted of 36 experimental items and 36 filler items. The experimental items began with a definite nominative in the Vorfeld that was followed by the auxiliary wird 'will' which was followed by a second noun phrase. The nouns had


319

occurred in the sentence fragments from Completion Questionnaire 1 as well, referring to animate entities and displaying masculine gender. Definiteness (definite vs. indefinite) and Case (dative vs. accusative) of the second noun phrase were varied. Examples are given in (9) and (10). (9)

Der Doktor wird dem/den Krankenpfleger .... theNOM doctor will theDAT/ACC nurse ... (10) Der Doktor wird einem/einen Krankenpfleger.... nurse ... theNOM doctor will aDAT/ACC

Sentence fragments were distributed over four lists according to a Latin square. Each item occurred once in each list, in different conditions across lists. Lists included an equal number of items per condition. Filler items were added. The lists were pseudorandomized, and each subject completed fragments of one list. 4.2.2 Results The completions were coded as to whether they resulted in a double object structure, a grammatical error, another syntactic structure or whether they were missing (cf. Completion Questionnaires 1 and 2). 100 90

84,1

83,9

80 70 60


57,3

56,5

50 38,3

40

35,9

30 20 10 0

13,8

13,5

6,3

4,7 0,0

nom-dat, def-def

0,3

nom-dat, def-indef

2,1

0,0

nom-acc, def-def

2,3

0,3

nom-acc, def-indef

Figure 4: Proportion of completions per Definiteness by Case Condition from Completion Questionnaire 3 (in %).

320


Figure 4 shows the relative proportion of completions per condition. The Case effect found in Completion Questionnaires 1 and 2 was replicated. Subjects produced less double object completions after an accusative than after a dative constituent. In addition, the accusative conditions yielded more errors than the dative conditions. No effect of Definiteness could be discerned. Double object completions were categorized as to whether the inserted object was definite or indefinite. Criteria were the same as in Corpus Study 3. Table 6 reports the number of double object completions per Case by Definiteness. definite before definite definite before indefinite indefinite before definite indefinite before indefinite



Table 6: Number of double object completions per Definiteness by Case Condition from Completion Questionnaire 3.

Data on double object structures with the accusative preceding the dative was again sparse. There was more data on double object structures with the dative before the accusative. They did not show an advantage of the definite before indefinite order over the indefinite before definite order. However, completions revealed a tendency to have exactly one indefinite object per sentences. Given a definite object, an indefinite one was added and vice versa. 4.2.3 Discussion Completion Questionnaire 3 tested the impact of Case and Definiteness on word order in double object structures. The Case effect found in Completion Questionnaires 1 and 2 was replicated. There was no evidence for a modulation of the Case effect by Definiteness. In double object completions, the order indefinite before definite was as frequently produced as the order definite before indefinite. This pattern conflicts with the Definiteness constraint. However, the data from Corpus Study 3 favour the Definiteness constraint. This seeming contradiction may be resolved. Sentences from the corpus were produced in the context of newspaper articles whereas the completions were elicited out of the blue. We propose that the Definiteness constraint holds for sentences produced in context only.


321

4.3 Reading Experiment 2 Corpus Study 3 and Completion Questionnaire 3 investigated the impact of Definiteness and Case on word order variation. Both yielded results that were in line with the Dative constraint, but the completion data cast some doubts on the applicability of the Definiteness constraint to sentences produced in isolation. In the current section, a self-paced reading experiment (cf. Reading Experiment 1) will be reported in which Case and Definiteness of the objects in double object sentences were manipulated. As far as we know, Reading Experiment 2 is the first experiment that tested Definiteness effects on word order preferences for full noun phrases on-line. 4.3.1 Method Forty-seven subjects (17 male and 30 female) participated in this experiment. All had German as their native language and reported normal or corrected-to-normal vision. None had participated in Reading Experiment 1 and all were naive with respect to the purpose of the study. There were 32 experimental double object sentences (cf. Reading Experiment 1) with an auxiliary in second and the main verb in final position. The definite nominative subject occupied the Vorfeld, while the two objects were in the Mittelfeld. All referents were animate. Case marking of the nominal constituents was unambiguous. Due to the small number of ditransitive verbs that allow for animate referents only, each verb appeared twice. Nouns were not repeated across sentences. The determiner of the first object was always indefinite. Case of the first object (dative or accusative) and Definiteness of the second object (definite or indefinite) were varied. The resulting word orders either satisfied both, the Case and the Definiteness constraint (cf. 11), or they violated only one of them (cf. 12 and 13), or both (cf. 14). (11) Der

Doktor wird einem Krankenpfleger einen Rollstuhlfahrer zeigen. aACC wheel-chair user point out to theNOM doctor will aDAT nurse The doctor will point a wheel-chair user out to a nurse. (12) Der Doktor wird einem Krankenpfleger den Rollstuhlfahrer zeigen. theACC wheel-chair user point out to theNOM doctor will aDAT nurse The doctor will point the wheel-chair user out to a nurse. (13) Der Doktor wird einen Krankenpfleger einem Rollstuhlfahrer zeigen. aDAT wheel-chair user point out to theNOM doctor will aACC nurse The doctor will point a nurse out to a wheel-chair user.

322


(14) Der Doktor wird einen Krankenpfleger dem Rollstuhlfahrer zeigen. theDAT wheel-chair user point out to theNOM doctor will aACC nurse The doctor will point a nurse out to the wheel-chair user.

There were 32 single object filler sentences that resembled the experimental sentences up to and including the first object and that served to obscure the lengths of the sentences. In addition, there were 28 filler sentences of different syntactic structure. Each sentence was followed by a content question. Content questions after the experimental items were case-critical. Sentences were presented word by word in a central window for self-paced reading. Subjects were instructed to push a button as soon as they were ready to read the next word. To answer the content questions they had to push a yes or a no button. 4.3.2 Results Trials for which the content question was incorrectly answered were discarded. Subject- and condition-specific error rates were calculated. The data from subjects who performed badly in answering the content question (more than three missing values in a cell) or who showed extremely long reading times (individual reading times twice as long as mean reading times) was excluded from further analyses. This held for 15 subjects. The data from the remaining 32 subjects entered the analysis. In the next step, reading and decision times that exceeded two standard deviations of the subject- and condition-specific mean were discarded. 800 750 nom-dat-acc, def-indef-indef nom-dat-acc, def-indef-def nom-acc-dat, def-indef-indef nom-acc-dat, def-indef-def

700 650 600 550 500 450 400 350 Det1 N1 Aux Det2 N2 Det3 N3

V

Figure 5: Mean reading times per Definiteness by Case Condition from Reading Experiment 2 (in ms).


323

Mean reading times per condition are depicted in Figure 5 for the first determiner (Det1), the first noun (N1), the auxiliary (Aux), the second determiner (Det2), the second noun (N2), the third determiner (Det3), the third noun (N3), and the subcategorizing verb (V). Analyses of variance were calculated. Mean reading times per condition from the first determiner to the second noun did not differ significantly. On the determiner of the second object, the analyses of variance revealed a main effect of Case that was significant in the item-specific analysis only (F1 (1, 31) = 2.19, p > .10; F2 (1, 31) = 6.49, p < .05), a marginally significant main effect of Definiteness (F1 (1, 31) = 4.17, p = .050; F2 (1, 31) = 3.20, p = .083), and a significant interaction of the factors (F1 (1, 31) = 5.79, p < .05; F2 (1, 31) = 8.92, p < .01). The Scheffé test showed that reading times in the condition nom-acc-dat, def-indefindef were significantly longer than in all other conditions (p < .05). On the noun of the second object, the factor Case yielded a significant effect in the item-specific analysis only (F1 (1, 31) = 3.43, p = .073; F2 (1, 31) = 13.78, p < .01), the factor Definiteness showed a significant effect in the subject-specific analysis only (F1 (1, 31) = 4.98, p < .05; F2 (1, 31) < 1), and the interaction of the factors did not approach significance at all. On the sentence-final main verb, there was no significant main effect of Case or Definiteness, but an interaction of the factors that was significant in the subject-specific analysis only (F1 (1, 31) = 5.35, p < .05; F2 (1, 31) = 3.39, p = .075). The Scheffé test identified significantly longer reading times in the condition nom-acc-dat, def-indef-indef than in the condition nom-acc-dat, def-indef-def (p < .05). Finally, accuracy in the content question was analysed. Table 7 specifies the error rates per condition. indefinite before definite indefinite before indefinite

dative before accusative 24.8 18.6

accusative before dative 30 31.9

Table 7: Mean error rates per Definiteness by Case Condition from Reading Experiment 2 (in %)

The analysis of variance revealed a significant main effect of Case (F1 (1, 31) = 5.20, p < .05; F2 (1, 31) = 6.95, p < .05) with more errors after the order accusative before dative than after the order dative before accusative, no significant main effect of Definiteness and no significant interaction of the factors.

324


4.3.3 Discussion The Case effect found in Reading Experiment 1 was replicated for error rates in Reading Experiment 2. However, this time there was no main effect of Case during on-line processing of double object sentences with indefinite determiners. The interaction of the factors Case and Definiteness was significant on the second determiner, instead. This effect could be traced back to longer reading times in the accusative before dative condition with two indefinite objects. We propose that this is an information structural effect. The two indefinite determiners signal neutral focus which is incompatible with the marked object order (cf. Abraham 1994; Haider & Rosengren 1998). Moreover, there was no main effect of Definiteness. Even though the order indefinite before definite violates the Definiteness constraint, this object order did not invoke longer reading times. In the accusative before dative condition, two indefinites were even dispreferred as compared to the order indefinite before definite. It should be noted that a potential interference by a length confound (einem is longer than dem) does not seem probable because the effect persisted on the noun and because there was no similar difference in reading times in the dative before accusative condition (where einen is longer than den as well). Thus, Reading Experiment 2 brought no evidence of the Definiteness constraint. Future reading time experiments that present sentences in contexts instead of in isolation may help to shed further light on the impact of Definiteness on sentence processing. 4.4 Evidence for the Definiteness constraint To summarize, corpus counts on sentences produced in context favour the Definiteness constraint whereas the data from a completion questionnaire and a self-paced reading experiment on sentences presented out of context speaks against its validity for object order in German. This suggests a reformulation of the Definiteness constraint in terms of context. Moreover, corpus data indicated that Definiteness and Case are correlated. Dative objects in double object structures are mostly realized with a definite determiner whereas accusative objects tend to be realized with an indefinite determiner. In addition, reading times increased when the two objects were indefinite and the Dative constraint was violated. This suggests that information structural factors are relevant for Definiteness. Maybe the Focus constraint might succeed in


325

accounting for corpus and behavioural data. Further studies will help to shed light on these issues.

5 Conclusions The validity of linguistic constraints on German word order variation was investigated in a series of corpus queries, completion questionnaires, and self-paced reading experiments. The constraints tested were the Dative constraint, which states that datives tend to precede accusatives, the Animacy constraint, which states that constituents referring to animate entities tend to precede constituents referring to inanimate referents, and the Definiteness constraint, which states that definite constituents tend to precede indefinite ones. The scope of these constraints was restricted to the relative order of the two objects in double object sentences with the subject in the Vorfeld. The validity of the Dative constraint was confirmed. Datives tended to precede accusatives even if Animacy as a potentially interfering factor is controlled. The impact of Animacy on the order of the objects was not as easily ascertained. It turned out that Animacy and Case were indeed not independent of each other because datives typically referred to animate entities and accusatives to inanimate entities. The inverse distribution of Animacy over datives and accusatives was rarely found in double object structures in the corpus data. Consequently, it became evident that it will not be practicable to disentangle these effects in an experiment with plausible materials. Finally, the validity of the Definiteness constraint irrespective of contextual factors has to be questioned. Only in the corpus sample, there was evidence for a tendency of definite objects to precede indefinite ones. Structures violating the Definiteness constraint were produced rather frequently in a completion questionnaire, and they did not seem to induce processing difficulties in a self-paced reading experiment. Moreover, there was evidence for an association of Case and Definiteness in the corpus. Datives were mostly introduced by definite determiners whereas accusatives were mostly introduced by indefinite determiners. Some of the observed patterns suggest information structural effects on word order variation. These might better be captured by a Focus constraint stating that background information tends to precede new information.

326


To summarize, the results reported here suggest that the Dative constraint and the Animacy constraint are both valid but that they cannot be disentangled and that they must be ranked together. In contrast, the Definiteness constraint does not account for object order as it stands. It might be replaced by an information structural constraint taking contextual factors into account. To decide on this, further empirical evidence is needed. The association of Case, Animacy, and Definiteness raises the question of a common underlying factor. Lamers and de Hoop (2005) suggest that there is one underlying constraint referring to Prominence that governs the realization of objects. The data reported here indicate that a precedence rule stating that prominent constituents (i.e., definite dative constituents with animate referents) tend to precede non-prominent constituents (i.e., indefinite accusative constituents with inanimate referents) might account for the prevalent object order pattern. However, this assumption does not really provide an explanation for the impact of prominence on processing. Kempen and Harbusch (2003b) propose that definite and animate referents tend to be conceptualized earlier than indefinite and inanimate ones. Incremental production then results in the realization of be it prominent constituents before non-prominent ones. A subsequent explanation for parallel effects in comprehension might be found in the Tuning Hypothesis according to which parsing preferences depend on the pattern of statistical prevalence in linguistic input (Cuetos, Mitchell & Corley 1996). Finally, it must be stressed that in this study empirical evidence was given for the relative order of the objects in double object structures only. Consequently, the validity of the discussed constraints cannot be estimated for the relative order of other constituents and the conclusions drawn here do not necessarily apply to alternative structures.

References Abraham, W. (1994): Fokusgrammatik und Indefinitheit. In: B. Haftka (ed.): Was determiniert die Wortstellungsvariation? Studien zu einem Interaktionsfeld von Grammatik, Pragmatik und Sprachtypologie. Opladen: Westdeutscher Verlag, 273–245. Bader, M. (1994): Syntactic-function ambiguities. Folia Linguistica 28, 5–66. Bresnan, J., A. Cueni, T. Nikitina & R. H. Baayen (in press): Predicting the dative alternation. Proceedings of the Royal Netherlands Academy of Science Workshop on Foundations of Interpretation.


327

Büring, D. (2001): What do definites that indefinites definitively don't? In: C. Féry & W. Sternefeld (eds): Audiatur vox sapientiae: A Festschrift for Achim von Stechow. Berlin: Akademie Verlag, 70–100. Cuetos, F., D. C. Mitchell & M. M. B. Corley (1996): Parsing in different languages. In: M. Carreiras, J. García-Albea & Núria Sebastian-Gallés (eds): Language processing in Spanish. Mahwah, NJ: Erlbaum. Dowty, D. (1991): Thematic proto-roles and argument selection. Language, 67, 547–619. Featherston, S. (no year): Experiment on word order in the Mittelfeld. Retrieved 01. 11. 2005 from http://www.sfb441.uni-tuebingen.de/~sam/db/wotan.exp.html Gibson, E. (1998): Linguistic complexity: Locality of syntactic dependencies. Cognition 68, 1–76. Gries, St. Th. (2003): Towards a corpus-based identification of prototypical instances of constructions. Annual Review of Cognitive Linguistics 1, 1–27. Grimshaw, J. (1990): Argument structure. Cambridge, MA: MIT Press. Haider, H. (1993): Deutsche Syntax – generativ: Vorstudien zur Theorie einer projektiven Grammatik. Tübingen: Narr. Haider, H. & I. Rosengren (1998): Scrambling. Sprache und Pragmatik: Arbeitsberichte 49, 1–104. Hemforth, B., L. Konieczny & G. Strube (1993): Incremental syntax processing and parsing strategies. Proceedings of the 15th Annual Conference of the Cognitive Science Society. Hillsdale, NJ: Erlbaum, 539–545. Höhle, T. N. (1982): Explikationen für "normale Betonung" und "normale Wortstellung". In: W. Abraham (ed.): Satzglieder im Deutschen: Vorschläge zur syntaktischen, semantischen und pragmatischen Fundierung. Tübingen: Narr, 75–153. Jacobs, J. (1988): Probleme der freien Wortstellung im Deutschen. Sprache und Pragmatik: Arbeitsberichte 5, 8–37. Keller, F. (2000): Gradience in grammar: Experimental and computational aspects of degrees of grammaticality. PhD thesis, University of Edinburgh. Kempen, G. & K. Harbusch (2003a): An artificial opposition between grammaticality and frequency: Comment on Bornkessel, Schlesewsky, and Friederici (2002). Cognition 90, 205–210. Kempen, G. & K. Harbusch (2003b): Word order scrambling as a consequence of incremental sentence production. In: H. Härtl & H. Tappe (eds): Mediating between concepts and grammar. Berlin: Mouton De Gruyter, 141–164. Kempen, G. & K. Harbusch (2004): A corpus study into word order variation in German subordinate clauses: Animacy affects linearization independently of grammatical function assignment. In: Th. Pechmann & Chr. Habel (eds): Multidisciplinary approaches to language production. Berlin: Mouton De Gruyter, 173–181.

328


Kempen, G. & K. Harbusch (2005): The relationship between grammaticality ratings and corpus frequencies: A case study into word order variability in the midfield of German clauses. In: St. Kepser & M. Reis (eds): Linguistic evidence: Empirical, theoretical, and computational perspectives. Berlin: Mouton De Gruyter, 229-249. Lamers, M. & H. de Hoop (2005): Animacy information in human sentence processing: An incremental optimization of interpretation approach. In: H. Christiansen, P. R. Skadhauge, & J. Villadsen (eds): Constraint Solving and Language Processing. Berlin: Springer, 158–171. Lenerz, J. (1977): Zur Abfolge nominaler Satzglieder im Deutschen. Tübingen: Narr. Löbner, S. (1985): Definites. Journal of Semantics 4, 279–326. Müller, G. (1999): Optimality, markedness, and word order in German. Linguistics, 37, 777–818. [Negra2] http://www.coli.uni-sb.de/sfb378/negra-corpus/ Pappert, S., J. Schließer, D. P. Janssen & Th. Pechmann (2005): Word order effects in production and comprehension. Poster. Conference on Architectures and Mechanisms for Language Processing, Gent. Pechmann, Th., H. Uszkoreit, J. Engelkamp & D. Zerbst (1996): Wortstellung im deutschen Mittelfeld: Linguistische Theorie und psycholinguistische Evidenz. In: Chr. Habel, S. Kanngießer & G. Rickheit (eds): Perspektiven der Kognitiven Linguistik. Wiesbaden: Westdeutscher Verlag, 257–299. Primus, B. (1999): Cases and thematic roles: Ergative, Accusative and Active. Tübingen: Niemeyer. Rösler, F., Th. Pechmann, J. Streb, B. Röder & E. Hennighausen (1998): Parsing of sentences in a language with varying word order: Word-by-word variations of processing demands are revealed by event-related brain potentials. Journal of Memory and Language 38, 150–176. [Tiger] http://www.ims.uni-stuttgart.de/projekte/ TIGER/ Weber, A. & K. Müller (2004): Word order variation in German main clauses: A corpus analysis. Proceedings of the 20th International Conference on Computational Linguistics, Geneva. Uszkoreit, H. (1986): Constraints on order. Linguistics 24, 883–906. Uszkoreit, H. (1987): Word order and constituent structure in German. Stanford, CA: Center for the Study of Language and Information. VanNice, K. & R. Dietrich (2003): Task-sensitivity of animacy effects: Evidence from German picture descriptions. Linguistics 41, 825–849. [Verbmobil] http://verbmobil.dfki.de/