2.1 Argument animacy

The animacy of a noun’s referent is a semantic (and possibly conceptual) property that is expressed differently across languages. Animacy is reflected, for example, in the choice of interrogative pronouns (English who or German wer for animates, but English what or German was for inanimates), number marking (Croft 1990; Corbett 2000; Haspelmath 2013) or case marking (in Differential Object Marking languages, in which morphologically overt object marking is obligatory for one semantic class of objects, but not for another one – the semantic property requiring overt marking often being animacy or humanness; see, e.g., Bossong 1985; 1991; Næss 2004). Based on observations from corpus linguistics and typology (Silverstein 1976; Bossong 1985; Comrie 1989; Dixon 1994; Aissen 2003; Jäger 2004; Dahl 2008; Malchukov 2008), sentences with an animate subject and an inanimate object are often considered the “most natural” transitive constructions in psycholinguistics.

In sentence production and comprehension, argument animacy is known to play a crucial role. In production, argument animacy influences word order and the assignment of syntactic function (Bock & Warren 1985; Bock 1987; Bock & Loebell 1990; McDonald et al. 1993; Ferreira 1994; Branigan & Feleki 1999; Prat-Sala & Branigan 2000; Branigan et al. 2008), interacting with processes of thematic role assignment (Ferreira 1994; van Nice & Dietrich 2003, see Verhoeven 2009; 2014 for the interplay with other factors like case marking in a variety of languages).

In sentence comprehension research, argument animacy has been used even earlier (e.g., Frazier & Rayner 1982) to assess the relative timecourse of the processing of semantic and syntactic information, with mixed results. While some studies support syntax-first approaches to parsing, with semantic information like animacy only being used during second-pass parsing (e.g., Ferreira & Clifton 1986; Clifton et al. 2003), other studies show that argument animacy is processed immediately, and influences structure building from the beginning of the parsing process (MacDonald et al. 1994; Trueswell et al. 1994). Over time, argument animacy manipulations have turned out to be different from many other semantic properties, in that contrasts in subject and object animacy seem to play an important role in the assignment of subject and object roles. In general, sentences with animate objects increase processing costs relative to sentences with inanimate objects, especially so for structures that are challenging to process, like relative clauses or ungrammatical clauses (Trueswell et al. 1994; Weckerly & Kutas 1999; Frisch & Schlesewsky 2001, but see Paczynski & Kuperberg 2011; Czypionka 2014; Czypionka et al. 2017, for object animacy effects with comparatively simple, grammatical sentences). For German, Frisch & Schlesewsky (2001) monitored the comprehension of embedded verb-final transitive sentences with inanimate and animate objects, using ERP measurements. Both arguments were marked nominative, making the sentences ungrammatical in German. With animate-inanimate argument sequences, they found a P600 on the clause-final verb. With animate-animate sequences, they found an N400 in addition to the P600, although the verbs’ semantic selectional restrictions concerning animacy were not violated. The authors suggest that the assignment of thematic roles can proceed either via morphosyntactic information or via argument animacy contrasts, and that the N400 arises only when neither is possible.

In EEG studies on the comprehension of English transitive sentences, object animacy has also been shown to affect processing, even in the absence of violations. For grammatical sentences without violations of semantic selectional restrictions, Paczynski & Kuperberg (2011) report an enhanced N400 component at centroposterior sites for postverbal animate compared to inanimate objects. The authors argue that this finding suggests that animacy effects in sentence comprehension are not caused by the higher conceptual accessibility of animate objects (as this should have attenuated, not enhanced, an N400). In their second experiment, Paczynski & Kuperberg (2011) find no difference for animate objects that are assigned patient and experiencer roles, suggesting that animacy effects are not caused by semantic matches or prototypicality (as this would have attenuated an N400 for animate experiencers, but not for atypically animate patients), at least in SVO languages like English. The authors interpret their findings as showing that animacy hierarchies are mapped directly on syntactic hierarchies during sentence comprehension.

The special role of argument animacy in sentence comprehension is further supported by observations collectively referred to as semantic illusions (see Hoeks et al. 2004, for Dutch, and Kuperberg et al. 2003; Kim & Osterhout 2005; Kuperberg et al. 2007, for English, among others). This term refers to the results of a number of studies monitoring the effects of different types of semantic violations of verb-argument-relationships. The general pattern reported in these studies is the following: For subject-verb relations that violated world knowledge and for weak lexico-semantic associations between subjects and objects, an N400 is reported. However, for violations of subject animacy restrictions, no N400, but a P600 is reported. The animacy restriction violation even causes a P600, but no N400, when there is no close lexico-semantic relationship between subject and verb (Kuperberg et al. 2007). These findings lend further support to the idea that argument animacy contrasts are used to assign subject-object relationships, and are processed in a different way from other types of semantic processing like world knowledge or lexico-semantic associations.

Findings like the ones outlined above have informed different models of sentence comprehension that focus on the assignment of grammatical and thematic roles. The comprehension model proposed by Kuperberg (Kuperberg 2007) distinguishes between semantic and formal processing pathways. Two formal processing pathways operate in parallel, one using morphosyntactic information like subject-verb-agreement, the other using lexical-thematic information like argument animacy. According to Kuperberg, processing difficulties in the semantic pathway are reflected in enhanced N400 components, whereas processing difficulties in one of the formal pathways are reflected in enhanced P600 components. Therefore, argument animacy is one of the formal cues for sentence comprehension, although semantic (and possibly conceptual) in nature. In the extended Argument Dependency Model, (eADM, Bornkessel-Schlesewsky & Schlesewsky 2006; 2009; 2013), animacy is one of the contributors that help calculate a potential argument’s prominence in the “compute prominence” step. Increased processing load during this step is linked to an enhanced N400, whereas the P600 is related to processing steps involving agreement and syntactic reanalysis.¹

In sum, argument animacy is one of the relevant cues in sentence comprehension, and the processing of argument animacy contrasts is different from other types of semantic processing (like, e.g., the checking of semantic appropriateness, see Paczynski & Kuperberg 2011, outlined above). Animate-animate argument sequences cause a higher processing load than animate-inanimate sequences, usually reflected in enhanced N400 components.

In the following, we will refer to the increased processing cost of two animate arguments relative to animate-inanimate argument combinations as the object animacy effect. We will use this well-established effect as a tool to understand more about the processing of different types of verbs in German, and to better understand another factor in the comprehension of German sentences, namely, case marking.

2.2 Case marking

In German, case is morphologically marked on arguments. Due to widespread case syncretism, many nouns and some pronouns are case-ambiguous, so that case-ambiguous, but grammatical sentences are still possible. In the comprehension of German sentences, case marking is one of the main cues for assigning grammatical roles: Subjects are marked with nominative, direct objects are marked with accusatives, and indirect objects are marked with dative. However, a small group of German two-place verbs assign dative instead of nominative to their direct objects. In the following, we will refer to these verbs assigning noncanonical case as lexical case marking verbs or NOM-DAT verbs; the verbs assigning canonical case will be called structural case marking verbs or NOM-ACC verbs.

Lexical case marking verbs occur in different case-marking languages, and encode a special type of argument semantics (Blume 2000): The situations encoded by German NOM-DAT verbs are not prototypically or maximally transitive (Dowty 1991; Grimm 2010). This means that they do not describe situations with a prototypical agent doing something to a prototypical patient; the semantic features associated with the proto-agent role (in the sense of semantic properties contributing to these roles, see Dowty 1991) are usually more evenly distributed among the participants. A rough way of putting this would be to say that their objects in general are more agentive relative to their subjects than in prototypically transitive situations.² It is important to mention that although the role of animacy is discussed in this part of the literature, animacy is not included in the list of semantic properties making up the agent and patient proto-roles in Dowty’s account, and neither in the set of properties defining prototypical agents in Grimm (2010). Some of the semantic properties that contribute to Dowty’s agent proto-role strongly imply animacy (e.g., sentience and volitional involvement). However, the semantic property of affectedness, which is associated with the patient proto-role, could also be argued to be linked to animacy. Therefore, it cannot be said that lexical case marking signals a preference for animate objects; if anything, it signals a deviation from prototypical transitivity (which only roughly matches an animate-inanimate argument pattern).

While there is no consensus on the syntactic structure of NOM-DAT verbs, many analyses suggest that their structure is different from NOM-ACC verbs. Dative for direct objects is considered to be an idiosyncratic or lexical case (Haider 1993; Czepluch 1996; Haider 2010; Woolford 2006) (in contrast to dative for indirect objects, see Woolford 2006; but see Meinunger 2007 for a different position outlined above). Some analyses hold that lexical datives need to be assigned by lexical heads like V⁰ (see Fanselow 2000 for lexical case marking verbs, and Woolford 2006 for datives assigned by lexical case marking verbs and prepositions). Another suggestion is that dative is always assigned in the position of the indirect object, even for two-place verbs; however, this approach assumes that there are different base positions for indirect objects, depending on the verb (McFadden 2004). An approach that has already been tested psycholinguistically is the proposal that lexical case is assigned in an additional syntactic layer of projection called KP (for Kase Phrase, Bader et al. 2000; Bayer et al. 2001); outlined in more detail below).

In sentence comprehension, NOM-DAT verbs cause increased processing cost compared to NOM-ACC verbs, but only when morphological case marking on the arguments is ambiguous (see Bader et al. 2000 for data from speeded grammaticality judgments). In ERP studies on German sentence comprehension, the reassignment of dative instead of structural case is reflected in an enhanced N400 to clause-final NOM-DAT verbs relative to NOM-ACC verbs (Hopf et al. 1998; 2003), again only when case marking on the arguments is morphologically ambiguous.

Another effect of lexical case marking is that NOM-DAT verbs license noncanonical word orders. In contrast to NOM-ACC verbs, NOM-DAT verbs come with different unmarked argument orders; some occurring with unmarked subject-object, others with unmarked object-subject word orders.³ Judgment studies show that while object-subject-structures cause garden path effects with accusative objects, these processing difficulties are attenuated for dative objects (Bader 1996). These findings are further supported by EEG evidence reported by Bornkessel et al. (2004). For NOM-ACC verbs, the authors found a P600 for OS word orders compared to SO word orders. The same contrast elicited no P600, but an N400 for NOM-DAT verbs with unmarked SO word order. Although no direct comparison between NOM-ACC and NOM-DAT verbs was offered in this study, these findings support the idea that NOM-DAT case marking licenses noncanonical word orders, irrespective of the argument order associated with the specific NOM-DAT verb.

The processing of NOM-ACC and NOM-DAT verbs differs in another respect, namely, the processing of argument animacy. In Czypionka et al. (2017), grammatical subordinate clauses with SOV word order showed an effect of object animacy in various behavioral and ERP measurements. ERPs to the clause-final verb were more negative-going with animate than with inanimate objects at left-posterior sites. This contrast was only visible with NOM-ACC verbs. With NOM-DAT verbs, there was no contrast between animate and inanimate objects, and waveforms for dative verbs were more negative-going than the inanimate-accusative condition, irrespective of the objects’ animacy.

Case marking plays an important part for assigning grammatical and thematic roles in models of sentence comprehension; however, the underlying assumptions about structural and lexical case marking verbs differ between accounts. The eADM (Bornkessel-Schlesewsky & Schlesewsky 2006; 2009; 2013) predicts a processing difference between NOM-ACC and NOM-DAT verbs: “(…) constructions including a dative argument should give rise to processing behaviour that is measurably distinct from that observable for nominative-accusative structures” (Bornkessel-Schlesewsky & Schlesewsky 2006: 792). The explanations for this predicted processing difference center around the special semantic behavior of NOM-DAT verbs. While the accusative arguments of NOM-ACC verbs must always be mapped on the “undergoer” role, this relationship is more flexible with NOM-DAT verbs (Bornkessel-Schlesewsky & Schlesewsky 2009: 43–44). It is assumed that the objects of NOM-DAT verbs are not associated with a specific type of thematic role or set of semantic properties (“GR”s in the eADM), and neither with a specific structural position.⁴ In its current state, the model does not make any more specific predictions for the type of difference to be found between NOM-ACC and NOM-DAT verbs, and does not go into detail about possible effects of different types of dative-assigning structures on sentence comprehension measures.

Another proposal about the processing of NOM-ACC verbs assumes that upon encountering a NOM-DAT instead of a NOM-ACC verb, two additional processing steps are performed (Bader et al. 2000; Bayer et al. 2001). The first step is a mild reanalysis of the syntactic structure to insert an additional projection where lexical case can be assigned to the object (KP; Bayer et al. 2001 – unlike assumed in the eADM, there is a specific syntactic position for objects of NOM-DAT verbs in this proposal). The second is a reaccess to the lexical entry of the preverbal object to check if the morphology licenses lexical case (i.e., dative). It is this latter step that is assumed to cause an N400, as reported in Hopf et al. (1998; 2003). While this proposal does not tie in with any broader model of sentence comprehension, it makes falsifiable predictions for contrasts between NOM-ACC and NOM-DAT verbs, and takes into account syntactic differences between both verb classes.

In sum, the reassignment of lexical instead of structural case enhances processing load. This additional processing load is reflected in more negative-going waveforms, mostly an enhanced N400, rather than an enhanced P600 (as could be expected for a morphosyntactic process like case assignment to objects). At the same time, NOM-DAT verbs seem to license noncanonical word orders, and seem to attenuate object animacy effects.

What is it that makes the processing of NOM-DAT verbs special? Specifically, what is it that attenuates the object animacy effect with NOM-DAT verbs? Different possible lines of explanation could account for the case marking effects reported in the psycholinguistic literature, and more specifically for the absence of object animacy effects with dative verbs:

1. The first is centered around the non-prototypically transitive semantics of NOM-DAT verbs. These verbs encode transitive situations where the arguments do not have the prototypically transitive distribution of semantic properties (e.g., objects that can be volitional or agentive because they are animate; or subjects that are affected or moved by the action denoted by the verb). It is possible that the non-prototypically transitive semantics associated with the NOM-DAT case marking pattern license the processing of noncanonical argument animacy patterns, i.e., the processing of two animate arguments, thereby attenuating the object animacy effect, irrespective of the specific verb’s semantic selectional restrictions (roughly parallel to the licensing of OS word orders by NOM-DAT verbs, even those with unmarked SO word orders). This explanation is in line with the eADM and its underlying assumptions for the processing of NOM-DAT verbs (Bornkessel-Schlesewsky & Schlesewsky 2006; 2009).

2. The second is centered around the syntax of NOM-DAT verbs: Lexical case marking effects in general, and the interaction of lexical case marking and object animacy effects in particular, could be caused by the syntactic differences between NOM-ACC and NOM-DAT verbs. The effects of encountering a NOM-DAT verb (rebuilding the syntactic structure, rechecking the dative morphology on the object) override the measurable effects of the processing of animacy contrasts; not because of an interaction between the verb’s semantics and the object animacy, but simply because of the increased processing difficulty. This explanation is in line with the KP-based account of NOM-DAT verb processing (Bader et al. 2000; Bayer et al. 2001).

Currently, the studies cited above do not allow us to distinguish between these two explanations. To disentangle the semantic and syntactic contributions to lexical case marking effects, and to find an explanation for the interaction of lexical case marking and object animacy effects, we will use another distinction between different types of German verbs: simple and particle verbs. Both simple and particle verbs come with NOM-ACC and NOM-DAT case marking patterns. For simple verbs, NOM-DAT verbs signal non-standard argument semantics, and additional syntactic and lexical processing load when compared to NOM-ACC verbs. For particle verbs, we assume that NOM-DAT verbs also have non-standard argument semantics, but not necessarily higher syntactic processing load than NOM-ACC particle verbs.

In the following, we will give a short overview of our assumptions about case marking with particle verbs. After this overview, we will present our research question again, in more detail than above, outlining how the contrast between simple and particle verbs can help us to better understand the processing differences between NOM-ACC and NOM-DAT verbs.

2.3 Particle verbs

In addition to their case marking pattern, German verbs can be distinguished by their separability. Non-separable verbs (like simple folgen, ‘to follow’, or prefixed verfolgen, ‘to pursue’) are realized as the second constituent in main clauses, and in sentence-final position in subordinate clauses. Separable or particle verbs (nachlaufen, ‘to run after’) consist of a particle (nach), often homologous to a preposition, and a base (laufen). In subordinate clauses, particle verbs are realized as one word in the clause-final position: … dass Peter dem Hund nachläuft (‘… that Peter pursues the dog’; literally: … that Peter the dog after-runs). In main clauses, however, the particle and the base are split. Only the base is moved to the second constituent position in the clause, while the particle remains behind in the sentence final position: Peter läuft dem Hund nach (‘Peter pursues the dog’; literally: Peter runs the dog after). Particle verbs thus are compounds that are potentially realized as multiple words in the sentence (McIntyre 2007), but share one lexical entry (Jackendoff 2002; Cappelle et al. 2010). In the following, we will refer to separable verbs as particle verbs, and to simple and prefix verbs as simple verbs for the sake of readability. (See Olsen 1996 for a definition of particle verbs, and Dehé 2015 for an overview of particle verbs in Germanic.) Particle verb formation is highly productive in German (see Smolka et al. 2014: 32), meaning that there is no exhaustive list of German particle verbs, since speakers can always make up new ones.

Currently, there is no single agreed-upon syntactic analysis of German particle verbs. Different analyses can roughly be divided into two groups (see McIntyre 2007, and, more recently, McIntyre 2015, for an overview of some of the descriptions). One group of analyses holds that the objects of particle verbs are not objects of the verb, but of the particle (see, e.g., Zeller 2001: 21, for a syntactic description of this latter view). Another view holds that the objects of particle verbs are objects of the verb or of a complex predicate that includes the particle and the verb, not of the particle alone (see, e.g., Neeleman & Weerman 1993; Neeleman 1994, for Dutch, Stiebels & Wunderlich 1994, for German). While these views differ in many important respects, all of them describe particle verbs with structures that are more complex than those of simple, structural case assinging verbs.

Just like simple two-place verbs, two-place particle verbs come with both canonical NOM-ACC and noncanonical NOM-DAT case marking patterns. Importantly, although many particles are homologous to prepositions, the case marking pattern of particle verbs cannot be reduced to case assignment by incorporated prepopositions (Zeller 2001: 219), for which case encodes spatial semantics, (see, e.g., Svenonius 2010), and neither to any case reliably assigned by a specific particle. The lexical differences between accusative and dative case marking seem to be the same for simple and particle verbs (e.g., lexical case like dative, but not structural case like accusative, being retained under passivization). The relationship between noncanonical case marking and non-prototypical semantic transitivity described above holds for both particle verbs and simple verbs; i.e., particle verbs assigning NOM-DAT are never prototypically transitive and never describe true agent-patient scenarios (Meinunger 2007).

In the current study, we will not adopt a specific syntactic analysis of particle verbs, and will not answer the question whether the objects of particle verbs are best described as objects of the base verb, of the particle, or of a combination of both.

We will make the following assumptions:

1. All particle verbs are morphologically and syntactically more complex than simple NOM-ACC assigning verbs.

2. Particle verbs assigning NOM-ACC and NOM-DAT do not differ from each other in syntactic complexity to the same extent as simple verbs assigning NOM-ACC and NOM-DAT differ from each other.⁵

3. The semantic differences between particle verbs assigning NOM-ACC and NOM-DAT are the same as for simple verbs, and, following the analysis of Meinunger (2007), NOM-DAT case assignment signals a deviation from prototypically transitive argument semantics.

We will discuss the implications of this assumption for sentence processing measurements below, when restating our research question.

Sentence comprehension research so far has mostly focused on the processing of separated base-particle combinations in Dutch and German (Urban 2001; 2002; Isel et al. 2005; Cappelle et al. 2010; Piai et al. 2013). An exception is the study by Roehm & Haider (2009), which directly compared the processing of nonseparable prefix verbs and complete particle verbs in the verb-second position of sentences. This study reports an N400 for particle verbs compared to prefix verbs at the position of the verb, suggesting enhanced processing cost for particle verbs. To our knowledge, no study so far has tested the influence of case marking patterns on the processing of particle verbs, and no model of sentence comprehension has made explicit predictions as to their processing.

To sum up the preceding paragraph, German two-place particle verbs come with canonical NOM-ACC and noncanonical NOM-DAT case marking patterns. Just like simple verbs, noncanonical case marking is a signal of non-prototypically transitive semantics with particle verbs. However, both NOM-ACC and NOM-DAT verbs are syntactically more complex than the standard structure of simple NOM-ACC verbs.

2.4 Research question

Because of their special syntactic and semantic characteristics, particle verbs allow us to disentangle semantic and syntactic contributions to lexical case marking effects in comprehension.

• For simple verbs, effects of the processing of NOM-DAT verbs in contrast to NOM-ACC verbs could reflect one or more of the following properties: (1) They reliably signal non-prototypically transitive argument semantics, (2) their syntactic structure is more complex than the standard structure of NOM-ACC verbs, (3) and, according to the KP-analysis/lexical reaccess hypothesis proposed by Bayer et al. (2001), they cause lexical reaccess to the object entry to check for dative morphology.

• For particle verbs, effects of the processing of NOM-DAT verbs in contrast to NOM-ACC verbs could reflect the fact that (1) they signal non-prototypically transitive argument semantics, and (2) possibly cause lexical reaccess to the object. They are unlikely to reflect differences in syntactic complexity, given that both NOM-ACC and NOM-DAT particle verbs differ morphosyntactically from the “standard” structure of simple NOM-ACC verbs.

It is important to note that none of the earlier studies on the processing of NOM-DAT verbs cited above (for ERP, see Hopf et al. 1998; Hopf et al. 2003; Czypionka et al. 2017; Bornkessel et al. 2004) distinguished between simple and particle verbs. Both the sets of NOM-ACC and NOM-DAT verbs in those studies consisted of a mix of simple verbs and particle verbs that appeared as one orthographic unit in the clause-final position. Therefore, these earlier findings could theoretically reflect the semantic, lexical and syntactic effects of NOM-DAT in contrast to NOM-ACC verb processing.

To disentangle the semantic and syntactic contributions to effects of lexical case marking, we will focus on one of the effects of lexical case marking, namely, the attenuated object animacy effect (Czypionka et al. 2017). As outlined above, animate objects have been shown to cause higher processing costs than inanimate objecs in transitive sentences. While this object animacy effect in general seems to be relatively stable, it is attenuated for NOM-DAT verbs – because of their special syntax, special semantics, or a mix of various factors.

Based on the previous studies, we expect a main effect of object animacy for NOM-ACC verbs, with more negative-going waveforms for animate than for inanimate conditions. This should reflect the slightly increased processing cost of animate-animate arguments sequences when compared to animate-inanimate argument sequences. Based on the semantic and syntactic properties of different types of verbs, we make the following predictions:

1. If the modulation of the object animacy effect for datives in previous studies was mainly due to the semantic differences between NOM-ACC and NOM-DAT verbs, or due to the lexical reaccess to the object necessary for NOM-DAT verbs, we predict an interaction of animacy and case marking for both simple verbs and particle verbs. Based on previous studies, we expect a reduced difference between animate and inanimate object conditions for NOM-DAT verbs compared to NOM-ACC verbs.

2. If, however, the source of the modulation in previous studies was mainly due to the increased syntactic processing cost caused by NOM-DAT verbs, we predict an interaction of animacy and case marking for simple verbs, but not for particle verbs. For particle verbs, we expect a main effect of object animacy that is not influenced by verbal case marking pattern.

In the following, we report the results of two EEG experiments. In the first experiment, we measure the interaction of the effects of object animacy and case marking patterns in sentences with simple verbs. In the second experiment, we measure the interaction of the same effects in sentences with particle verbs. For the sake of readability, the language material used in both experiments is described in detail in the next section, followed by the reports of the experiments.

3.1 Simple verb stimulus set

The simple verb stimulus set (used in Experiment 1) consisted of 144 sentences, with 36 items in four different conditions. In each sentence quartet, we paired either NOM-ACC or NOM-DAT simple verbs with either inanimate or animate objects. All critical sentences were grammatical, verb-final embedded sentences with SOV word order, surrounded by a matrix clause. The two arguments were bare plural NPs. Because of case syncretism, the argument NPs do not carry overt morphological case marking in their plural forms, so neither case marking nor subject-verb number congruency allowed grammatical role assignment. Importantly, the sentences with animate objects are theoretically ambiguous because both arguments are animate and could be interpreted as the grammatical subject of the verb. However, an interpretation of these sentences as OSV would result in a dispreferred, pragmatically marked word order. NOM-DAT verbs were selected from a list of German dative-assigning verbs (Meinunger 2007), chosing only verbs with standard linking patterns like folgen, ‘to follow’ (no object experiencer verbs like gefallen, ‘to please’). All NOM-ACC and NOM-DAT verbs semantically allow inanimate and animate objects, so that all conditions are syntactically and semantically well-formed. An adverb was inserted between the object NP and the critical verb.

We constructed 36 critical sentence quartets using 33 animate subject NPs, 33 inanimate and 29 animate object NPs, 22 NOM-ACC and 16 NOM-DAT verbs. Due to the limited number of German NOM-DAT verbs, we repeated some verbs with different subject and object NPs. Animate and inanimate object NPs in a sentence quartet were controlled for length (t (60) = 1.49, p > .1) and frequency (t (52.3) = –.35, p > .5; frequencies unavailable for five objects) according to the dlexDB corpus (Heister et al. 2011). NOM-ACC and NOM-DAT verbs in a sentence quartet were also controlled for length (t (74) = –.18, p > .8) and frequency (t (73) = 1.23, p > .2, frequency unavailable for eight verbs) according to the dlexDB corpus.⁶

A typical sentence quartet is given in Example 1. A full list of the simple verb stimulus set is given in the Supplementary Material.

3.2 Particle verb stimulus set

The particle verb stimulus set (used in Experiment 2) was constructed in parallel to the simple verb stimulus set, consisting of 144 sentences, with 36 items in four different conditions. The structure of the sentences was identical to the ones in the simple verb stimulus set. However, there was one important difference to the simple verb stimulus set: The NOM-ACC and NOM-DAT verbs were particle verbs, not simple verbs. The particle verbs were presented as one orthographic word in the clause-final position. The NOM-DAT particle verbs were selected from the same list of German dative-assigning verbs as the NOM-DAT simple verbs (Meinunger 2007).

We constructed 36 critical sentence quartets using 34 animate subject NPs, 36 inanimate and 28 animate object NPs, 25 NOM-ACC and 21 NOM-DAT verbs. Just like in the simple verb stimulus set, we repeated some particle verbs with different subject and object NPs. A typical sentence quartet is given in Example 2. A full list of the particle verb stimulus set is given in the Supplementary Material. Animate and inanimate object NPs in a sentence quartet were controlled for length (t (58.05) = –.86, p > .3) and frequency (t (51.3) = –1.19, p > .2; frequencies unavailable for ten objects) according to the dlexDB corpus (Heister et al. 2011). NOM-ACC and NOM-DAT verbs in a sentence quartet were controlled for frequency (t (42.75) = –.96, p > .3, frequency unavailable for one verb) according to the dlexDB corpus and for length (t (37.9) = –2.64, p < .05). The mean length of NOM-ACC assigning particle verbs was 9.1 letters, while the mean length of NOM-DAT assigning particle verbs was 10.7 letters. We chose to accept this descriptively small, but statistically significant difference in length, mostly because there is only a limited choice of NOM-DAT verbs in German, and stimulus construction is further restricted by the need to find argument-verb combinations that are meaningful and acceptable for accusative and dative conditions.

Please note that assessing the frequency of verbs in German corpora is problematic because of the separability of German particle verbs (see, e.g., Smolka et al. 2014: 33). Since particles and base verbs are separated in main clauses (present, simple past and imperative), the frequencies of potential base verbs are overestimated, and the frequencies of particle verbs are underestimated in German corpora like dlexDB (Heister et al. 2011). This means that lemma frequencies for particle verbs are unrealistically low, and both lemma and surface frequencies for simple verbs are unrealistically high. Therefore, the frequencies of simple and particle verbs cannot be used for direct quantitative comparisons in a meaningful way (the same holds for measures that are influenced by frequencies). Comparisons between different types of simple verbs, and different types of particle verbs, however, can still be informative.

Example 1 Example of a typical sentence quartet with simple verbs. Note that case morphology is not marked overtly on the arguments.

Example 2 Example of a typical sentence quartet with particle verbs. Note that case morphology is not marked overtly on the arguments.

3.3 Sentence completion task

To better characterize our stimuli, and to be able to interpret the results of the subsequent EEG studies, we ran a sentence completion tasks to assess the relative likelihood of different object-verb combination patterns in sentences parallel to our stimulus material. The aim of this study was to monitor which kinds of objects are expected to occur with NOM-DAT and NOM-ACC verbs.⁷ This was necessary because the results of currently available data on object case and animacy (roughly: datives are animate more often than inanimate) cannot be transferred directly to the specific case of our stimuli.⁸

For the sentence completion task, participants saw the critical sentence printed until the final verb of the subordinate clause (i.e., minus the spillover region). The space of the object was left blank. The sentence completion task was prepared in two versions: For each sentence quartet, participants saw either the NOM-ACC or the NOM-DAT conditions. In each version of the task, participants saw 36 sentences from the stimulus set with non-separable verbs, and 36 sentences from the set with separable particle verbs, leading to overall 72 sentences, with 18 sentences in each combination of conditions. Unlike in the EEG experiment, the separable and non-separable stimuli were presented in the same task.

The sentence completion task was performed as a pen-and-paper task as part of a course assignment at Constance University. Participants were instructed to fill in the blank with the first word that came to mind and that would complete the sentence in a meaningful way. Completions were coded as either “inanimate”, “animate” or “not applicable” (when the completion was not an argument, or when the space had been left blank). 28 participants successfully completed the task. 3 participants were male. Mean age was 23.3 years (SD = 2.0).

Before data analysis, the responses coded as “not applicable’ were removed from the data set. The majority of the completions were inanimate or animate arguments (93% for simple verb conditions, 89% for particle verb conditions), suggesting that a second argument is natural in this position in our stimulus material. An overview of the completions per condition is given in Table 1.

The results of the sentence completion task were analyzed in R (R Development Core Team 2005) with a binomial generalized linear mixed model, using the packages lme4 (Bates et al. 2015, glmer function for binomial data) and LMERConvenienceFunctions (Tremblay & Ransijn 2015, summary function); the packages plyr (Wickham 2011) and reshape (Wickham 2007) were used for data preparation. For the first model, we specified the main effects and interactions of CASE and SEPARABILITY as fixed effects. Participants and items were specified as random intercepts. In addition, random slopes were defined for participants (main effects of CASE and SEPARABILITY) and items (main effects of CASE). A full table of the statistical results for the fixed effects of the first model (prepared using the summary function) is given in Table 2. There was a statistically significant main effect of CASE (p < .05), with more animate completions for dative verbs than for accusative verbs. There was no interaction of CASE and SEPARABILITY, suggesting that this effect of case marking did not differ systematically between simple and particle verbs.

Since the two stimulus sets were to be run in separate EEG experiments and analyzed separately, we performed planned comparisons for simple and particle verbs. The planned comparisons were performed with a second model. For the second model, we specified the main effect of CASE as fixed effect. Participants and items were specified as random intercepts, and random slopes were defined for participants and items (main effects of CASE). A full table of the results for the fixed effects of the second model is given in Table 3.

For simple verbs, there was a statistically significant main effect of CASE on the likelihood of choosing an animate object (p < .05). Overall, participants chose more animate than inanimate objects. This tendency was more pronounced for dative than for accusative conditions. For particle verbs, there was a statistically significant main effect of CASE on the likelihood of choosing an animate object (p < .05). Participants chose more animate than inanimate objects for dative conditions. For accusative conditions, there was no marked preference for animate or inanimate objects.

The general pattern for simple verbs and particle verbs fits the findings in the literature: there are more animate completions for dative than for accusative verbs, and more inanimate completions for accusative than for dative verbs. However, animate accusative objects seem to be more likely in our stimuli than in larger corpora, and inanimate datives still make up over 20% of the completions.

We will briefly return to the results of the sentence completion task in the General Discussion, when discussing possible influences of expectation-based parsing on the outcome of later EEG experiments.

4.2 Results

Mean answer accuracy was 83% (SD = 7%).

ERP results are reported for latency windows timelocked to the beginning of the presentation of the critical verb. Mapping views for selected time windows and curves for selected electrodes are given in Figures 1 and 2.

Figure 1

Experiment 1 (simple verbs): Grand average ERPs for selected electrode sites and difference maps are shown. Accusative conditions are depicted in the upper part, dative conditions are depicted in the lower part. Mean voltage difference maps (animate minus inanimate) for the marked time window from 400–600 ms are given on the left side. The electrodes selected for illustration are marked in the maps.

Figure 2

Experiment 1 (simple verbs): Grand average ERPs for selected electrode sites and difference maps are shown. Inanimate conditions are depicted in the upper part, animate conditions are depicted in the lower part. Mean voltage difference maps (dative minus accusative) for the marked time window from 1400–1700 ms are given on the left side. The electrodes selected for illustration are marked in the maps.

400–600 ms: There was a main effect of MEDIAL-LATERAL position (F (4,80) = 24.5, ɛ = .55, p < .05), ANTERIOR-POSTERIOR position (F (4,80) = 19.9, ɛ = .55, p < .05), an interaction of MEDIAL-LATERAL position and ANTERIOR-POSTERIOR position (F (16,320) = 8.3, ɛ = .42, p < .05) as well as an interaction of ANIMACY, CASE and MEDIAL-LATERAL position (F (4,80) = 3.3, ɛ = .66, p < .05).

To assess whether animacy effects were similar or different for accusative and dative conditions, we resolved the three-way interaction by calculating separate three-way ANOVAs for the two cases.

For the accusative conditions, there was a main effect of MEDIAL-LATERAL position (F (4,80) = 21.6, ɛ = .57, p < .05), ANTERIOR-POSTERIOR position (F (4,80) = 14.5, ɛ = .50, p < .05), an interaction of MEDIAL-LATERAL position and ANTERIOR-POSTERIOR position (F (16,320) = 6.2, ɛ = .47, p < .05), and an interaction of ANIMACY and MEDIAL-LATERAL position (F (4,80) = 4.2, ɛ = .62, p < .05) as well as an interaction of ANIMACY, MEDIAL-LATERAL position and ANTERIOR-POSTERIOR position (F (16,320) = 2.7, ɛ = .39, p < .05).

To resolve which positions drove the three-way interaction, we calculated separate two-way ANOVAs of ANIMACY and ANTERIOR-POSTERIOR position for each MEDIAL-LATERAL position, and separate two-way ANOVAs of ANIMACY and MEDIAL-LATERAL position for each ANTERIOR-POSTERIOR position. This ANOVA revealed the following effects for medial-lateral positions:¹⁰ lateral-right: ANIMACY (F (1,20) = 5.1, p < .0.5), ANTERIOR-POSTERIOR position (F (4,80) = 8.6, ɛ = .52, p < .001); midline: ANIMACY (F (1,20) = 7.0, p < .0.5), ANTERIOR-POSTERIOR position (F (4,80) = 7.7, ɛ = .60, p < .001); lateral-medial-left: ANIMACY (F (1,20) = 6.2, p < .0.5), ANTERIOR-POSTERIOR position (F (4,80) = 12.61, ɛ = .64, p < .001.

For anterior-posterior positions, this analysis revealed the following effects: medial-anterior: MEDIAL-LATERAL position (F (4,80) = 9.7, ɛ = .65, p < .001) and an interaction of ANIMACY and MEDIAL-LATERAL position (F (4,80) = 3.2, ɛ = .61, p < .05). t-tests for simple main effects of ANIMACY at individual electrode sites did not show significant effects of ANIMACY. medial: MEDIAL-LATERAL position (F (4,80) = 14.7, ɛ = .64, p < .001) and an interaction of ANIMACY and MEDIAL-LATERAL position (F (4,80) = 5.0, ɛ = .61, p < .01). Resolving the effect of ANIMACY for individual sites showed statistically significant effects at electrode sites Cz (t (20) = –2.4, p < .05), T7 (t (20) = 2.4, p < .05), and T8 (t (20) = 2.5, p < .05). medial-posterior: MEDIAL-LATERAL position (F (4,80) = 18.7, ɛ = .63, p < .001) and an interaction of ANIMACY and MEDIAL-LATERAL position (F (4,80) = 3.9, ɛ = .45, p < .05). t-test for simple main effects of ANIMACY at individual electrode sites did not show significant effects of ANIMACY. posterior: MEDIAL-LATERAL position (F (4,80) = 33.9, ɛ = .69, p < .001).

For the dative conditions, there was a main effect of MEDIAL-LATERAL position (F (4,80) = 18.8, ɛ = .46, p < .05), of ANTERIOR-POSTERIOR position (F (4,80) = 13.8, ɛ = .50, p < .05), and an interaction of MEDIAL-LATERAL position and ANTERIOR-POSTERIOR position (F (16,320) = 7.8, ɛ = .44, p < .05). No statistically significant effects including the factor ANIMACY were found for dative conditions.

The original four-way ANOVA showed an interaction of ANIMACY and CASE. As illustrated in the difference maps of Figure 1, this was most clearly visible at central electrodes. In the time window from 400 to 600 ms, waveforms for central electrodes were more negative-going for the animate-accusative than for the inanimate-accusative condition. In contrast, waveforms for the two dative conditions did not differ from each other, and ran close to the waveforms for the inanimate-accusative condition.

1400–1700 ms: In the time window from 1400–1700 ms after the onset of the verb (which corresponds to 500 ms after the onset of the first postverbal word in the spillover region, und), there was a main effect of ANTERIOR-POSTERIOR position (F (4,80) = 18.4, ɛ = .34, p < .05), an interaction of ANTERIOR-POSTERIOR position and MEDIAL-LATERAL position (F (16,320) = 2.9, ɛ = .46, p < .05) as well as an interaction of CASE and ANTERIOR-POSTERIOR position (F (4,80) = 3.4, ɛ = .46, p < .05). To reveal which of the anterior-posterior positions drove the case effect, we performed t-tests for amplitudes cumulated across medial-lateral positions. The effect of CASE was statistically significant for anterior-posterior positions ANTERIOR-MEDIAL (t (20) = 2.5, p < .05) and POSTERIOR (t (20) = –2.1, p < .05).

As seen in Figure 2, waveforms for dative conditions were more negative-going than for accusative conditions at left-anterior sites. Waveforms for animate and inanimate conditions did not differ visibly from each other. While the distinction between accusative and dative conditions looked slightly less pronounced for animate than for inanimate conditions, the interaction did not reach statistical significance.

4.3 Discussion Experiment 1

In Experiment 1, we found an interaction of object animacy and verbal case marking on the verb, and a main effect of case marking on the first postverbal word, for the simple verb stimulus set.

Starting around 400 ms after the presentation of the verb, and continuing for about 200 ms, waveforms for the animate-accusative condition were more negative-going that waveforms for the inanimate-accusative condition, most visibly so at central electrodes. We interpret this finding as reflecting the slightly enhanced processing load caused by the animate compared to the inanimate objects, i.e., an object animacy effect. This fits earlier findings reporting enhanced N400 amplitudes to sequences of two animate arguments when compared to animate-inanimate argument sequences (Frisch & Schlesewsky 2001; Paczynski & Kuperberg 2011). Interestingly, this object animacy effect is visible even in the absence of semantic or syntactic violations.

On the contrary, for NOM-DAT verbs, no difference is visible between animate and inanimate objects. Waveforms for inanimate-dative and animate-dative conditions ran close together, and close to the inanimate-accusative condition.

At a general level, the object animacy effect for accusative, but not dative verbs fits earlier findings (Czypionka et al. 2017). However, there are a number of important differences relative to the previous study. In the earlier study, the animacy effect surfaced as a longer lasting left-posterior negativity, while in the present study, negativities were constrained to latency windows close to those for an N400, and showed an N400-like topography. Another difference between the studies is that waveforms for the dative conditions were close to the animate-accusative condition in the earlier study, but close to the inanimate-accusative condition in the current study. A direct comparison of the outcomes of both studies would be difficult, since the stimuli of the earlier study mixed simple verbs and particle verbs, and therefore the stimuli were morphosyntactically more heterogenous than the ones in the current study. The main pattern of results, however, is similar: In both studies, there are object animacy effects for accusative, but not dative conditions. If anything, the outcome of the current study is closer to earlier findings on animacy effects of Frisch & Schlesewsky (2001) or Paczynski & Kuperberg (2011) than the results of Czypionka et al. (2017).

In the spillover region, 1400 ms after the onset of the verb, there was a main effect of case, with more negative-going waveforms for dative compared to accusative conditions in the left-anterior and some central electrodes. This effect has not previously been described for NOM-DAT compared to NOM-ACC verbs. Interestingly, no N400-like dative effect was found, as could have been expected from earlier ERP studies on the processing of lexical case in German (Hopf et al. 1998; Hopf et al. 2003; Bornkessel et al. 2004). These earlier studies reported an N400 starting already on the verb, either for the direct comparison of accusative and dative conditions, (Hopf et al. 1998; 2003), or for OS in contrast to SO word orders (while the same contrast elicited a P600 for accusative conditions, Bornkessel et al. 2004). Data from the word directly following the critical verb are not reported in any of these studies.

In general, a comparison of the current and earlier studies is difficult due to the many differences in the stimuli used. All of these three earlier studies used a syntactically heterogenous mix of simple and particle verbs, whereas our own stimuli in Experiment 1 only contained simple verbs. In addition, dative effects in earlier studies were elicited using very different grammatical constructions (e.g., following embedded relative clauses with an accusative-assigning verb, Hopf et al. 2003, or object-subject-sentences, Bornkessel et al. 2004), and not always discussed in a direct comparison of accusative- and dative assigning verbs (Bornkessel et al. 2004, where verbal case marking pattern was a between-subjects factor that was not statistically pursued).

At this point, we cannot offer a conclusive interpretation of the late effect of case marking in our data. Therefore, we limit ourselves to the tentative interpretation that this effect of case in our results reflects the additional workload caused by NOM-DAT verbs.

The fact that we found an interaction between ANIMACY and CASE in the latency window from 400–600 ms supports our initial hypothesis, suggesting that the effect of object animacy is indeed different for simple verbs assigning NOM-DAT than it is for simple verbs assigning NOM-ACC. As we outlined in the Introduction, there are different possible explanations for the underlying linguistic causes of this interaction. The absent object animacy effect for dative verbs could reflect semantic differences (i.e., the non-prototypically transitive argument semantics); this explanation would be in line with the eADM. It could also reflect lexical differences (lexical reaccess to the object for NOM-DAT, but not NOM-ACC verbs) and syntactic differences (the increased processing load associated with the build-up of a more complex syntactic structure in the case of NOM-DAT verbs). This explanation would be in line with the KP-based account proposed by Bayer et al. (2001).

The first and the second, but not the third explanation should also hold for particle verbs.

5.2 Results

Mean answer accuracy was 86% (SD = 8%).

ERP results are reported for latency windows timelocked to the beginning of the presentation of the critical verb. Mapping views for selected time windows and curves for selected electrodes are given in Figure 3.

Figure 3

Experiment 2 (particle verb stimulus set): Grand average ERPs for selected electrode sites and difference maps are shown. Accusative conditions are depicted in the upper part, dative conditions are depicted in the lower part. Mean voltage difference maps (animate minus inanimate) for the marked time window from 400–600 ms are given on the left side. The electrodes selected for illustration are marked in the maps.

400–600 ms: In the time window from 400 to 600 ms, there was a main effect of MEDIAL-LATERAL position (F (4,80) = 37.6, ɛ = .63, p < .001), ANTERIOR-POSTERIOR position (F (4,80) = 14.9, ɛ = .43, p < .001), and an interaction of MEDIAL-LATERAL position and ANTERIOR-POSTERIOR position (F (16,320) = 6.6, ɛ = .29, p < .001). In addition, there was also an interaction of ANIMACY and ANTERIOR-POSTERIOR position (F (16,320) = 3.6, ɛ = .50, p < .05) and an interaction of ANIMACY, MEDIAL-LATERAL position and ANTERIOR-POSTERIOR position (F (16,320) = 2.3, ɛ = .45, p < .05).

To resolve which positions drove the three-way interaction, we calculated separate two-way ANOVAs of ANIMACY and ANTERIOR-POSTERIOR position for each MEDIAL-LATERAL position, and separate two-way ANOVAs of ANIMACY and MEDIAL-LATERAL position for each ANTERIOR-POSTERIOR position. This ANOVA revealed the following effects for medial-lateral positions:¹¹ lateral-left: ANTERIOR-POSTERIOR position (F (4,80) = 19.8, ɛ = .38, p < .001) lateral-medial-left: ANTERIOR-POSTERIOR position (F (4,80) = 9.9, ɛ = .58, p < .001) and an interaction of ANIMACY and ANTERIOR-POSTERIOR position (F (4,80) = 4.2, ɛ = .78, p < .01). Resolving the effect of ANIMACY for individual sites showed a statistically significant effect of ANIMACY at O1 (t (41) = 2.1, p <. 05). midline: ANIMACY (F (1,20) = 4.4, p <.05; ANTERIOR-POSTERIOR position (F (4,80) = 6.1, ɛ = .61, p < .001) and an interaction of ANIMACY and ANTERIOR-POSTERIOR position (F (4,80) = 4.3, ɛ = .70, p < .01). Cz (t (41) = –2.1, p < .05), Fz (t (41) = –3.3, p < .01) lateral-medial-right: ANTERIOR-POSTERIOR position (F (4,80) = 8.2, ɛ = .63, p < .001) and an interaction of ANIMACY and ANTERIOR-POSTERIOR position (F (4,80) = 3.5, ɛ = .79, p < .05). sites: F2 (t (41) = –2.8, p < .01). lateral-right: ANTERIOR-POSTERIOR position (F (4,80) = 17.9, ɛ = .30, p < .001).

For anterior-posterior positions, the following effects were found: medial-anterior: ANIMACY (F (1,20) = 12.1, p <.01; MEDIAL-LATERAL position (F (4,80) = 14,9 ɛ = .61, p < .001), and an interaction of ANIMACY and MEDIAL-LATERAL position (F (4,80) = 6.6, ɛ = .56, p < .01). Resolving the effect of ANIMACY for individual sites showed a statistically significant effect of ANIMACY at F1 (t (41) = –3.0, p <.01), Fz (t (41) = –3.3, p <.01) and F2 (t (41) = –2.8, p <.01). posterior: ANIMACY (F (1,20) = 4.6, p <.05; MEDIAL-LATERAL position (F (4,80) = 19.2, ɛ = .59, p < .001).

As illustrated in the difference maps of Figure 3, waveforms at central-anterior sites were more negative-going for the animate than for the inanimate conditions, with no visible differences between waveforms for accusative and dative conditions. No main effects of CASE and no interactions of ANIMACY and CASE were found in this time window, and none were visible during the visual inspections of other time windows until the end of the presentation of the first postverbal word.

5.3 Discussion Experiment 2

The results of Experiment 2 show a clear main effect of object animacy, starting about 400 ms after the presentation of the verb, and persisting for about 200 ms. Waveforms for animate conditions were more negative-going than for inanimate conditions in central eletrodes. We interpret our findings as reflecting the enhanced processing load in sentences with animate compared to inanimate objects. These findings fit earlier reports of enhanced N400 components to animate compared to inanimate objects in the literature (Frisch & Schlesewsky 2001; Paczynski & Kuperberg 2011), and also fits the findings for the accusative conditions in Experiment 1 of the current study. The findings of Experiment 2 again strengthen the point that object animacy effects are subtle, but visible even in grammatical sentences without syntactic or semantic violations. They also match the predictions made by the models of sentence comprehension outlined above (Bornkessel-Schlesewsky & Schlesewsky 2006; Kuperberg 2007; Bornkessel-Schlesewsky & Schlesewsky 2009; 2013).

No main effects of verbal case marking, and no interactions of verbal case marking with object animacy were visible in Experiment 2. Our interpretation is that the contrast between NOM-ACC and NOM-DAT verbs does not affect the processing of particle verbs; at least, not to an extent that is detectable with our current paradigm. This outcome is difficult to explain in the framework of the eADM, as the difference between NOM-ACC and NOM-DAT verbs is assumed to be caused by their semantic differences, and these differences should hold for particle verbs, too. They fit better with the predictions implicit in the KP-based account, since this account assumes that the reason for lexical case marking effects lies in lexical and syntactic differences, and the syntactic difference between particle verbs assigning NOM-ACC and NOM-DAT should be smaller than for simple verbs.