Discontinuous particle verbs in German sentence comprehension Anna Czypionka 1,2 , Felix Golcher 3 , Carsten Eulitz 1 & Josef Bayer 1 Affiliations: 1) Constance University, Department of Linguistics 2) Wrocław University, Center for General and Comparative Linguistics 3) Humboldt-Universität zu Berlin, Department of German Studies and Linguistics Please see the handout for full references. German particle verbs = particle + base Subordinate clauses (SOV): orthographical unit at clause-final position ... dass ich morgen mit Peter den Film im Kino ansehe. ... that I tomorrow with Peter the movie in.the cinema PART.see `That I (will) watch the movie with Peter at the cinema tomorrow.‘ Main clauses (SVO): base verb = 2 nd constituent, particle clause-final Ich sehe morgen mit Peter den Film im Kino an. I see tomorrow with Peter the movie in.the cinema PART `I (will) watch the movie with Peter at the cinema tomorrow.‘ Particle verb formation is productive, many possible combinations: machen (`to make‘) combines with 31 particles (auf, zu, weg, klar, aus, rüber, .... ) Particles can change the base‘s • semantics: machen `to make‘, einmachen = `to pickle‘ / `to arrest‘ • syntax: machen 2-place, mitmachen 1-place, klarmachen 3-place -> German main clauses are semantically and syntactically ambiguous until the final position. Background Experiment 3: EEG Conclusion 1) accusative vs dative: • subtle case effects for separated particle verbs - different from nonseparated, different from simple verbs • no N400 - different from ungrammatical conditions -> no simple lexical reaccess to object as source of case effects with simple verbs -> fits interpretation of simple verb case effects as reflecting syntactic differences 2)+3) accusative vs intransitive and illegal: • N400 fits expectations / literature • P600 not previously reported -> lexical access is costly for intransitive and nonexisting particle verbs -> parser attempts syntactic repair / reanalysis with existing and nonexisting particle verbs More fine-grained analysis: • reassess case effects before final conclusions • quantitative differences intransitive vs illegal conditions? Experiments Stimuli GRAMMATICAL CONDITIONS UNGRAMMATICAL CONDITIONS (A) accusative: ( baseline) ~~~ Polizisten hören Unterhaltungen von Kriminellen heimlich ab, damit sie ... policeman.PL.(NOM) hear.3.PL conversation.PL.(ACC) of criminals secretly PART so.that. they ... `Policemen secretly wiretap criminals‘ conversations so that they are in the know.‘ (C) intransitive: Polizisten hören Unterhaltungen von Kriminellen heimlich weg, ... policeman.PL.(NOM) hear.3.PL conversation.PL.(INTR) of criminals secretly PART ... `Policemen secretly do not listen on purpose (*to criminals‘ conversations) ...‘ (B) dative: Polizisten hören Unterhaltungen von Kriminellen heimlich zu, ... policeman.PL.(NOM) hear.3.PL conversation.PL.(DAT) of criminals secretly PART ... `Policemen secretly listen to criminals‘ conversations ... (D) illegal: Polizisten hören Unterhaltungen von Kriminellen heimlich über , ... \\ policeman.PL.(NOM) hear.3.PL conversation.PL.(ILL) of criminals secretly PART ... \\ # `Policemen secretly do verb? criminals‘ conversations ...‘ \\ 40 sentence quartets built using 16 base verbs, repeated once to 3 times. The stimuli were used in three experiments. 400-550 ms: acc vs intr: CONDITION (F(1,21)=4.54, p<.05) acc vs nons: CONDITION (F(1,21)=9.72, p<.01, CONDITION:MINIROI (F(11,231)=3.84, p<.05). CONDITION sign. in 5 mini-ROIs. N400 for both conditions compared to accusative 600-900 ms: acc vs intr: CONDITION (F(1,21)=4.72, p<.05), MINIROI (F(11,231)=5.17, p<.01 acc vs nons: CONDITION (F(1,21)=8.31, p<.01, MINIROI (F(11,231=7.10, p<.001), CONDITION:MINIROI (F(11,231)=3.23, p<.05). CONDITION sign. in 5 mini- ROIs. More positive-going P600-like for both conditions compared to accusative -0.75 μV 0 μV 0.75 μV 600 ms - 900 ms -0.75 μV 0 μV 0.75 μV 600 ms - 900 ms -0.75 μV 0 μV 0.75 μV 400 ms - 550 ms -0.75 μV 0 μV 0.75 μV 400 ms - 550 ms P1 -200 200 4 6 -2 -1 1 2 3 μV ms P2 -200 200 4 600 -2 -1 1 2 3 μV ms 0 0 CP2 -200 200 4 600 80 -2 -1 1 2 3 μV ms 0 0 8 0 0 0 accusative dative intransitive illegal 22 participants, 64 electrodes, filtered bandpass 0.5-70Hz before segmentation Analysis: 32 electrode subset in 12 mini-ROIs (3 electrodes for lateral, 2 for midline); ANOVA mean amplitude in time windows for CONDITION*MINIROI (1) Case: Accusative vs. dative 200-300 ms: CONDITION (F(1,21)=6.27, p<.05), MINIROI (F(11,231)=13.49, p< .001), BED:MINIROI (F(11,231)=3.63, p<.05). CONDITION significant in 4 mini-ROIs, left-anterior and left-posterior. more positive-going for dative than accusative (left-anterior) 700-900 ms: MINIROI (F(11,231)=,6.70 p < .01), BED:MINIROI (F(11,231)= 2.59, p<.05). CONDITION significant in 3 mid/right-anterior mini-ROIs more positive-going for dative than accusative (anterior) -> No visible link to behavioral results -> No N400 - doesn‘t fit lexical reaccess from literature -> No left-anterior negativities on next word - unlike dative effects for simple verbs Effects are stat. significant, but very subtle, depend on electrode selection -0.75 μV 0 μV 0.75 μV 700 ms - 900 ms accusative dative Voltage difference map: dative-accusative F5 -200 200 400 600 -2 -1 1 2 3 ms 0 0 0 0 0 FP1 -200 200 400 600 -2 -1 1 2 3 μV ms 0 0 0 0 0 FP2 -200 200 400 600 -2 -1 1 2 3 μV ms 0 0 0 0 0 F6 -200 200 400 600 -2 -1 1 2 3 μV ms 0 0 0 0 0 0 0 0 -0.75 μV 0 μV 0.75 μV 200 ms - 300 ms Voltage difference map: intransitive-accusative Voltage difference map: illegal-accusative CP1 -200 200 4 600 -2 -1 1 2 3 μV ms 0 0 8 -> Lexical access is costly for non-matching and non-existing base-particle combinations. This fits findings from the literature [7,9]. -> Syntactic integration is difficult for non-matching and non-existing base-particle combinations. So far, no P600 has been reported for base-particle mismatches. Research Questions What happens when a sentence-final particle changes the verb from two-place nominative-accusative to 1) lexical dative? • earlier studies: accusative vs dative causes N400 [5,6] interpretation: additional synt. structure and lexical reaccess to object [1] • newer studies: no case marking effects for non-separated particle verbs! [4] - but lexical reaccess to object should still be necessary with dative part. verbs Dative effects with separated particle verbs? -> would confirm lexical reaccess to object 2) intransitive? • semantically implausible, but existing lexicon entry -> N400 • syntactically anomalous -> P600? 3) illegal base-particle combination? • semantically implausible, no lexicon entry -> bigger N400 than in 2)? • attempts at reanalysis / structure repair -> P600? Or only for existing verbs? Literature: no P600 for base-particle mismatches. one lexical entry [2,3] expressed by multiple words [8] Dutch main clauses [9]: • LAN to potential base verbs, not affected by number of existing base-particle combinations • graded N400 to clause-final particles: plausible < semantically implausible < nonexisting German main clauses with nonexisting base-particle combinations [7]: • late frontal neg. and N400 -> difficulties for lexical retrieval • but no P600 -> no synt. repair / reanalysis? • German subordinate clauses: No effects of case marking (unlike simple verbs) [4] Particle verbs in sentence processing Experiment 1: Acceptability ratings in comparison to accusative: • no difference to dative (t=1.05, p > .3) • difference to both ungrammatical conditions (intr.: t=-11.56, p<.001; ill.: t=-11.98, p<.001) illegal rated worse than intransitive (t=3.24, p<.01) Magnitude Estimation task, 57 participants Analysis: LMM, fixed eff. CONDITION, random terms PARTICIPANT (rand. slo- pe CONDITION) and ITEM. -> case effects in other experiments do not reflect acceptability -> ungrammatical conditions are ungrammatical -> syntactic / semantic mismatch is bad, nonexistence is worse Experiment 2: Self-paced reading times Self-paced reading times with occasional questions, 49 participants Analysis: LMM fixed effect CONDITION, random terms PARTICIPANT and ITEM. Sequence effect up to second order in fixed and random parts. -> ungrammaticality affects reading times, case does not -> acceptability difference in ungrammatical conditions is not reflected in reading times in comparison to accusative: • no difference to dative • difference to ungrammati- cal conditions on particle and second, third and fourth spill- over position intransitive vs illegal: no difference (see handout for stat. report) -2.0 -1.5 -1.0 -0.5 0.0 acc dat intr ill Condition rating / mean(log(rating/reference)) condition acc dat intr ill Grammaticality ratings -0.0032 -0.0028 -0.0024 -0.0020 part part1 part2 part3 part4 position model fit (-1/RT) Condition acc dat intr ill Self-paced reading times (2) and (3) Accusative vs intransitive, accusative vs illegal