While it is well-known that syntactic constraints are rapidly processed (e.g., Hagoort et al., 1999; Pearlmutter et al., 1999), world knowledge (e.g., Ferretti et al., 2001; Troyer & Kutas, 2020) and social context (e.g., Münster & Knoeferle, 2018; Nieuwland & Van Berkum, 2006) are also integrated to resolve ambiguity. Uncovering how context information, such as formality, affects sentence comprehension in real time, and whether and how this interacts with morphosyntactic constraints, can help better understand variability in language processing (e.g., Adger, 2006). We tracked eye movements during sentence reading to investigate: (i) whether context formality-register congruence is rapidly processed; (ii) whether and how sensitivity to formality-register congruence and subject-verb agreement interact during sentence comprehension. We expected rapid, incremental integration costs for mismatches (relative to matches; e.g., Hagoort et al., 2004; Van Berkum et al., 2008), with additive or interactive effects reflecting, respectively, distinct versus shared underlying mechanisms/representations (Hagoort, 2003). We also explored whether variability in perceived formality modulates such effects.

Methods: In two eye-tracking pilot studies, monolingual German adults (aged 18-31) read 40 critical items interleaved with 56 filler items. Each item (see Table 1) comprised two context sentences conveying a formal or informal situation, followed by a target sentence, which featured an animate subject, a high- or low-register verb (matching or mismatching context formality) and a direct object. In the register-only pilot (N=8), we manipulated formality-register congruence (match vs. mismatch). In the register-by-subject-verb-agreement pilot (N=8), we additionally manipulated subject-verb agreement (match vs. mismatch), such that the verb would be incorrectly inflected, in the infinitive form, in the mismatch conditions. Linear mixed-effects models were fitted to log-transformed first-pass duration, regression path duration, and total reading time at the verb and object (NP2) regions in the target sentence. Average (independent) target and context sentence formality ratings served as covariates; random intercepts were fitted for items and participants.

Results: In the register-only pilot, as predicted, longer verb total reading times were observed (Fig. 1) for register mismatches (vs. matches, t=-2.05, p=.040, d=-.25), and for higher target sentence formality ratings (t=3.22, p=.001, d=.37; post-verbal effects n.s.). In the register-by-subject-verb-agreement pilot, as expected, verb total reading times (Fig. 2) were longer for subject-verb agreement mismatches (vs. matches, t=-1.98, p=.048, d=-.24); contrary to predictions, total reading times were longer for register-matching verbs (vs. mismatching, t=3.24, p=.001, d=.39; formality rating effects n.s.). In the NP2 region (Fig. 3), we observed first-pass reading effects of subject-verb agreement (t=2.07, p=.038, d=.25), register congruence (t=-2.50, p=.012, d=-.31), and of their interaction (t=2.00, p=.045, d=.25). Finally, longer regression path durations in NP2 regions were observed following agreement mismatching verbs (t=-2.83, p=.005, d=-.35), register-matching verbs (t=2.55, p=.011, d=.31), and as a function of higher context formality ratings (t=2.45, p=.014, d=.29, Fig. 4).

Conclusions: Effects of formality-register congruence, as well as subject-verb agreement in the second pilot, emerged at a late processing stage in the verb region, and at earlier stages in the object/spillover region (second pilot). While rapid integration of context effects is not fully supported by our findings, when both factors were manipulated, the direction of the register congruence effect was opposite to that predicted. This may suggest an interference between context and subject-verb agreement processing during sentence reading. The ongoing register-by-subject-verb-agreement main experiment, with a larger N (=40) will be presented at the conference, helping to further address these questions.