Secret Agents or Sleeping Beauties: What Happens to Repaired Constituents?

Secret Agents or Sleeping Beauties: What Happens to Repaired Constituents? Noa Shuval (noa.shuval@gmail.com) Laboratoire de Psychologie et Neuropsychologie Cognitives, CNRS, Universite Paris Descartes, 71 ave Edouard Vaillant, 92774 Boulogne-Billancourt, France Lars Konieczny (lars.konieczny@frias.uni-freiburg.de) FRIAS, Universitat Freiburg Starkenstr. 44 79098 Freiburg, Germany Barbara Hemforth (barbara.hemforth@parisdescartes.fr) Laboratoire de Psychologie et Neuropsychologie Cognitives, CNRS, IUPPD, Universite Paris Descartes, 71 ave Edouard Vaillant, 92774 Boulogne-Billancourt, France Abstract (1) J’irai chez le boucher, euh non, le boulanger. J’ai besoin de pain. I’m going to the butcher, eh no, the baker. I need some bread. In two experiments we investigated the comprehension of sentences with repaired NPs. In our first experiment, we applied an acceptability task after speeded auditory presentation of French versions of sentences with and without repairs like “I will go to (the butcher, uh no,) the baker. I need some bread/meat”. While repairs led to reduced acceptability for consistent continuations, the inconsistent continuation was more acceptable when a compatible but repaired constituent had been mentioned before, suggesting that the to-be-repaired constituent was not fully overwritten by the correction. In our second experiment, the visual world paradigm was used to auditorily present participants with the stimuli compiled for Experiment 1, while they looked at corresponding visual stimuli. This time, evidence from eye fixation patterns suggests that the to-be-repaired constituent was actually suppressed online during sentence processing. To settle this contradicting evidence we would like to suggest that the acceptability judgments are mainly the result of offline reconstruction of memory traces following Gimenes et al., Keywords: Sentence Processing; Disfluencies; Acceptability Judgments; Visual World Paradigm Introduction The human sentence processing system has to be extremely robust since it does not only have to cope with highly standardized and edited to correct input, but very often (probably more often than not) also with deficient input caused by various, often non-linguistic, situational factors. In this paper, we will look at the comprehension of repaired utterances like (1). It has been proposed that disfluencies such as silent or filled pauses or repairs may lead to undesirable effects in sentence processing, leaving misparses harder to detect, possibly by providing cues which are interpreted as prosodic structuring information (Bailey & Ferreira, 2003; Maxfield, Lyon, & Silliman, 2009). For repairs, it has been proposed that the to-be- repaired constituent may continue influencing listeners’ comprehension, the so-called lingering effect (e.g., Lau & Ferreira, 2005). It is this latter effect that we investigate in our experiments. Disfluencies are highly frequent in natural language production. They include editing terms such uh and um as well as repeats (“I – uh - I wouldn’t”, e.g. Clark & Wasow, 1998) as well as revisions. Typically, in spoken language, disfluencies can be found in about six out of 100 words (Fox Tree, 1995). In the corpus used by Levelt (1983), 25 % of the annotated disfluencies were repairs similar to the structures under investigation in our studies. Of these, 62 % included editing expressions like Dutch versions of “I mean” or “that is” or mostly (30 % of all repairs) the Dutch version of “uh”. Since disfluencies in general and repairs in particular are so frequent, listeners have to find ways to process them, they have to detect the disfluency, see what the problem is, and edit out the part of speech to-be-repaired to arrive at the intended meaning of the utterance. Research on error processing in spelling has provided evidence, that recently processed incorrect information (Brown, 1988; Dixon & Kamisnka, 2007; Jacoby & Hollingshead, 1990) may affect subsequent performance even in cases, where the error has been explicitly recognized as such (Perruchet, Rey, Hivert, & Pacton, 2006). Editing out explicitly marked repairs may, equally, not always work perfectly well. Lau and Ferreira (2005, see also Bailey & Ferreira, 2003; Ferreira, Lau, & Bailey, 2004) claim that the to-be-repaired constituent in repetitions and corrections introduces lexical content and local syntactic structure not fully overwritten by the correction. They studied a disfluency involving the repair of a verb (like chosen vs. selected) in sentences like (2 a, b). (2) The little girl a. chosen-uh/b. picked-uh selected for the role celebrated with her parents and friends. Sentences like these, with verbs like “selected” which are ambiguous between a main verb and a past participle reading, usually lead to comprehension difficulty (e.g.,