When negative statements are easier: Processing polarity items in Japanese

Abstract:

Many previous studies of sentence processing have demonstrated that negative statements present more difficulty than affirmative statements (Johnson-Laird 1983).  In contrast, in this study I present evidence from Japanese which shows that under certain circumstances negative elements are easier to process, due to the need to license negative polarity items which precede their licensers.

Japanese has two suffixes meaning 'only', which have contrasting licensing requirements.  The suffix sika is a negative polarity item (NPI), which must be followed by a licensing negative morpheme; on the other hand dake is a non-NPI, which does not require licensing by NEG.  In contrast to English, NPIs precede their licensers in Japanese.  Therefore, using the logic of Miyamoto & Takahashi's (2001) Type Mismatch Effect shown in wh-scope markers, we expected that following the NPI sika, the parser should set up an expectation for a negative marker.  Following the non NPI dake, however, no such expectation should be created.  This prediction was tested in a self-paced reading study in Japanese (n=28).

Comparing (1a) and (1b), we expect that the reader will slow down at the fifth region in (1b) but not in (1a).  In (1a), the NPI, which created a prediction for the licenser, is licensed at this position.  However, in the non-NPI condition (1b), the reader does not expect the negative element to appear.  Thus, a slowdown is predicted in (1b).  In (1c) and (1d), however, we do not expect the slow down at the fifth region because the verb is in affirmative form so there is no causes of difficulty.  We further expect that at the fourth region, reading times should be slower in (1d) than (1c) because in (1c), NPI creates an expectation for NEG, but in (1d) non NPI does not create an expectation for NEG and there are additional processing demands.  The results of on-going experiment strongly support these claims.  There are significant differences in reading times between (1a) and (1b) at the fifth region (p<.0001), and also there are significant differences between (1c) and (1d) at the fourth region (p<.0001).  This study will shed light on the processing of Polarity and Non Polarity items with respect to clause typing.  We can see how the polarity items or non-polarity items establish predictions in on-line sentence processing.