We present the results from an Event Related Potentials (ERP) study on the processing of anaphoric reference to quantified expressions (QEs) in Swedish. QEs pick out proportions of possible members of some set for which a property holds. In (1a) and (1b), for example, some or few members of the set of students attended the lecture.

(1) a. Some students attended the lecture.    b. Few students attended the lecture.

(2) a. They found it very interesting.    b. They stayed at home instead.

Some and few differ in polarity: some is positive (upward entailing) while few is negative (downward entailing) (Peters and Westerståhl, 2006) and this is of importance when referring back to the QE using anaphoric expressions. The sentence in (1a) is naturally followed by (2a), which is about the students attending the lecture (the reference set, REFSET). The sentence in (1b), in contrast, is naturally followed by (2b), which is about the students not attending the lecture (the complement set, COMPSET) (e.g. Moxey and Sanford, 1987). While (1b) can in fact be followed either by (2a) or (2b), (1a), cannot be followed by (2b).

Filik et al. (2011) is one of few studies of anaphoric reference to QEs in English using online measures (ERP). They report results for positive and negative QEs separately. Each type of QE shows REFSET and COMPSET effects, as described above, on the disambiguating word. A larger N400 for COMPSET vs. REFSET continuations for positive QEs, and the opposite for negative QEs. However, they do not report any results for the contrast between positive and negative QEs in the COMPSET condition. Since this is a very important condition and since it is known that QEs differ across languages (Nouwen, 2010; Tsai et al., 2014), we investigated this issue for Swedish.

160 experimental items of four sentences each were manipulated along two dimensions: polarity (positive vs negative quantifier, några vs få in (3)), and set (REFSET vs COMPSET targeting disambiguating adjective, duktiga vs dåliga in (3)). The quantifiers included were: några (‘some’) ,få (‘few’), många (‘many’), inte många (‘not many’), alla (‘all’), inga (‘no’), nästan alla (‘almost all’), inte alla’ (‘not all’).

(3) Några/Få studenter skrev bra på tentan igår och att de_CW var så duktiga/dåliga_CW förbryllade professorn.

some/few students wrote well on the exam yesterday and that they were so good/bad confused the professor.

There were four lists with 40 sentences from each condition. Each participant (29 in total, results reported below based on the first 13) only saw one sentence from each item, but saw all types of manipulation. In total, each participant read 400 sentences (160 test items, 240 fillers).

Unlike Filik et al. (2011) we found that positive QEs showed a pronounced positivity over the central region (FCZ, CZ, CPZ, PZ) in the COMPSET condition relative to negative QEs, in the P600 time span (500–800 ms) after the onset of the critical word (the disambiguating adjective, ‘bad’). A linear mixed effects model analysis (LmerTest) showed a highly significant main effect of polarity in the central region and the P600 time span above. We interpret this to mean that for positive QEs, a new discourse referent needs to be introduced following COMPSET reference, while for negative QEs this discourse referent is already available (Burkhardt, 2007).