Truth conditions & Entailment

4.2 Trigger study (Tiemann, 2011, Experiment 1)

For this study, the goal was to compare what happens when comprehenders encounter triggers and non-trigger words. A target sentence varied in three ways: trigger, non-trigger, and semantically unacceptable. The target sentence was always preceded by one sentence to provide some context, but the content of that sentence didn't change the interpretation of the target sentence variants.

Context sentence
Tina ist mit einer guten Freundin shoppen.
Tina is with a good friend shopping.
'Tina is shopping with a good friend.'

Target sentence variants:

Target sentence: trigger
Sie kauft wieder rote Handschuhe.
She buys again red gloves.
'She buys red gloves again.'

Target sentence: no trigger, acceptable
Sie kauft heute rote Handschuhe.
She buys today red gloves.
'She buys red gloves today.'

Target sentence: no trigger, semantically unacceptable
Sie kauft freundlich rote Handschuhe.
She buys friendly red gloves.
'She buys red gloves friendly.'

What were the predictions again? Which condition should be hardest? Easiest? [click to reveal]
If presupposition triggers signal to comprehenders that they will need to check the context to see if it contains the presupposed content, the trigger condition is predicted to be more difficult than the non-trigger condition. Most difficult of all should be the condition with the unacceptable sentence.

The target items were interspersed with fillers to distract participants so they'd be less likely to guess the goals of the experiment or to develop strategies in the task. Each sentence was followed by a comprehension question to encourage participants to pay attention. The experiment used a within-subject design whereby every participant saw all 3 conditions (in fact every subject saw every item in every condition over 3 days, a rare level of rigor in psycholinguistic studies).

Figure 1 below shows the results from the Trigger rating study. The y-axis shows acceptability ratings (from 1 to 4). Each position on the x-axis corresponds to a different item type with a different type of trigger (or non-trigger or semantically unacceptable word). The height of the bars indicates the average rating across participants for the unacceptable condition (black), the presupposition condition (light grey), and the fully acceptable no-trigger condition (dark grey).

Figure 1: Trigger study, Tiemann et al. 2011 (PSP = presupposition)

The ratings results follow the predictions: The unacceptable items consistently get the lowest ratings; items with presuppositions are given higher ratings, and the highest ratings go to fully acceptable non-trigger items. The reading times at the trigger word (or corresponding non-trigger or unacceptable word in the other conditions) showed the same pattern: Semantically unacceptable words took longer to read than presupposition triggers which took longer to read than non-triggers.

KEY POINT: Offline acceptability ratings and online reading times both point to a gradation in ease of processing: Non-trigger words are easier than trigger words which are easier than semantically unacceptable words.

Is it okay to compare reading times on 3 different words (wieder, heute, freundlich)? [click]
No. In general, psycholinguistic studies try to avoid comparing apples and oranges, which is what happens in this experiment. Though unlikely to explain the systematic biases that were observed, it is possible that the items with the non-trigger word were always rated highest because those words are just shorter or more frequent or more fun along some dimension.
- Can you think of a way to avoid that problem? [click]
  Ideally, the target sentence would stay the same across conditions (comparing apples to apples) while the preceding context is what is manipulated. It's hard to do that to compare trigger and non-trigger words.
  
  The next study, however, does try to get at a similar question about when comprehenders engage in pragmatic processing of presuppositions (wow alliteration!) and it does it with an experimental design that avoids comparing apples to oranges.

To go on to section 4.3 "Verification study", click here.