Wojciech Rostworowski, Katarzyna Kuś and Bartosz Maćkiewicz
Affiliation: University of Warsaw
Category: Linguistics
Keywords: attitude ascription, disjunction, free choice, semantics
Date: Thursday 4th of September
Time: 14:30
Location: Room 161 (161)
View the full session: Concepts & Propositional Attitudes
The talk focuses on the interpretation of natural-language disjunction (‘or’) in the scope of different attitude verbs, i.e. verbs that describe an agent’s mental state or ‘attitude’ towards a proposition. Based on experimental methods, we argue that the so-called ‘free-choice’ interpretation is accessible for attitude ascriptions, which suggests that free choice is a more universal phenomenon than it is assumed.
Roughly speaking, the free-choice (FC) reading of ‘or’ implies its conjunctive force. Typical examples include possibility modals, e.g., (1) You are allowed to eat an apple or a pear. (permission) ⇒ So, you are allowed to eat an apple. ⇒ So, you are allowed to eat a pear. (2) It might be in the kitchen or in the bathroom. (epistemic possibility) ⇒ So, it might be in the kitchen. ⇒ So, it might be in the bathroom. In contrast, the FC reading does not arise when ‘or’ is in the scope of necessity modals, e.g., the following sentence does not licence an inferences analogical to (1): (3) You must (= are required to) eat an apple or a pear. ⇏ So, you must eat an apple. ⇏ So, you must eat a pear. Various semantic accounts have been proposed in order to explain the presented data (see Zimmerman 2000, Simons 2005, Fox 2007, Aloni 2022).
An important empirical question from the viewpoint of any theoretical analysis of FC is what kinds of embeddings allow for FC inferences. Most accounts assume that FC is connected to existential quantification of different sorts (e.g.., over deontic or epistemic possibilities). Experimental literature already confirms this assumption with regards to the cases with permissions (e.g., Chemla & Bott 2014). There is also evidence that speakers tend to perform FC inferences under epistemic possibility modals, existential quantifiers, and in the cases in which a modal operator is further embedded under quantification (yet, these effects are not as strong as in the case of deontic modals, see van Tiel 2012). Interestingly, Liu (2019) shows that people perform a weaker sort of FC inference in ‘must’-embeddings, that is, draw the conclusion that X can do p as well as q, provided that they are told ‘X must do p or q’.
Our study adds to these projects by investigating whether FC readings are available when ‘or’ is embedded under attitude verbs, in particular, the ones expressing desires, e.g., (4) She wants to drink tea or coffee. ⇒? So, she wants to drink tea. ⇒? So, she wants to drink coffee. While the formulated conclusions may sound too strong, the embedded disjunction in (4) triggers an implication that offering her a tea would satisfy her wish as well as offering her a coffee. Some theoretical considerations suggest that a kind of FC-inference is valid in the context of desire ascriptions (e.g., Blumberg and Hawthorne 2021). To our best knowledge, there is no empirical evidence in favor of this hypothesis.
Experiments: Our two experiments address the question whether people perform FC inferences for attitude ascriptions with ‘want’ and ‘fear’. In our studies, modelled after Chemla’s & Bott’s (2014) work on FC inferences, the participants were presented with a set of facts about attitudes of two groups of inhabitants of an imaginary island. This setup enabled us to construct experimental and control sentences such as the following example: SINGLE CONTROL (SC): Paul-the-Urbanite wants to eat a toffee. DOUBLE CONTROL (DC): Patricia-the-Nomad wants to eat a fudge or a gingerbread. EXPERIMENTAL: James-the-Urbanite wants to eat a fudge or a sausage. Both types of control sentences were designed to be unambiguous when it comes to their logical value. In the DC sentences, both disjuncts were of the same category of entities (fruits, sweets, animals, or vehicles) to ensure that the sentences are unquestionably true or false. The experimental sentences were predicted to be true, if the classical logical interpretation of ‘or’ is employed, and to be false under the FC interpretation. Following Chemla’s and Bott’s studies, the sentences were presented in segments in the centre of the screen and the judgement was made after the last segment.
Results: The participants generally correctly judged true control sentences as ‘true’ and their false versions as ‘false’. The experimental sentences were judged on average as true in less than 50% of cases and they were accepted less often than true DC (p < 0.001) and true SC (p < 0.001) items and more often than control ‘false’ items (DC: p < 0.001, SC: p < 0.001). The analysis of RTs for control items revealed an expected pattern of correct responses being significantly faster than incorrect ones (DC: p=0.005, SC: p=0.005). On average, the target sentences were typically judged as ‘false’ slightly faster than as ‘true’.
Discussion: The results indicate that ‘want’- and ‘fear’- ascriptions with ‘or’ are treated as ambiguous and some participants interpret them as expressing falsehoods in the experimental condition, which is compatible with the hypothesis that those people perform a sort of FC reasoning. This can be further explained in two ways: (a) the participants derive a conclusion that S wants/fears p and S wants/fear q from the target ascription (‘S wants/fears p or q’), (b) they draw a weaker conclusion to the effect that both p and q are positive (in the case of ‘want’) or negative (in the case of ‘fear’) for the attitude holder, and given that these weaker conclusions are still false in the light of the background story, they reject the target ascriptions. The (b)-type of inference would be similar to the kind of FC inference with a weaker modal conclusion (‘can’), identified by Liu (2019) in ‘must’-embeddings. Our second experiment focuses on the question which particular kind of FC-inference ((a) vs (b)) is preferably performed. Altogether, our findings indicate that FC is present in attitude-verb contexts. Also, the result concerning response times gives limited support to the hypothesis that the FC reading is a default interpretation, which could be overridden in a time- and resources- consuming process of reinterpretation.