The semantics and pragmatics of counterfactual sentences at the discourse level : Sobel-sequences and the licensing of negative polarity items

Abstract

In this thesis, I explore the question whether conditionals can be more accurately modelled with a (semi-)dynamic strict (von Fintel, 2001; Gillies, 2007) or a variably-strict (Stalnaker, 1968; D. K. Lewis, 1973) semantics. To this end, I examine two linguistic phenomena that are associated with conditionals: (weak) negative polarity items and (reverse) Sobel sequences. Negative polarity items are words with a varyingly restrictive distribution that mostly (but not exclusively) occur in negative contexts (e.g., any and ever ). Sobel sequences are conditional sequences that adhere to the pattern of ‘If φ, χ; but if φ ∧ ψ, ¬χ’. For negative polarity items, I rule out the traditional account that stipulated that they are licensed by downward monotone environment (Fauconnier, 1975a,b; Ladusaw, 1980) due to the contextually changing licensing of negative polarity items in non-monotone environments (Crnič, 2011a). This account would have restricted us to a (semi-)dynamic conditional semantics, which is (Strawson) downward monotone by nature, as the variably-strict semantics is non-monotone by nature. I then evaluate an alternative approach to negative polarity item licensing: the operator-based approach that stipulates that negative polarity items are licensed by a covert even-like operator that imposes a probability-based scalar presupposition on its associated sentences (see, amongst others, Crnič, 2014a,b). I improve the viability of this model by solving an issue that this model had with deriving a difference in question bias between questions containing unfocused weak negative polarity items and questions that contain the expression even ONE (see, amongst others, Crnič, 2014a,b; Krassnig, 2018; Jeong and Roelofsen, 2021, 2023). I then evaluate the interaction of this licensing model with a variably-strict conditional semantics and with a (semi-)dynamic strict conditional semantics. I show that the latter has a slight explanatory advantage over the former, being able to account for all of the known data, but conclude that this advantage might be mitigated by future research. (Reverse) Sobel sequences have traditionally been considered another key piece in the debate on how to accurately model conditionals because regularly ordered Sobel sequences are felicitous but reverse Sobel sequences are infelicitous. This status originally spoke in favour of a (semi-)dynamic strict semantics, which was designed to predict reverse Sobel sequences to be infelicitous, because the variably-strict semantics predicts that either sequence type should be felicitous. However, recent developments have shown that some reverse Sobel sequences might be consistently rendered felicitous (Moss, 2012), prompting a return to the variably-strict approach to conditionals with the addition of different selective pragmatic mechanisms that rule out some but not all reverse Sobel sequences (Moss, 2012; Klecha, 2014, 2015; Krassnig, 2017; K. Lewis, 2018; Ippolito, 2020; Krassnig, 2020, 2022). I survey the current empirical landscape on reverse Sobel sequences and improve upon it by conducting a reverse Sobel sequence felicity experiment. I conclude that the crucial factor for reverse Sobel sequence felicity is contrastive stress in the antecedent of the φ-conditional and isolate a number of highly influential sub-factors for felicity that were partly already proposed in the literature in isolation: namely, (i) counterfactuality, (ii) a lack of a causal link between φ and ψ, and (iii) the use of overt or covert exclusion of ψ as an epistemic possibility (the last of which acts as a felicity rescue operation for contrastively stressed reverse Sobel sequences). I propose a pragmatic model constructed around the effect of contrastive stress and show how this model is able to account for all of the known empirical data on reverse Sobel sequences via the interaction of contrastive stress and the aforementioned factors (where all components can be independently motivated by other phenomena in the literature). Crucially, I show that this proposed supererogatory pragmatic mechanism may be combined with either the variably-strict model or the (semi-)dynamic strict model without a change in the predicted felicity distribution (though the former requires a few additional independently motivated pragmatic mechanisms to do so). In the end, I conclude that the (semi-)dynamic strict account appears to have a slight explanatory advantage over the variably-strict approach overall, but I also conclude that this advantage can likely be eliminated with further modifications to the variably-strict approach, rendering both approaches equally viable with respect to (reverse) Sobel sequences and negative polarity items.