Metasemantics without semantic intentions

ABSTRACT

The most common answers to metasemantic questions regarding context-sensitive expressions appeal primarily to speakers' intentions. Having rejected intentionalism in Lewis [(2020. “The Speaker Authority Problem for Context-Sensitivity (Or: You Can't Always Mean What You Want).” Erkenntnis 85: 1527–1555.], this paper takes a non-intentionalist perspective in answering the metasemantic question: how does a context determine the value of context-sensitive expressions? It focuses on the case of gradable adjectives, i.e. expressions like ‘tall’, ‘expensive’, and ‘rich’, which require a contextually determined standard in the unmarked positive form, as in ‘Pia is tall’. I argue that this standard is determined by a salient comparison class, which, when embedded in the relevant facts, provides input into statistical reasoning which outputs a standard in accordance with conversational domain goals.

KEYWORDS:

• Metasemantics
• context-sensitivity
• gradable adjectives
• speaker intentions

Acknowledgments

Thanks are due to Michael Glanzberg, Mona Kleinberg, and Emmanuel Viebahn for helpful discussions. Thank as well to audience members at the Variables and the (Meta)semantics of Context-Sensitivity workshop at the University of Vienna in September 2019, the Knowledge, Context, and Responsibility conference at Peking University in October 2019, and the Language and Cognition University Seminar at Columbia University in February 2020 for comments, questions, and discussion, as well as an anonymous reviewer for this journal for comments, all of which helped improve this paper. All mistakes belong to the author.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 Absolute gradable adjectives like ‘flat’, ‘empty’, ‘impure’, or ‘wet’ have a minimal or maximal point connected to their meaning, unlike relative gradable adjectives. For example, it is possible to be completely empty or perfectly flat but not completely tall or perfectly big. From now on, I will drop the term ‘relative’ and just say ‘gradable adjectives’.

2 Some other recent work arguing against intentionalist metasemantics includes Bach (2017) and Mion and Gauker (2017) Bach does not think a non-intentionalist metasemantics will fair any better and argues against there being any semantic values for these sorts of context-sensitive expressions at all. Mion & Gauker, on the other hand, are much in favor of a non-intentionalist metasemantics. Glanzberg (2007, 2009, 2020) argues for an indirect metasemantics for gradable adjectives, and his view overlaps in many ways with my own.

3 See Rett (2015), chapters 2 and 3, for an excellent overview.

4 This presentation of the semantics makes traditional assumptions, i.e. that semantics works by function application and ‘is’ in these constructions is inert. There is no obvious way in which giving up these assumptions solves the problem presented.

5 For an overview of the various popular treatments of pos, see Lewis (2020).

6 The metasemantic account would equally apply to Kennedy (2007)'s semantics in which the variable is a function that takes adjective meanings and returns a degree.

7 Though I will not pursue this here, this picture could perhaps be developed further within a coherence theory framework. See Stojnić (2018); Stojnić and Stone (ms) for work in this vein.

8 The notion originates with Rosch et al. (1976). For a good overview of research on the basic-level category, see Murphy and Lassaline (1997).

9 Interestingly, Croft and Cruse (2004, 96-97) also point out that there is some contextual variation in basic-level categories, and that people who are familiar with two versions of a hierarchy can operate within both systems, depending on the context. To take one of their examples, it is plausible that for a dog breeder spaniel, collie, terrier, etc. will be basic level (as these will satisfy the above criteria for basic-level categories) and dog superordinate. But when the dog breeder has a conversation with a non-specialist, she will treat dog rather than spaniel as the basic-level category, as non-specialists do. This seems to yield the right prediction for the role of basic-level categories in the current theory. If someone says ‘he is tall’ regarding a particular spaniel, the conversational context, i.e. whether it is dog breeders conversing or non-specialists, will determine whether the natural comparison class property raised to salience is spaniel or dog. Thank you to an anonymous referee for pointing out the relevance of this passage in Croft & Cruse to my proposed metasemantics.

10 See Roberts (2004, 2012a) for more on questions under discussion.

11 In the case of something like is Marjorie tall? the basic-level category that Marjorie belongs to may be the thing raised to salience. But this is nothing special about its role as the question under discussion. This is just another example of a default interpretation.

12 I changed Glanzberg's example slightly here. Glanzberg has this sentence as “he is too tall”, but as an anonymous reviewer for this journal pointed out, it is difficult to isolate the effects of context if we also vary the sentence between the two contexts.

13 I take it that in some cases of gradable adjectives such as ‘intelligent’ or ‘beautiful’, social rather than physical facts will be the relevant facts in which the comparison class property is embedded and which help determine the input values to the relevant calculations that determine the standard of comparison. I leave this for future work.

14 There is much work to be done here. Solt and Gotzner (2012) and Schmidt et al. (2009) actually come to incompatible conclusions. Schmidt et al. (2009) conclude that a model based on ordinal ranking (relative height by range) or a clustering model is best, while Solt and Gotzner (2012) come to the conclusion that these sorts of models are inadequate and we require models that take measurement degrees into account, such as the standard deviation model, rather than merely ordinal ranking.

15 Furthermore, unlike Fara, I do not think that interests or domain goals enter into the semantics of the proposition expressed. They are only part of the metasemantics.

16 I am borrowing the idea of goal-conduciveness from Dobler (2019), though I am implementing it in a different way and applying it to something different. Dobler is concerned with occasion-sensitive, objective predicates and not contextually saturated variables. Her way of implementing goal conduciveness does not naturally transfer to the case of gradable adjectives.

17 Of course, we can look at this individual's budget and see what is the most she can possibly pay in rent. I agree there is some sense in which this is the appropriate cut-off. This is the standard we might use if we said that every apartment in a certain neighborhood was (in)expensive (and we were not comparing it to another neighborhood). I will return to this in the next section.

18 Sometimes when we say things like this we are implicitly comparing one area to another or one time period to another, but there is also a sense in which people do this without an implicit comparison, meaning rents are (too) expensive, period.

19 An anonymous reviewer for this journal suggested that perhaps what we have here is a comparison class that invokes counterfactual apartments with rents that the individual in question can afford (or actual apartments with counterfactual rents). Thus there is a comparison class here, relative to which the rents of the actual apartments stand out as expensive. I have no in principle problem with this suggestion. As the reviewer notes, it has the advantage of utilizing a comparison class for all cases, yielding a more uniform view. On the other hand, it's not clear to me there is a substantive philosophical difference between my position and the suggested one. (There is indeed a real difference in the way in which the two versions would be formally implemented.) On the reviewer's suggestion, the standard is determined by the comparison class of counterfactual apartments which is in turn determined by the domain goal. It is not like there is a natural comparison class relative to which there is a notion of standing out. In both my view and this one, in these cases the domain goal is doing the real work rather than a comparison class. Compare this to a case in which we have 50 identical widgets, 40 of which cost under $20 and 10 of which cost above $20. In this case, it is clear that costing more than $20 is what it takes to stand out as a standard for expensive. But take another case in which we have maximum $20 to spend on a widget. There is a sense of ‘expensive’ as used in natural language in which one might say ‘these widgets are expensive’ if they all cost more than $20. One might treat this as a matter of the cutoff being determined by the domain goal, as I have. One might alternatively treat this as a matter of the standard being determined by the price that stands out relative to some counterfactual cheaper widgets as determined by our domain goal. But neither of these amount to the same notion of standing out as in the first case. My hunch is that there is not a substantive philosophical difference between the two ways of thinking about the latter case.

20 On a view in which the semantics of gradable adjectives required a comparison class property, here the property would just be being an object that has a temperature, i.e. the comparison class is just all objects that have a temperature.

21 I am making no claims here about how automatic or pure any indexicals actually are.

22 King does argue that there are no cases of felicitous underspecification in which there are no candidate semantic values given by context. That is, if the possibilities for the semantic value are not constrained at all by the context, there is no possibility for felicity. King calls these cases of catastrophic failure.

23 Neale (2016), for example, worries about the plausibility of speaker's intentions about unpronounced variables.