📄 The semantics and pragmatics of nouns in concealed questions

Summary

The paper asks what makes a determiner phrase usable as a concealed question (CQ), i.e., a DP that is understood as an embedded question, as when "I know your age" means knowing what that age is. Earlier semantic accounts (Nathan 2006, Barker 2016, Frana 2013/2017) tie CQ readings to the head noun: relational nouns such as age or capital allow them, while unmodified sortal nouns such as brick do not. Kalpak gathers counterexamples that go both ways. "I know your card" works even though card is sortal, and "#Alex knows the mother" fails even though mother is relational. She concludes that the explanation must be pragmatic.

Her account extends Aloni & Roelofsen's (2011) conceptual-cover semantics in two ways. It adds a standard relational/sortal noun semantics, and it adopts Barker's idea of alternative domains, but treats the domain as set by context rather than fixed by the lexicon. A CQ reading needs a salient way of identifying individuals within that domain. The theory explains why relational nouns usually help (they always supply a default domain), why this tendency sometimes fails, and it removes the need for derived covers.

Key points

• CQ availability depends mainly on whether context makes a method of identification salient over a domain of alternatives. The head noun only proposes a default domain, which context can override so that the question is non-trivial.
• The noun taxonomy follows Partee 1997 (sortal = sets of individuals; relational = two-place relations). It also uses de Bruin & Scha 1988 to split relational nouns into functional and non-functional, and functional ones further into one-to-one (nose) and merely onto (temperature).
• For quantified DPs, pair-list readings no longer rely on derived covers. As a result, non-functional relational nouns also get them, as in "Sam knows every phone number."
• An unmodified sortal noun gets a pair-list reading only when context supplies two independent basic covers over the same domain. For cards, these could be suit/rank and position on the table ("Kim knows every card").
• Modifiers help CQ readings for specific reasons:
  • favorite supplies the missing relation;
  • relative clauses make the domain resolution consistent;
  • superlatives make uniqueness easier to satisfy.
• These modifiers are not sufficient on their own: "the reddest apple" stays odd because apples have no context-independent way of being identified.

Interesting ideas

• I1. Instead of treating "can this DP stand for a question?" as a lexical yes/no feature, the paper treats it as a default from the noun that context may override, which avoids brittle hard rules.
• I2. A question is only sensible if there are distinct ways to pick out the same entities (name vs. description, rank vs. position); if only one exists, the question collapses into something trivial.
• I3. The split into sortal, relational, functional and one-to-one nouns predicts which readings appear (unique values, multiple values, pair-lists).
• I4. Each modifier is explained by what it contributes (a relation, a consistent domain, uniqueness) rather than by being on a list of licensers.
• I5. The paper tests its theory against minimal pairs in both directions: sortal nouns that succeed and relational nouns that fail.

For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

• I1 (frames): when a DP is the object of know/reveal/find out/guess, prag could try a concealed-question frame ("know what X is") preferred for relational nouns, but also allowed for sortal nouns whenever the world offers a way to identify entities.
• I2 (world): entities in the text world could store several identification keys (name, attributes, position), so that the system can check whether a "which one / what is it" question is non-trivial before trying to answer it.
• I3 (world): adding these flags to the compiled noun ontology (e.g., capital functional, phone number non-functional) would help decide whether "every capital" means a single answer or a list of pairs.
• I4 (frames): the parser could record which piece of information a modifier like favorite or a superlative supplies to a DP, so that it is clear in the trace why a definite description now picks out one referent.
• I5 (learn): evaluation suites could include such two-sided negative controls (e.g., "knows your card" vs. "#knows the mother"), so that a rule learned from noun type alone is caught when it overgeneralises.