📄 Contrastive topics decomposed

Summary

The paper revisits how contrastive topics (CTs) should be analysed. The standard account by Büring (1997b, 2003) marks a CT feature and a FOC feature. A special non-compositional rule then builds a "topic semantic value", a set of question-sets. Under that rule, CT and focus could appear in any structural relation to each other. Wagner argues that this freedom is not what languages show.

He collects judgments from English, Italian, Korean and other languages. The data show that CT constructions follow the same strict syntax as sentences with two overt, nested focus operators such as even and only. His generalization is that the CT must take scope over the focus.

He then gives a compositional analysis. A CT is the associate of a higher, unpronounced focus-sensitive operator, and the sentence focus belongs to a lower one. The analysis needs only ordinary Roothian focus alternatives, with no nested topic value. It also extends to scope effects of CTs and suggests a new view of wh-intervention effects. This matters because it brings information structure under general principles of scope.

Key points

• Target analysis. Büring's account uses a two-step procedure. Focus alternatives come first, then CT alternatives turn the result into a set of questions. It comes with a "disputability implicature": some question in the topic value must remain open after the assertion.
• Intonation. In English ToBI terms, CTs typically carry L+H* followed by L- H% (a fall-rise, the B-accent). Foci carry H* followed by L- L%.
• Configurations. Krifka (1992) lists five configurations for multiple foci. Wagner identifies the CT case with the one where each of two separate operators has its own associate, and the CT goes with the wider-scoping operator.
• Diagnostic example. In the test sentence where even attaches to "the best student" and only attaches to "one problem," the likelihood scale works only if even takes scope above only. Without only, the sentence is odd.
• Korean data. With cocha 'even' and man 'only', the wide-scope operator always comes first in linear order. On this basis, wh-intervention effects are recast as a wrong scope order between a Q operator and man, possibly causing a type mismatch.

Interesting ideas

• I1. Treat a CT plus focus as two stacked focus operators, each over its own alternatives, instead of one special nested object. This reuses a single mechanism for both.
• I2. Test a covert operator by checking whether its syntax matches a sentence with an overt operator in the same configuration. Here the overt case is even over only.
• I3. Use likelihood scales as a scope probe. Even only works if its alternatives are ordered the right way, so the felicity of a sentence reveals which operator is higher.
• I4. The CT's disputability requirement says some sibling question must remain unresolved after the answer.
• I5. Intervention effects are explained as wrong scope order between operators, here the Q operator below only. A type mismatch then blocks interpretation, rather than a broken association.

For Mova

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

• I1 (frames): prag could represent a CT answer as an outer operator over a focus frame, yielding the implied larger question ("who ate what?") and the open sub-questions without a separate CT module.
• I2 (learn): build evaluation pairs where an overt-operator paraphrase acts as a control, checking whether the system assigns the same scope reading to the covert version.
• I3 (world): when ranking alternatives for scalar particles (even, only, at least), compute likelihood or measure scales in the world model and flag readings whose ordering clashes.
• I4 (answer): in QA over tales, a fall-rise or contrastive answer can signal that other sub-cases are still open, so the reader should not infer exhaustivity for the remaining entities.
• I5 (tree): after parsing, a check could detect focus particles linearly before a wh-word in fixed-scope languages and mark the structure as uninterpretable, rather than forcing a frame.