Automated tableau generation using SPOT (Syntax Prosody in Optimality Theory)

Abstract

Much recent work on the syntax-prosody interface has been based in Optimality Theory. The typical analysis explicitly considers only a small number of candidates that could reasonably be expected to be optimal under some ranking, often without an explicit definition of GEN. Manually generating all the possible candidates, however, is prohibitively time-consuming for most input structures – the Too Many Candidates Problem. Existing software for OT uses regular expressions for automated generation and evaluation of candidates. However, regular expressions are too low in the Chomsky Hierarchy of language types to represent trees of arbitrary size, which are needed for syntax-prosody work. This paper presents a new computational tool for research in this area: Syntax-Prosody in Optimality Theory (SPOT). For a given input, SPOT generates all prosodic parses under certain assumptions about GEN, and evaluates them against all constraints in CON. This allows for in-depth comparison of the typological predictions made by different theories of GEN and CON at the syntax-prosody interface.

Match Theory: An overview

“…Similar considerations apply to most markedness constraints proposed in the MT literature so far (Bellik & Kalivoda 2019), including EqualSisters (Myrberg, 2013) and StrongStart (Selkirk, 2011). Chen (2021), Bellik (to appear), and Kalivoda (to appear b) discuss multiple versions of these constraints, and show that rendering previous definitions unambiguous requires analytical choices that make divergent typological predictions.…”

“…For the right-headed input, the prosodic trees admitted by (14b) are ( φ ( φ ω B ) ω A ) and ( φ ω B ω A ). In a typical Match-Theoretic analysis, there would be many more candidates admitted by Gen (Bellik & Kalivoda, 2019), but for the system M.Sys we are bound by (14).…”

“…The SPOT application includes versions of all of the major constraints invoked in the MT and Alignment literature, programed as functions that automatically assign violations to candidates. For constraints that have not been clearly Constraint definitions in MT are not mere technical details, but can have major consequences (Bellik, to appear;Bellik & Kalivoda, 2019;Bellik & Van Handel, to appear;Kalivoda, to appear a;Kalivoda & Bellik, 2018). Take for example the binarity constraint definitions in (11), repeated as ( 27).…”