Defining Syntax for Learner Language Annotation

We discuss making syntactic annotation for learner language more precise, by clarifying the properties which the layers of annotation refer to. Building from previous proposals which split linguistic annotation into multiple layers to capture non-canonical properties of learner language, we lay out the questions which must be asked for grammatical annotation and provide some answers. Our investigation points to the layer of distributional syntax being based on properties of the target language (L2) and largely redundant with the other layers. We show, for example, that subcategorization seems to better be able to underspecify annotation for situations where no single correct solution can be found. While this paves the way for applying the annotation to larger corpus efforts, it also represents a significant step in elucidating syntax for non-canonical language.