SuquamishGrammarIndexing

Notes on discussion of Grammar Indexing for Phenomena

Brief presentation

The problem:

• Analyses of phenomena typically touch many different types/rules, and each type/rule typically bears constraints related to multiple phenomena.
• Grammar engineers and others might have many reasons to want to see where in a grammar a phenomenon is handled:

  • Learning from someone else's grammar

  • Cross-linguistic comparison: How does phenomenon X differ across grammars?

  • Cross-grammar comparison: How many phenomena are handled in grammars X, Y and Z?

  • Measuring degree of interaction between phenomena

  • Discoverability: How can I find grammars that have treatments of phenomenon Z?

  • Grammar engineering for language documentation (=> finding phenomena in a treebank)

A solution:

• Annotate grammars with labels on constraints (i.e., pieces of types) indexing the analyses they belong to

• Ideally using a vocabulary drawn from GOLD or a similar ontology (for discoverability)

But how?

• Build on lextype DB?
• Embedded in tdl or stand-off?
• Grammar Matrix customization system could auto-generate for library-based analyses (and to get people off on a good start)
• Retrofitting might be harder than building in in new grammars

Discussion