TomarFormalism

Discussion: HPSG Formalism Harmonization: Implications, Relations, Macros?

Background

Within HPSG, currently, (at least) three sub-communities working on computational grammars:

• Alpino (Groningen):

• CoreGram (Berlin, et al.): TRALE (plus various add-ons, e.g. UTool and [incr tsdb()])

• DELPH-IN:

These groups assume different formalisms in their work (and, correspondingly, different interpretations of HPSG). Among the above, the DELPH-IN Reference Formalism arguably is the most restrictive selection of descriptive devices found in the theoretical HPSG literature; this selection was largely made in the mid-1990s.

Alpino is somewhat special, as local to Groningen and focused on Dutch only; very close collaboration between tool developers and grammarians over two decades: broad-coverage grammar and fairly efficient parsing. In comparison to CoreGram, several DELPH-IN grammars have much broader coverage, and can be processed magnitudes more efficiently.

Stefan Mueller argues (forcefully) that, in order to allow “rather direct implementation of analyses that are proposed by theoretical linguists”, the following descriptive devices are indenspensible:

• phonologically empty elements
• relational constraints
• implications with complex antecedents
• cyclic feature structures
• macros (aka templates)
• a more expressive morphological componet

From an in-depth discussion that Stephan had with Stefan, formalism differences are a major obstacle to collaboration with the CoreGram group and, possibly, also a barrier to establishing agreement within the larger HPSG community on the ‘standard’ approach and infrastructure for computational grammar engineering.

Do we feel like revisiting the DELPH-IN reference formalism, presumably in a spirit of backwards-compatible, monotonic extension?

Are any of the above devices (strongly) attractive to DELPH-IN members?

Would we know how to implement them (with reasonable efficiency)?

As background, here are some slides from an earlier discussion in a similar spirit.

Discussion Notes