Are there NLP applications (either natural language understanding or generation) which make use of the Construction Grammar framework? On a related note, does there exist a logically organized list of constructions in English which could be applied to an NLG task, or a method of obtaining such a list?
I'm not at all up-to-date on theory-driven NLP, but…
First, "Construction Grammar" (CxG) isn't a framework; it's a family of distinct (sometimes very distinct) frameworks or approaches that share the same core idea.
Even sticking to the ones that are most theoretically aligned (usage-based, unification-driven, etc.), you can't take a set of constructions from Adele Goldberg in her CxG and mix and match with a set of constructions from Charles Fillmore in Berkeley Construction Grammar (BCG), without doing a lot of careful work to translate.1
There's also some question about which frameworks are "really" CxG, and some of the most useful work is in those borderline frameworks.
The most obvious example is Head-driven Phrase Structure Grammar (HPSG). While HPSG was originally designed by people outside the CxG universe, and described as a unification-driven bottom-up generative PSG, not a CxG—but it turns out to be very CxG-like in practice. In fact, it's easier to translate data between Berkeley Construction Grammar (BCG) and HPSG than between BCG and some other CxGs. So, is HPSG really a CxG after all? Or a hybrid?
If you don't care about the answer to that question, then there's a ton of NLP work driven by HPSG—projects like DELPH-IN and CoreGram/Babel, and the theoretical and data-collection work behind them—which could be directly useful to you.
But if it matters to you (and you don't believe that HPSG qualifies as a CxG), then you can't use any of that work. Some of it has translations to or successors in Sign-Based Construction Grammar, which is definitely a CxG, and can be seen as sort of a merger of BCG and HPSG, but definitely not all of it.
1. Correlating two CxG frameworks informally is usually pretty simple—Goldberg can write a paper with Ray Jackendoff that fits into both of their frameworks, with only a few footnotes to explain where the rough edges are. But doing so formally is not so easy.
I am not an expert on this, but plan on writing my dissertation on something similar. Since no one has answered yet, here are a couple of resources:
- https://www1.icsi.berkeley.edu/~miriamp/fillmore-tribute/slides/Collin_Baker.pdf - This talks about some NLP applications of FrameNet, which is based on CG
- https://www.fcg-net.org/ - Fluid CG, which is an instantiation of CG.
I'd love to hear about more, though.