jennhu / tag

Automating tree-adjoining grammar analyses of linguistic phenomena.

Automating synchronous multicomponent tree-adjoining grammar derivations

The system allows a user to specify a synchronous multicomponent tree-adjoining grammar (STAG) using an intuitive, text-based metagrammar. It then automates the specified derivations and outputs the derived trees.

This README is only meant to get you started. Refer to doc/mctagdoc.pdf for in-depth documentation. These slides may also be helpful.

Getting started

Dependencies

You will need OCaml 4.02 (or later) and Menhir. Refer to the OCaml docs for installation instructions. If you install the package manager OPAM with OCaml (highly recommended), Menhir can be installed with the following command:

opam install menhir

You may compile the files with your favorite OCaml compiler, but I suggest OCamlbuild, which can be installed with the following command:

opam install ocamlbuild

Building

To build all the files for the first time, run the following command:

ocamlbuild -use-menhir mctag.byte

This command creates a _build directory within your current working directory. The executables will be stored there, but a symlink should be created in your current directory. If you can't find the target mctag.byte file, look under _build.

Usage

To help the user get off the ground, the repo contains several sample grammars in the grammars folder. Try running

./mctag.byte -i grammars/toy.txt

to print the derived syntax and semantics trees for the sentence 'John saw a girl'.

A full list of example grammars is provided below.

Path	Description
grammars/toy.txt	'John saw a girl'
grammars/example.txt	Reflexives, topicalization, quantifier scope ambiguity, adjunction at substitution nodes, etc.
grammars/paper.txt	Trees from Aggazzotti & Shieber (2017)
grammars/features.txt	Basic analyses of features -- see doc for more details

Writing your own grammars

You can specify your own STAGs using the metagrammar specified in the full documentation under doc/mctagdoc.pdf.

Acknowledgments

This system was developed in collaboration with Stuart Shieber and Cristina Aggazotti during the summer of 2016.