Grounding statistical machine translation with semantic parsing
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Failed to load latest commit information.


Code for grounded SMT on geoquery data or Free917 data.

(N.b. This has nothing to do with the programming language REBOL [0]!)


###Geoquery: WASP-1.0 includes the geoquery knowledge base and scripts for querying it. The evaluation scripts were slightly modified to produce full outputs. These scripts can be found in data/geoquery/wasp/, and they go into wasp-1.0/data/geo-funql/eval/. WASP-1.0 can be downloaded from here [1].

You'll also need a Prolog environment for geoquery, e.g. SWI-Prolog [2].

We use the a slightly modified implementation of smt-semparse, as described in Semantic parsing as machine translation (Andreas et al, ACL 2013). Our fork can be found here [3]. This software depends on more stuff, e.g. the Moses decoder and SRILM.

###Free917: The parser used for Free917 is sempre, as described in Semantic Parsing on Freebase from Question-Answer Pairs (Berant et al, EMNLP 2013). It can be downloaded here [4] and further dependencies can be found in

###Both: For translation we use the cdec toolkit, [5].

As semantic parsing is quite slow and rebol does it quite often, results are cached with memcached [6].

You'll need the following ruby gems: