AceExperimental

Experimental Features in ACE

This page describes some features that have been added to ACE over the course of minor revisions that are not well-tested enough to be considered robust, but may be of interest to some users.

• The LogonTransfer system -- as of version 0.9.11, ACE has relatively complete support for this (see AceTransfer for implementation details). To translate from Japanese to English, for example:

$ ace -G jacy.dat -g logon/dfki/jacy/ace/config.tdl
[various output...]
$ ace -G erg.dat -g erg-1111/ace/config.tdl
[various output...]
$ ace -G jaen.dat -g logon/uio/tm/jaen/ace/config.tdl
[various output...]
$ echo "犬 が 猫 を 食べる" | ace -g jacy.dat -1Tf 2>/dev/null | ace -g jaen.dat 2>/dev/null | ace -g erg.dat -e1 2>/dev/null
The dog eats the cat.
$

Note that JaCY does not work properly out-of-the-box with ACE; you will need an SVN update at least as recent as late January 2013 of JaCY, and you will additionally need need to comment out the line including "lex/generics.tdl" in the file japanese.tdl. Also, to compile the ERG from a working directory other than the ACE directory (as shown above), you will need ace-0.9.13, and unknown word handling will not work.

• Detailed time profiling. When run with the -i option, ACE produces a report of how much time was spent in various system components. For parsing, chart mapping, and transfer, the amount of time spent in each rule is included. Most of the other measurements are somewhat coarser, such as "semantic lookup" and "idiom checking".

• Pre-generation transfer phase: input MRSes can undergo a set of transfer rules in a .mtr file defined by the 'generation-fixup-rules' configuration setting. This is fairly robust.

• Post-unpacking transfer phase: output MRSes can undergo a set of transfer rules in a .mtr file defined by the 'cleanup-rules' configuration setting.

• Post-generation token mapping phase: output surface strings can undergo a set of token mapping rules, defined by TDL status "post-generation-mapping-rule". This can be used, for example, to conjoin tokens like the possessive 's in English and to adjust uppercase/lowercase.

• Token mapping flow control: token mapping rules can have a +JUMP property that enables nonlinear processing through a set of rules (e.g. skipping a section if a condition is met, or looping).

There may not be enough information on this page to successfully try out some of these features. Contact WoodleyPackard if you want to deploy something and need help.