ACE Command-line Options

Below are the most common command-line usages of ACE, followed by a complete reference for all the command-line options.

Parsing (input is one sentence per line): ace -g grammar.dat [input-file] [-1 | -n count]

Generating (input is one MRS per line): ace -g grammar.dat -e [input-file] [-1 | -n count]

Compiling a grammar: ace -G grammar.dat -g path-to-config.tdl

Complete Reference:




-g filename

Identify the grammar to be used. Normally, a compiled grammar image. When compiling a grammar, the location of that grammar's config.tdl file.


-G filename

Enable grammar-compilation mode, and specify where to save the compiled grammar.



Generate, rather than parse.


-n count

Use selective unpacking to enumerate only the top count results.


Equivalent to -n 1.

-r roots

Supplies a space-delimited list of root instances (overriding the setting compiled into the grammar image).


Enable LUI mode. You must have a working yzlui binary in your PATH.


Format each EP on its own line when printing MRSes.


Turn off packing.


Do not output results.


Only output MRSes, not trees.


Run as a [incr tsdb()] client.


Show a short synopsis of usage


Show ACE version number.


Increase verbosity; up to 3 -v options can be used if desired. The additional output is unlikely to be useful except for debugging.


Run the REPP preprocessor on each input and output the result; do not perform token mapping or any further parsing.

-m hostname

Join an arbiter grid of parsers.


Suppress printing the input sentence in cases of successful parsing.


Stop when the first result is found, rather than generating the complete packed forest. Not recommended.

-L lexemes

Disable certain lexemes when parsing.


Perform a fine-grained timing profile of the operation requested (experimental).



Enable YY input mode (experimental).



Use up to X megabytes of RAM on chart edges before terminating the forest creation portion of parsing.


Use up to X megabytes of total RAM for chart edges and unpacking operations before cancelling a parse.


Use up to X seconds of time before cancelling a parse. Off by default.


Turn off the subsumption check in generation (underspecified generation)


Turn off subsumption under generalization (useful for debugging)



enable ubertagging



use a pcfg approximation to parse (c-saw) and combine its results with the HPSG


include the root in the derivation (if multiple roots apply, the first listed in roots.tdl will be used)


decorate lexeme nodes with their type; "--udx=all" decorates rule nodes too



produce structured output as expected by tsdb



add a tree with the labels (e.g., DT, VP as specified in labels.tdl)



when in YY mode, require input tokens to undergo specific lexical rules



ignore sentences with more than X words (default 70)


AceOptions (last edited 2018-06-22 09:04:48 by FrancisBond)

(The DELPH-IN infrastructure is hosted at the University of Oslo)