ucto(1)

NAME

ucto - Unicode Tokenizer

SYNOPSIS

ucto [[options]] [input‐file] [[output‐file]]

DESCRIPTION

ucto tokenizes text files: it separates words from punctuation, splits sentences (and optionally paragraphs), and finds paired quotes. Ucto is preconfigured with tokenisation rules for several languages.

Those rules are provided by uctodata

OPTIONS

-c configfile

read settings from a 'configfile'

-B

run in batch mode. Process all inputfiles to an output directory specified with -O.

-d value

set debug mode to 'value'

-e value

set input encoding. (default UTF8)

-I value

set the input directory to 'value'. (batch mode only)

-O value

set the ouput directory to 'value'. (Required for batch mode)

-N value

set UTF8 output normalization. (default NFC)

--filter=[YES|NO]

disable filtering of special characters, (default YES) These special characters can be specified in the [FILTER] block of the configuration file.

-L language

Automatically selects a configuration file by language code. The language code is generally a three-letter iso-639-3 code. For example, 'fra' will select the file tokconfig‐fra from the installation directory

--detectlanguages=<lang1,lang2,..langn>

try to detect all the specified languages. The default language will be 'lang1'. (only useful for FoLiA output).

All values must be iso-639-3 codes.

You can also use the special language code `und`. This ensures there is NO default language, and any language that is NOT in the list will remain unanalyzed.

Warning: To be able to handle utterances of mixed language, Ucto uses a simple sentence splitter based on the markers '.' '?' and '!'. This may occasionally lead to surprising results.

-l

Convert output text to all lowercase

-u

Convert all input text to all uppercase

-n

Emit one sentence per line on output

-m

Assume one sentence per line on input

--normalize=class1,class2,..,classn

map all occurrences of tokens with class1,...class to their generic names. e.g --normalize=DATE will map all dates to the word {{DATE}}. Very useful to normalize tokens like URL's, DATE's, E-mail addresses and so on.

-T value or --textredundancy=value

set text redundancy level for text nodes in FoLiA output:
'full' - add text to all levels: <p> <s> <w> etc.
'minimal' - don't introduce text on higher levels, but retain what is already
there.
'none' - only introduce text on <w>, AND remove all text from higher levels

--allow-word-correction

Allow ucto to tokenize inside FoLiA Word elements, creating FoLiA Corrections

--ignore-tag-hints

Skip all tag=token hints from the FoLiA input. These hints can be used to signal text markup like subscript and superscript

--add-tokens="file"

Add additional tokens to the [TOKENS] block of the default language. The file should contain one TOKEN per line.

--passthru

Don't tokenize, but perform input decoding and simple token role detection

--filterpunct

remove most of the punctuation from the output. (not from abreviations and embedded punctuation like John's)

-P

Disable Paragraph Detection

-Q

Enable Quote Detection. (this is experimental and may lead to unexpected results)

-s <string>

Set End‐of‐sentence marker. (Default <utt>)

-V or -- version

Show version information

-v

set Verbose mode

-F

The input file(s) are assumed to be FoLiA XML. Text in the correct 'inputclass' will be tokenized. For files with an '.xml' extension, -F is the default.

In batch mode, this forces to only select files with the '.xml' extension from the input directory.

--inputclass="cls"

When tokenizing a FoLiA XML document, search for text nodes of class 'cls'. The default is "current".

--outputclass="cls"

When tokenizing a FoLiA XML document, output the tokenized text in text nodes with 'cls'. The default is "current". It is recommended to have different classes for input and output.

--textclass="cls"(obsolete)

use 'cls' for input and output of text from FoLiA. Equivalent to both --inputclass='cls' and --outputclass='cls')

This option is obsolete and NOT recommended. Please use the separate --inputclass= and --outputclass options.

--copyclass

when ucto is used on FoLiA with fully tokenized text in inputclass='inputclass', no text in textclass 'outputclass' is produced. (A warning will be given). To circumvent this. Add the --copyclass option. Which assures that text will be emitted in that class

-X

All output will be FoLiA XML. Document id's are autogenerated.

Works in batch mode too.

--id <DocId>

Use the specified Document ID for the FoLiA XML. (not allowed in batch mode) When not provided, a document is is generated based on the nema of the input file.

BUGS

likely

AUTHORS

Maarten van Gompel

Ko van der Sloot

e-mail: [email protected]