Spanish TimeNorm

This is a fork of the TimeNorm SCFG normalization system that adds the Spanish TimeNorm SFCG. It works as the English and Italian versions, which are also included in this fork.

The project is related to XTN Multilingual Timex Detection and Normalization, as described in

Nayla Escribano, German Rigau and Rodrigo Agerri. 2023. A modular approach for multilingual timex detection and normalization using deep learning and grammar-based methods. In Knowledge-Based Systems, 110612, ISSN 0950-7051, https://doi.org/10.1016/j.knosys.2023.110612.

Preparing the environment for using TimeNorm SCFG

To use the TimeNorm SCFG system from the Scala console:

1. Clone this repository.

2. Check that your Scala version is the one in build.sbt.

3. Compile the program from the main directory using sbt package.

4. Open the Scala console using scala -cp target/scala-[version]/timenorm_[version].jar.

Once you open the Scala console, you should be able to use TimeNorm SCFG to normalize single or multiple timexes.

How to normalize single timexes

To normalize a single timex, just choose the language of the grammar, set a temporal anchor and enter the timex. This provides a TimeNorm normalization value, which can be converted to TimeML format.

scala> import org.clulab.timenorm.scfg._, scala.util.Success
import org.clulab.timenorm.scfg._
import scala.util.Success

scala> val parser = TemporalExpressionParser.es // Choose the grammar
parser: org.clulab.timenorm.scfg.TemporalExpressionParser = org.clulab.timenorm.scfg.TemporalExpressionParser@1815577b

scala> val anchor = TimeSpan.of(2013, 1, 4) // Set the anchor
anchor: org.clulab.timenorm.scfg.TimeSpan = TimeSpan(2013-01-04T00:00Z,2013-01-05T00:00Z,Period(Map(Days -> 1),Exact),Exact)

scala> val Success(temporal) = parser.parse("hace dos semanas", anchor) // Input an expression
temporal: org.clulab.timenorm.scfg.Temporal = TimeSpan(2012-12-17T00:00Z,2012-12-24T00:00Z,Period(Map(Weeks -> 1),Exact),Exact)

scala> temporal.timeMLValue // Obtain TimeML value
res0: String = 2012-W51

How to normalize multiple timexes

To normalize and evaluate multiple timexes from a file, use the Evaluator. Input file must be in "timex [type] gold_value" tab-separated format, timexes from different documents should be separated by newlines and the anchor has to be the first timex of each set of document timexes. Files in the datasets repository can be used as input files for testing.

scala> val evaluator = Evaluator
evaluator: Evaluator.type = Evaluator$@29f3185c

scala> evaluator.main("es", "input_file.tsv", "output_file.tsv")

The output of the Evaluator should be something like this, showing the expression, the gold value and the predicted value of each timex and the final evaluation:

...

2000-09-15	2000-09-15	2000-09-15
2008	2008	2008
julio del próximo año	2001-07	2001-07
hoy	2000-09-15	2000-09-15
el año 2007	2007	2007
1970	1970	1970
2002	2002	2002

Number of timexes (also DCTs):    234
Correct normalizations:           195
Incorrect normalizations:          39
Accuracy:                       83.33

More info

See the original TimeNorm page for more information.

This fork adapts the TimeNorm SCFG version presented in

Steven Bethard. 2013. A Synchronous Context Free Grammar for Time Normalization. In: Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing, pp. 821-826.