getRange() of SUTime.Temporal created with "this week" returns date with wrong year

[mirosval]

Looks like an off-by-one error somewhere.

When I initialize a TimeExpression with strings containing "week" such as "this week", "last week", "next week", the TimeExpression itself is correct, but then when I try to get the beginning of the range (see code) I get the correct week, but a year earlier (i.e. now for 2016-07-26 when I say "this week" I get 2016-07-W30 from TimeExpression, but 2015-07-20 from TimeExpression.getRange.begin)

Please note that stuff like "next monday" "today" "tomorrow" "last wednesday" works as expected.

import java.time.LocalDate

import edu.stanford.nlp.ling.CoreAnnotations
import edu.stanford.nlp.pipeline.{Annotation, AnnotationPipeline, POSTaggerAnnotator, TokenizerAnnotator, WordsToSentencesAnnotator}
import edu.stanford.nlp.time.{TimeAnnotations, TimeAnnotator, TimeExpression}

import scala.collection.JavaConversions._

val annotationPipeline = new AnnotationPipeline()
annotationPipeline.addAnnotator(new TokenizerAnnotator(false))
annotationPipeline.addAnnotator(new WordsToSentencesAnnotator(false))
annotationPipeline.addAnnotator(new POSTaggerAnnotator(false))
annotationPipeline.addAnnotator(new TimeAnnotator())

val annotation = new Annotation("this week")
annotation.set(classOf[CoreAnnotations.DocDateAnnotation], LocalDate.now.toString)
annotationPipeline.annotate(annotation)

val timeAnnotations = annotation.get(classOf[TimeAnnotations.TimexAnnotations]).map { ann =>
  ann.get(classOf[TimeExpression.Annotation]).getTemporal
}

println(timeAnnotations) // <- correct
println(timeAnnotations.head.getRange.begin) // <- off by one year

[mirosval]

I traced it further to JodaTimeUtils

import edu.stanford.nlp.time.JodaTimeUtils
import org.joda.time.{DateTimeFieldType, Partial}

val p = new Partial()
  .`with`(DateTimeFieldType.year(), 2016)
  .`with`(DateTimeFieldType.weekOfWeekyear(), 30)
  .`with`(DateTimeFieldType.dayOfWeek(), 1)

val t = JodaTimeUtils.resolveDowToDay(p) // <- 2015-07-20

But:

import edu.stanford.nlp.time.JodaTimeUtils
import org.joda.time.{DateTimeFieldType, Partial}

val p = new Partial()
  .`with`(DateTimeFieldType.year(), 2016)
  .`with`(DateTimeFieldType.weekOfWeekyear(), 30)

val t = JodaTimeUtils.resolveDowToDay(p) // <- [year=2016, weekOfWeekyear=30]

[mirosval]

More details:

The getInstant() function of JodaTimeUtils takes a Partial and if this Partial has a year, weekOfYear and a dayOfWeek set, it will not translate those values into the instant, using just the Year value it can understand (there are no attempts to even read weekOfYear).

And so the instant coming out is 2016-01-01 instead of mid 2016 and then Joda's ISOChronology.set() will resolve partial + instant to 2015 for some reason.

[J38]

Thanks for doing such a detailed analysis. I will look deeper into this and try to fix the issue soon.

[tkgesis]

A similar effect occurs when parsing Date ranges such as "March to July 1990".
cm.get(TimeExpression.Annotation.class).getTemporal().getRange().begin() will resolve to XXXX-3-XX ,
cm.get(TimeExpression.Annotation.class).getTemporal().getRange().end() to 1990-07

On the first, ..getTemporal().getTimexValue() will resolve to 2017-03 , while 1990-03 would be a better guess.

It could be addressed with an additional rule in defs.sutime.txt, which seems to be missing .

[kidchen]

Same issue found when parsing Date ranges like "last weekend". Year is not correct (one year less).

[tkgesis]

In an expression such as "... between January and April 1988" will be resolved to January 2016(Current year) to April 1988.

[J38]

Look at the current code I am seeing the correct behavior for these three sentences:

"I went to Los Angeles last weekend."
"I am going to Los Angeles this week."
"I went to Los Angeles last week."

We still have not fixed "between January and April 1988"...

[mujtabahasan]

11th dec gives year 2017. i checked with different dates. any date after 31 aug gives previous year in the output.