Errors with geological time data

[seongjinpark-88]

@MihaiSurdeanu

Hi,

I am currently working on geological data, and the data contains the temporal expressions like "cold period", "2.65 million years ago", or "since Paleogene" (Paleogene: 66 million years ago ~ 23.03 million years ago). When I try to normalize those temporal expressions in the data, it showed me the following errors:

scala> import org.clulab.timenorm.scate._
import org.clulab.timenorm.scate._
scala> val parser = new TemporalNeuralParser
parser: org.clulab.timenorm.scate.TemporalNeuralParser = org.clulab.timenorm.scate.TemporalNeuralParser@35d1457d

scala> val anchor = SimpleInterval.of(2019, 11, 20)
anchor: org.clulab.timenorm.scate.SimpleInterval = SimpleInterval(2019-11-20T00:00,2019-11-21T00:00)

scala> val text = "We searched for correlations between timing in diversification and timing of (1) a period of marked volcanism across the Trans-Mexican Volcanic Belt in central Mexico 37.5 million years ago (Ma)"
text: String = We searched for correlations between timing in diversification and timing of (1) a period of marked volcanism across the Trans-Mexican Volcanic Belt in central Mexico 37.5 million years ago (Ma)

scala> for (timex <- parser.parse(text, anchor)) timex match{
     | case interval: Interval =>
     | val Some((charStart, charEnd)) = interval.charSpan
     | println(s"${interval.start} ${interval.end} ${text.substring(charStart, charEnd)}")
     | }
scala.MatchError: periods (of class java.lang.String)
  at org.clulab.timenorm.scate.AnaforaReader.period(Readers.scala:94)
  at org.clulab.timenorm.scate.AnaforaReader.temporal(Readers.scala:306)
  at org.clulab.timenorm.scate.TemporalNeuralParser$$anonfun$parseBatch$1$$anonfun$apply$8.apply(TemporalNeuralParser.scala:154)
  at org.clulab.timenorm.scate.TemporalNeuralParser$$anonfun$parseBatch$1$$anonfun$apply$8.apply(TemporalNeuralParser.scala:154)
  at scala.collection.TraversableLike$$anonfun$map$1.apply(TraversableLike.scala:245)
  at scala.collection.TraversableLike$$anonfun$map$1.apply(TraversableLike.scala:245)
  at scala.collection.Iterator$class.foreach(Iterator.scala:750)
  at scala.collection.AbstractIterator.foreach(Iterator.scala:1202)
  at scala.collection.IterableLike$class.foreach(IterableLike.scala:72)
  at scala.collection.AbstractIterable.foreach(Iterable.scala:54)
  at scala.collection.TraversableLike$class.map(TraversableLike.scala:245)
  at scala.collection.AbstractTraversable.map(Traversable.scala:104)
  at org.clulab.timenorm.scate.TemporalNeuralParser$$anonfun$parseBatch$1.apply(TemporalNeuralParser.scala:154)
  at org.clulab.timenorm.scate.TemporalNeuralParser$$anonfun$parseBatch$1.apply(TemporalNeuralParser.scala:151)
  at scala.collection.TraversableLike$$anonfun$map$1.apply(TraversableLike.scala:245)
  at scala.collection.TraversableLike$$anonfun$map$1.apply(TraversableLike.scala:245)
  at scala.collection.IndexedSeqOptimized$class.foreach(IndexedSeqOptimized.scala:33)
  at scala.collection.mutable.ArrayOps$ofRef.foreach(ArrayOps.scala:186)
  at scala.collection.TraversableLike$class.map(TraversableLike.scala:245)
  at scala.collection.mutable.ArrayOps$ofRef.map(ArrayOps.scala:186)
  at org.clulab.timenorm.scate.TemporalNeuralParser.parseBatch(TemporalNeuralParser.scala:151)
  at org.clulab.timenorm.scate.TemporalNeuralParser.parse(TemporalNeuralParser.scala:137)
  ... 49 elided

or

scala> import org.clulab.timenorm.scate._
import org.clulab.timenorm.scate._

scala> val parser = new TemporalNeuralParser
parser: org.clulab.timenorm.scate.TemporalNeuralParser = org.clulab.timenorm.scate.TemporalNeuralParser@6ded3740

scala> val anchor = SimpleInterval.of(2019, 11, 20)
anchor: org.clulab.timenorm.scate.SimpleInterval = SimpleInterval(2019-11-20T00:00,2019-11-21T00:00)

scala> val text = "marked volcanism across the Trans-Mexican Volcanic Belt in central Mexico 37.5 million years ago"
text: String = marked volcanism across the Trans-Mexican Volcanic Belt in central Mexico 37.5 million years ago

scala> for (timex <- parser.parse(text, anchor)) timex match {
     | case interval: Interval =>
     | val Some((charStart, charEnd)) = interval.charSpan
     | println(s"${interval.start} ${interval.end} ${text.substring(charStart, charEnd)}")
     | }
2019-11-21 13:20:46.711084: I tensorflow/core/platform/cpu_feature_guard.cc:141] Your CPU supports instructions that this TensorFlow binary was not compiled to use: SSE4.2 AVX AVX2 FMA
scala.MatchError: FractionalNumber(37,1,2,Some((74,78))) (of class org.clulab.timenorm.scate.FractionalNumber)
  at $anonfun$1.apply(<console>:14)
  at $anonfun$1.apply(<console>:14)
  at scala.collection.IndexedSeqOptimized$class.foreach(IndexedSeqOptimized.scala:33)
  at scala.collection.mutable.ArrayOps$ofRef.foreach(ArrayOps.scala:186)
  ... 49 elided

And for the different sentence, it showed following error messages:

scala> val text = "major volcanic activity at DSDP Site 216 on Ninetyeast Ridge (Kerguelen hotspot) where volcanic sediments above basement basalt are dated ~69.5 Ma (nannofossil zone UC20a, planktic foraminiferal zones CF5CF4) and continued for about 2 million years."
text: String = major volcanic activity at DSDP Site 216 on Ninetyeast Ridge (Kerguelen hotspot) where volcanic sediments above basement basalt are dated ~69.5 Ma (nannofossil zone UC20a, planktic foraminiferal zones CF5CF4) and continued for about 2 million years.

scala> for (timex <- parser.parse(text, anchor)) timex match {
     | case interval: Interval =>
     | val Some((charStart, charEnd)) = interval.charSpan
     | println(s"${interval.start} ${interval.end} ${text.substring(charStart, charEnd)}")
     | }
org.clulab.timenorm.scate.AnaforaReader$Exception: cannot parse RepeatingInterval from "Some(million)" and Vector(<entity>
          <id>2@id</id>
          <span>235,242</span>
          <type>Frequency</type>
          <properties>
            <Number>1@id</Number><Type>million</Type>
          </properties>
        </entity>, <entity>
          <id>1@id</id>
          <span>235,242</span>
          <type>Number</type>
          <properties>
            <Value>1000000</Value>
          </properties>
        </entity>)
  at org.clulab.timenorm.scate.AnaforaReader.repeatingInterval(Readers.scala:280)
  at org.clulab.timenorm.scate.AnaforaReader.temporal(Readers.scala:316)
  at org.clulab.timenorm.scate.TemporalNeuralParser$$anonfun$parseBatch$1$$anonfun$apply$8.apply(TemporalNeuralParser.scala:154)
  at org.clulab.timenorm.scate.TemporalNeuralParser$$anonfun$parseBatch$1$$anonfun$apply$8.apply(TemporalNeuralParser.scala:154)
  at scala.collection.TraversableLike$$anonfun$map$1.apply(TraversableLike.scala:245)
  at scala.collection.TraversableLike$$anonfun$map$1.apply(TraversableLike.scala:245)
  at scala.collection.Iterator$class.foreach(Iterator.scala:750)
  at scala.collection.AbstractIterator.foreach(Iterator.scala:1202)
  at scala.collection.IterableLike$class.foreach(IterableLike.scala:72)
  at scala.collection.AbstractIterable.foreach(Iterable.scala:54)
  at scala.collection.TraversableLike$class.map(TraversableLike.scala:245)
  at scala.collection.AbstractTraversable.map(Traversable.scala:104)
  at org.clulab.timenorm.scate.TemporalNeuralParser$$anonfun$parseBatch$1.apply(TemporalNeuralParser.scala:154)
  at org.clulab.timenorm.scate.TemporalNeuralParser$$anonfun$parseBatch$1.apply(TemporalNeuralParser.scala:151)
  at scala.collection.TraversableLike$$anonfun$map$1.apply(TraversableLike.scala:245)
  at scala.collection.TraversableLike$$anonfun$map$1.apply(TraversableLike.scala:245)
  at scala.collection.IndexedSeqOptimized$class.foreach(IndexedSeqOptimized.scala:33)
  at scala.collection.mutable.ArrayOps$ofRef.foreach(ArrayOps.scala:186)
  at scala.collection.TraversableLike$class.map(TraversableLike.scala:245)
  at scala.collection.mutable.ArrayOps$ofRef.map(ArrayOps.scala:186)
  at org.clulab.timenorm.scate.TemporalNeuralParser.parseBatch(TemporalNeuralParser.scala:151)
  at org.clulab.timenorm.scate.TemporalNeuralParser.parse(TemporalNeuralParser.scala:137)
  ... 49 elided

When I fed the sentence I created, it showed me the result with the error message. Even though it showed me the result, I think the result is different from what it should be.

scala> val text = "I haven't done it since 4.5 million years ago"
text: String = I haven't done it since 4.5 million years ago

scala> for (timex <- parser.parse(text, anchor)) timex match{
     | case interval: Interval =>
     | val Some((charStart, charEnd)) = interval.charSpan
     | println(s"${interval.start} ${interval.end} ${text.substring(charStart, charEnd)}")
     | }
2019-11-21 13:06:46.315482: I tensorflow/core/platform/cpu_feature_guard.cc:141] Your CPU supports instructions that this TensorFlow binary was not compiled to use: SSE4.2 AVX AVX2 FMA
-997981-05-22T00:00 2019-11-21T00:00 since 4.5 million years ago
scala.MatchError: FractionalNumber(4,1,2,Some((24,27))) (of class org.clulab.timenorm.scate.FractionalNumber)
  at $anonfun$1.apply(<console>:14)
  at $anonfun$1.apply(<console>:14)
  at scala.collection.IndexedSeqOptimized$class.foreach(IndexedSeqOptimized.scala:33)
  at scala.collection.mutable.ArrayOps$ofRef.foreach(ArrayOps.scala:186)
  ... 49 elided

Do you have any thoughts on this type of issue?

Thanks,

[bethard]

The first one is a bug in our handling of the Type of Period operations (it should be UNKNOWN, not periods:

scala> parser.parseToXML("We searched for correlations between timing in diversification and timing of (1) a period of marked volcanism across the Trans-Mexican Volcanic Belt in central Mexico 37.5 million years ago (Ma)")
res5: scala.xml.Elem =
<data>
        <annotations>
          <entity>
          <id>0@id</id>
          <span>83,89</span>
          <type>Period</type>
          <properties>
            <Type>periods</Type>
          </properties>
        </entity>...

The second is because we very little training data on fractional numbers, so the model does not handle them well. In this case, it actually produces a parse, but the parse is wrong because it doesn't connect the 37.5 to anything.:

scala> parser.parse("marked volcanism across the Trans-Mexican Volcanic Belt in central Mexico 37.5 million years ago", anchor)
res8: Array[org.clulab.timenorm.scate.TimeExpression] = Array(FractionalNumber(37,1,2,Some((74,78))), BeforeP(SimpleInterval(2019-11-20T00:00,2019-11-21T00:00),SimplePeriod(Years,IntNumber(1000000,Some((79,86))),None,Some((87,92))),Some((93,96))))

The third is another wrong parse, where it has identified "million" as both a Frequency and a Number and doesn't know how to link it up propery:

scala> print(parser.parseToXML("major volcanic activity at DSDP Site 216 on Ninetyeast Ridge (Kerguelen hotspot) where volcanic sediments above basement basalt are dated ~69.5 Ma (nannofossil zone UC20a, planktic foraminiferal zones CF5CF4) and continued for about 2 million years."))
<data>
        <annotations>
          <entity>
          <id>0@id</id>
          <span>233,234</span>
          <type>Number</type>
          <properties>
            <Value>2</Value>
          </properties>
        </entity><entity>
          <id>1@id</id>
          <span>235,242</span>
          <type>Number</type>
          <properties>
            <Value>1000000</Value>
          </properties>
        </entity><entity>
          <id>2@id</id>
          <span>235,242</span>
          <type>Frequency</type>
          <properties>
            <Number>1@id</Number><Type>million</Type>
          </properties>
        </entity><entity>
          <id>3@id</id>
          <span>243,248</span>
          <type>Period</type>
          <properties>
            <Number>0@id</Number><Type>Years</Type>
          </properties>
        </entity>
        </annotations>
      </data>

Your last example is again the problem that the parser doesn't handle fractional numbers well.

@EgoLaparra: Any thoughts on how to fix any of these?