SUTIME: A Library for Recognizing and Normalizing Time Expressions
Detecting Temporal Expressions
There are 2 steps for extracting time from natural language expressions -
- Detecting temporal expressions.
- Extracting normalized time expressions from natural language phrases.
This paper is about the latter. They use TIMEX representation of time expression.
Types of Temporal Expressions
There are 4 types of temporal expressions -
Time - A time point indicating a particular instance on a time scale. Examples include Nov 20 1988, 5:30 PM, now, next weekend or Sunday Evening.
Duration - The amount of intervening time between the two end-points of a time interval. Examples include 5 days, 10 months, a decade, few years or 2-3 months.
Interval - A range of time defined by a start and end time points. Examples include Jan 2021 to Feb 2022.
Set - A set of temporals. Examples include “every 3rd Sunday”.
Datasets and Evaluation
- A popular dataset is TempEval-2. SUTime had the best Recall and F1 on this dataset but HeidelTime had the best precision. They’ve incorporated the work into Stanford CoreNLP.
Code. Learn. Explore