TIMEX2 was initially developed in 2000 under DARPA’s Translingual Information Detection, Extraction and Summarization (TIDES) program. The fundamental move forward here (compared to TIMEX) was the addition of a normalisation task to the recognition task: the annotation provided for the interpretation of dates and times by using ISO 8601 as the standard for the representation of normalized dates and times.
The TIMEX2 specification was then used in the DARPA-sponsored Automatic Content Extraction (ACE) contest in 2002 as the annotation standard for time attributes of relations. TIMEX2 has subsequently been used in many other projects, including the Time Expression Recognition and Normalisation (TERN) task introduced at the ACE program in 2004. There are newer annotation standards under development that allow for wider coverage of temporal phenomena (for, example, TimeML), however, TIMEX2 is now a robust standard that is expressively adequate for many practical applications.
Subsequently, the TIMEX2 standard has evolved through a number of versions:
Similarly to TIMEX, TIMEX2 is used as inline annotation using TIMEX2 xml tag with possible six possible attributes; see the table below for details on these attributes.
| Attribute | Description |
|---|---|
| VAL | Contains a normalized form of the date or time in the annotated expression. |
| MOD | Captures temporal modifiers, using values such as BEFORE, MORE_THAN, START, and APPROX. |
| ANCHOR_VAL | Contains a normalized form of an anchoring date or time. |
| ANCHOR_DIR | Captures the relative direction or orientation between VAL and ANCHOR_VAL attributes, as in WITHIN, STARTING, and BEFORE. It is used to express the information about when a duration is placed. |
| SET | Identifies expressions denoting sets of times; either takes the value YES or is empty. |
| COMMENT | Contains any comment that the annotator wants to add to the annotation; ignored from the point of view of automatic processing of the text. |
