NlpSemanticParsingDatetimeDateTime

For temporal expressions that consist of components with different types, the elements in the composition are nested according to the order they should be grounded/resolved. For example, "tomorrow at 8am" has a relative component [tomorrow] and an absolute datetime component [8am] so the annotation will be represented in the following way, indicating that the relative datetime for tomorrow should be grounded first, and then resolve the actual 8am point for that given date: point { hour: 8 hour_state: UNAMBIGUOUS properties { time_format: AM_PM_TOKEN } } composition_element { relative { fetched { ordinal: 1 target { unit: DAY } base_type: CURRENT_DATETIME } } grounding_stage: UNGROUNDED } grounding_stage: UNGROUNDED_COMPOSITION The composition_element field will be populated only when the grounding_stage is set to UNGROUNDED_COMPOSITION and it will hold the nested DateTime value for the rest the compositional expression. More details in: go/datetime-resolution-decoupling.

Deprecated fields. Do NOT use.

This field of the DateTime message should not in general be used by outside clients of the grammar. It is intended to be used internally in Aqua for evaluation purposes. The rationale is that token counts depend on the particular tokenization used in Aqua which may be different from the one used by the client and may change from time to time. Outside clients should not create a dependency on the current tokenization used in Aqua.

Note that there is a difference between this scenario and an ambiguous date/time expression. The latter is resolved to multiple proto messages, not multiple values within one proto message. To be concrete, consider "Monday" in "Monday football". It is ambiguous and can be reasonably resolved to "Monday last week," "Monday this week" and "Monday next week." The 3 values are represented as 3 separate DateTime messages, not 3 values within one DateTime message.

See comments of Span.