The Q-RKS time converter process

Whatever the types, theQ-RKS time converter process finds the definite DATE/-TIME for Q-RKS in the DATE/-TIMEX3 format. Below we describe the process for finding the definite DATE/TIME for each type of temporal keyword.

• Explicit temporal keyword: If Q-RKS input keywords hold the Type-1 keyword , i.e., keywords with a date and time, the Q-RKS time converter generates its definite DATE/TIME in TIMEX3 format by parsing thesignal

word following part of the Type-1 keyword . We refer to this TIMEX3 value as Q-RKS exp.

The Q#1 belongs to such a case, where the Q-RKS time converter process converts thesignal word following part of the Type-1 keyword (i.e., “the 29^th of August 1958”) to their corresponding TIMEX3 value as 1958-08-29.

• Event temporal keyword: We use BoTLRet to convert the event tem-poral keyword to its definite DATE/TIME. So, if Q-RKS input keywords hold the Type-2 keyword , we first execute BoTLRet over the signal word following part of the event temporal keyword. Then, by using a language parser, the output of BoTLRet is parsed. Over the parsed output, we cap-ture the DATE/TIME/DURATION type normalized NER (Named Entity Recognizer) value. These NER values are considered as definite DATE/-TIME of event temporal keyword. We convert the NER values in TIMEX3 format and refer as event temporal keyword corresponding Q-RKS exp.

For example, the Q#2 belongs to a case where the Q-RKS time converter converts the Type-2 keyword to the corresponding TIMEX3 values. The values are 1914-07-28, 1919-06-28, 1919-09-10, 1919-11-27, 1920-06-04, and 1920-08-10. Each of these dates are somehow related to “World War I”:

date 1914-07-28 is related to the armistice with Germany, date 1919-06-28 is related to signing the treaty of Versailles, and so on.

• Relative temporal keyword: Since the Type-3 keyword holds two tempo-ral values at the same time, finding the definite DATE/TIME requires some kind of adjustment. Therefore, to find the definite DATE/TIME for a Type-3 keyword, we retrieve its reference temporal value and adjustment temporal value. We adjust the reference temporal value by adding or subtracting the adjustment temporal value, which gives us the definite DATE/TIME. Be-low, first we describe the retrieval of the reference temporal value and the adjustment temporal value. Then, we describe their adjustment.

– Retrieval of thereference temporal value and theadjustment temporal value: To find thereference temporal value from the Type-3 explicit keyword, we execute the Q-RKS time converter process over the second signal word following part of the keywords. This gener-ates a TIMEX3 value that is used as the reference temporal value.

On the other hand, to find the adjustment temporal value from the

Type-3 explicit keyword, we execute a language parser over the key-words between the first signal word and the second signal word. From the parsed output, we collect the NER values for the DURATION/-DATE/TIME/NUMBER type NER, and then convert the NER val-ues as TIMEX3. However, we find that some language parsers, though able to identify the DURATION type NER, cannot generate a TIMEX3 value for the DURATION type NER. In such a case, we also execute the Tarsqi Toolkit⁴ over the keywords between the first signal word and the second signal word. The Tarsqi Toolkit confirms any missing DURATION type NER.

For example, for the Type-3 explicit keyword of the Q#3, the temporal keyword is “after 2 decades of Iranian revolution start date”, the first signal word is “after” and the second signal word is “of”. The execu-tion of theQ-RKS time converter process over the keywords “Iranian revolution start date” generates the TIMEX3 reference temporal value as 1978-01-07. On the other hand, the execution of the Stanford lan-guage parser ([22]) over the keywords “2 decades” can identify NER as the DURATION type, which should be theadjustment temporal value.

But the parser cannot generate the TIMEX3 value. Therefore, execu-tion of the Tarsqi Toolkit over the keywords “2 decades” generates the TIMEX3 value “P20Y”. By the TIMEX3 convention, this means 20 years duration, where “P” indicates duration and “Y” indicates year.

For the Type-3 implicit keyword, both the reference temporal value and theadjustment temporal value come from its temporal adverb. In such a case, thereference temporal value is always the current date and time. To find thereference temporal value, we convert the current date and time into their TIMEX3 value. To find the adjustment temporal value, we execute a language parser over the signal word following the keywords. Here, finding the adjustment temporal value in TIMEX3 is the same as described in the above paragraph. For example, for the Type-3 implicit keyword of the Q#4, we get the reference temporal value as 2016-03-05⁵ and the adjustment temporal value as “P100Y”.

– Getting the definite DATE/TIME: After finding thereference tem-poral value and the adjustment temporal value, we adjust them to get

4http://www.timeml.org/site/tarsqi/toolkit/

5Date of execution 5^th March 2016.

the definite DATE/TIME, Q-RKS exp. The adjustment between the reference temporal value and the adjustment temporal value is carried out by the (first) signal word of the Type-3 keyword and the guide-line of the TIMEX3 time convention⁶. If the first signal word means

“before”, we subtract the reference temporal value and the adjustment temporal value; if it means “after” we add these values. For exam-ple, as the definite DATE/TIME of the Q#3 Q-RKS (“after 2 decades of Iranian revolution start date”), we get 1998-01-07. Here the first signal word after thereference temporal value is 1978-01-07 and the ad-justment temporal value is “P20Y”. Therefore, adding 1978-01-07 and

“P20Y” generates the TIMEX3 value as 1998-01-07.

Finally, as the end result of phase 1, we get keywords for the Q-FKS and the definite DATE/TIME for the Q-RKS (Q-RKS exp).

Next, we talk about phase 2 of the TLDRet, which generates the final result. If it is found that the input keywords only belong to Q-RKS in phase 1, the output of the Q-RKS time converter process is considered as the final result. In such a case, we do not require phase 2.