The Reminder NLP API engine tries its best to resolve the intended date and time submitted in a phrase into a future reminder, however, machines are not perfect and sometime make mistakes. The JSON this API returns contains a very specific date and time, as well as hints about the resolution.
For example, let's say it is 9 p.m. and a user submits "don't forget to send this out before 6". Did the user mean 6 a.m. of the next day? or 6 p.m.? Our NLP will pick the next available option (6 AM) and the JSON will contain AMBIGUOUS_TIME_OF_DAY hint. This will allow the developer to take further actions to clarify ambiguities with the users - if needed.
If you believe that the resolution is wrong, post it here with:
The phrase (utterance) submitted The result received (you can copy/paste the JSON) The result expected
First, great API. I have been struggling with taking the spoken sentence apart and construct a logical time and date - there are sooooo many possibilities. I successfully integrated the API and ran some tests with very satisfying results. Of course, here is the "however"...
Example: "wake <name> up at 6 am on weekdays starting next Monday" gives me the following result:
body:"wake <name> up on weekdays starting" day:"4" hint:"AMBIGUOUS_TIME_OF_DAY,USE_CALL_TO_ACTION,TIME_UPDATE,ASK_FOR_TIME" hour:"6" minute:"0" month:"4" recurring:"no" reqtime:"3" status:"EVENT_DETECTED" timeZone:"GMT-04:00" type:"reminder" utcdate:"1459764031523" utterance:"wake <name>up at 6 am on weekdays starting next Monday" weekday:"MONDAY" year:"2016"
today's date is 3/22/2016 that would make next Monday the 28th of March, reccuring should be ON, the utcdate reflects that it is a week later than next Monday Mon, 04 Apr 2016 10:00:31.523 GMT
Also I noticed that on prior test that the hint:AMBIGUOUS_TIME_OF_DAY was set although the utterance read "wake me up every morning at 6" which should be resolved to 6 AM, also if the computer recognized Monday as a word but translates it to lower case monday the reply omits the word and it is not recognized.
I do verify the users request, but only verbally, in a question and answer routine. For example: user: "wake me up at 6am" computer: "here is what I understood: wake you up at 6am, say yes or no" => yes call the API. From my end I can perhaps check that weekdays are uppercase words, perhaps translate morning, afternoon etc into AM or PM. The hint feature is excellent for prompting for more detail. Although I found that the patience wears off quickly if the user has to go the loop again for more detail. Again, great work - I do realize that catching natural language is not by any means easy. Your thoughts are appreciated.
Hey Peter, all great suggestions, thank you! The case you are describing, the support for time spans (i.e. "weekdays", or "Tuesday through Wednesday", etc.) with a starting time is already in the backlog, i ran into a similar issue myself a little while back. The API is not quite smart enough with those yet, but it will handle those soon.
Of course you're right about the AMBIGUOUS_TIME_OF_DAY hint, I have few other tests that trigger that hint erroneously. I will be fixing it. It gets a little tricky as we enter more complex structure and a human understanding factor comes into play. But there is a lot that can still be done. I will move the issues you brought up to the top of priority list.
Cephalexin Sinus Propecia Xenadrine Where Can I Buy Furosemide In The Uk <a href=http://oc-35.com>lower price on finasteride camber</a> Fedex Amoxicilina In Internet Orlistat For Sale 60 Mg Forum Cialis Effetti Collaterali