Why does the URLGetRelations API mislabel some sentences as "future" tense?

77 Views Asked by At

I'm hoping to use the URLGetRelations API to help identify sentences in text that are future tense. However, I’ve seen inaccuracies in sentences the API is identifying as future tense. Examples below, all identified as "future" but these are arguably incorrect. I see in the API responses below that there are some garbled words (e.g. "revivehim" from the first article), maybe this is causing the mislabeling? If you look at the URLs I've directed the APIs to, though, the distortions are not present in the original source text.

One from: http://www.reuters.com/article/new-york-police-idUSL2N15R02C

{ "sentence": " \"Oh my God,someone's hit,\" a tearful Liang recalled saying upon finding a bleedingGurley lying on a landing, as his girlfriend frantically tried to revivehim.", "subject": { "text": "his girlfriend"}, "action": { "text": "tried to revive","lemmatized": "try to revive", "verb": {"text": "revive", "tense": "future" } },"object": { "text": "him","sentimentFromSubject": { "type": "negative","score": "-0.70197" } } },

Two from: http://www.cnn.com/2016/02/11/us/nypd-officer-trial/

{ "sentence": " On Thursdayevening, about an hour before the verdict, the jury asked Justice Danny Chun toread them the charges and legal definitions, the second time this week.","subject": { "text": "Justice Danny Chun" },"action": { "text": "to read","lemmatized": "to read", "verb": {"text": "read", "tense": "future" } },"object": { "text": "the charges and legaldefinitions", "sentiment": { "type":"negative", "score": "-0.597878" } } },

1

There are 1 best solutions below

1
On BEST ANSWER

I think it's the word "to" that is causing this confusion. It's very common to see future tense verb phrases that include the word "to", such as "I am going to eat that later" and "We are planning to fly tonight." You also see phrasing like "Joe to appear on TV tonight" that implies future tense even if it isn't 100% grammatically correct. In the cases you shared, the word "to" is acting to form an infinitive, but because of the very similar structure to future tense verb phrases they are being classified as future tense. It's especially easy to see how this is happening with the second example: the subject is "Justice Danny Chun", the action is "to read", and the object is "the charges...". The system is seeing this as one sentence that reads as "Justice Danny Chun to read the charges", which gives us future tense.