Perspective is an API that makes it easier to host better conversations. The API uses machine learning models to score the perceived impact a comment might have on a conversation. Developers and publishers can use this score to give realtime feedback to commenters or help moderators do their job, or allow readers to more easily find relevant information, as illustrated in two experiments below. We’ll be releasing more machine learning models later in the year, but our first model identifies whether a comment could be perceived as “toxic" to a discussion.

This is a project out of Google's think tank. You can test out how the API scores things by using the text box near the bottom of the page; interesting to see how a single word can change 'perceived toxicity'


The problem seems to be in the strictness of the rules. This phrase by C.S. Lewis was judged as 7% 'toxic'

    I believe in Christianity as I believe that the sun has risen. Not only because I see it, but because by it I see everything else.

Change every '.' with an exclamation mark, and lo, it's suddenly 1% more 'toxic'. End any sentence with more than one exclamation mark and you will see how it correlates it with toxicity. Add something like '#fail' and you are getting into the upper 40% with your toxicity. Lack or misused punctuation also counts as toxic. Some British English is also slightly less toxic than American English :P.

However, the shortest and most 'toxic' phrases I managed to make are:

80% "Pass a fag, mate"

78% "Death is murder!"

68% "Rape is a plant"

I could go on.

"Interesting, but not very smart." is both a good description and something that's only 2% 'toxic' ;)

EDIT: Also, ending the sentence with 'bro' will increase 'toxicity' only slightly (in the following example it even lowers it!) . Ending it with 'brah' will boost the 'toxicity' by quite a lot and it does so completely disregarding the preceding part of the sentence.

28% "Don't tase me, bro"

28% "Don't tase me, broseph"

29% "Don't tase me, bro!"

30% "Don't tase me, broseph!"

32% "Don't tase me" <- Baseline without exclamation

34% "Don't tase me!" <- Baseline for exclamation

45% "Don't tase me, brosephina"

47% "Don't tase me, brah!"

48% "Don't tase me, brosephina!"

posted by forwardslash: 84 days ago