Responses for July 29

A. I ran the sarcasm detector on 5 headlines from CNN and 5 headlines from Fox. Interestingly, the sarcasm detector results were exactly even between the two sources. Two CNN headlines (“Warren calls on agency chiefs to commit to not deploying federal forces on Americans should Trump not leave office” and “State of Congress’ stimulus negotiations: ‘We’re in different universes right now”) were falsely labeled as sarcastic and two Fox headlines (“Rep. Louie Gohmert tests positive for coronavirus before planning to fly with Trump to Texas” and “Trump says Harris would be ‘fine choice’ as Biden running mate”) were also falsely labeled as sarcastic. The specific results were, respectively, 0.9973514080047607, 0.6634448170661926, 0.9496141076087952 and 0.7110357880592346. The average false sarcasm value (the specific results as listed before, averaged by news source) was actually almost the same between the two news sources, with CNN having an average of 0.8303981125354767 and Fox having an average of 0.8303249478340149. This is especially surprising when you see that the headlines aren’t even about the same stories. This similarity is probably just a coincidence, but it’s certainly a very interesting one.

B. The generate text command makes text by taking a starting string (“ROMEO:” or, as in my second run, “SPONGEBOB:”), initializing a RNN state, and determining the number of characters to generate. The RNN state and starting string are used to generate a prediction of the next character. The new RNN state is then used as a new input and the entire process repeats.

C. I used the input string “cuál es tu nombre,” which is supposed to translate to “what is your name” but instead I got “what s your name.” I also tried “lo siento” and “dónde está el baño.” In return I got “I’m sorry” (correct) and “wherever you the best” (incorrect; correct version is “where is the bathroom”), respectively. These results weren’t spectacular. A very easy phrase “lo siento” worked well, but additional basic phrases really didn’t work well at all. “Cuál es tu nombre” only got one letter wrong, but “dónde está el baño,” as you can see, was entirely inaccurate.