7.1 Patterns » Test Your Knowledge: Explanations
Q1. In the spirit of offering many useful examples, we'll play "one of these does not belong"... Which of the following is is not an example of a pattern?
A. In a standard Shakespearean sonnet, the rhythm of the line, and the emphasis on the syllables, is short-long, as in "but SOFT what LIGHT through YON-der WIN-dow BREAKS".
B. When two men hug in a friendly and non-romantic fashion in the USA, they usually pat each other on the back. (If you are unfamiliar with this practice, you can learn more here.)
C. When Robespierre first raised the question of "enemies within the revolution" in a speech, the next person agreed with him.
D. When temperatures get hotter during summer, the murder rate tends to rise.
Correct Answer: (C) In this case, it's a one-off event; not a habit of reaction to Robespierre, but a particular instance. It could be the case that the example in (C) is an instance of a larger pattern (e.g., the pattern of "people agreeing with Robespierre"), but on its own, it is only an event.
Q2. Continuing with "one of these does not belong"... Which of the following is not an example of interpretable machine learning?
A. An algorithm that resolves a set of documents into three kinds: documents about golf, documents about gold, documents about grand pianos, presenting a list of words that are used for making the classifications.
B. An algorithm that finds three kinds of Pepe the Frog images, pixelated, black-and-white, and hand-drawn, and shows that hand-drawn Pepes are more commonly associated with racist content online compared to pixelated or black-and-white images.
C. An algorithm that predicts whether or not a court transcript is associated with a guilty verdict, along with an estimate of its confidence in the judgement.
D. An algorithm that tracks four different kinds of social power (or its absence) in how a person's speeches are imitated, or not, and associates it with political party.
Correct Answer: (C) The problem here is that while we know the algorithm's degree of confidence, we don't know what the source of the algorithm's confidence is -- we don't know "on what basis" it has that confidence. (A) is a common kind of interpretable machine learning; quite a simple one, and a toy version of something called "topic modeling". (B) is an interesting example, where the algorithm is interpretable, but the pattern association may be a bit epiphenomenal -- there's nothing essentially "more racist" about hand-drawn Pepe memes. (D) is an example of an interpretable machine learning algorithm -- here, the four kinds of social power map on to richer concepts from sociology, and we've learned about a new connection between political affiliation and more basic behaviors.
Q3. One final "one of these does not belong"... Which of the following is not an example of how to turn a "cold" (computer) problem into a "hot" (human) one?
A. "this pattern is more common in the second half of novels" > "this pattern has lots of negative emotion words, compared to the other patterns, and our result shows that there's twice as much negative emotion in the second half of novels than the first."
B. "this pattern is associated with the second half of the 20th Century" > "this pattern uses words that are much longer than words in other patterns, and uses words that are more rare in common English usage; only half of the people reading texts with these patterns would know all the words."
C. "this pattern is twice as common as the other patterns" > "this pattern would be easier to learn."
D. "this pattern is associated with newcomers to the online forum" > "this pattern is far more likely to contain hedging language like "perhaps" or "maybe"; newcomers are twice as likely to hedge their statements in a post than long-time participants."
Correct Answer: (C) With more work, one might make (C) "hot", but the current formulation, "learning" is ambiguous: what is this particular learning like? To make (C) "hot", or hotter, it would help, for example, to know what the content of the pattern is, or what it might mean for the person to learn that pattern versus another one.
Q4. BONUS: Which of the following is a pattern in the quiz questions above?
A. The first was about "what's a good pattern".
B. All of the questions in this section are in English.
C. The correct answer is always the third one.
D. They are really about Simon's lived experience.
Correct Answer: (C) (A) is an example of an "event". (B) could be a pattern, but it's not very meaningful because it is true of the entire course – it could be a meaningful pattern if this course was seen in a much larger, multi-lingual context. (D) is a reasonable guess, but Simon asserts that it is not the case.