Presentation is loading. Please wait.

Presentation is loading. Please wait.

Computational Models of Discourse Analysis Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute.

Similar presentations


Presentation on theme: "Computational Models of Discourse Analysis Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute."— Presentation transcript:

1 Computational Models of Discourse Analysis Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

2

3 Warm-Up Discussion What is social causality and what aspect of discourse analysis does it most naturally connect to? What is the connection between Girju’s approach and Engagement?  How would you characterize Girju’s approach?  What was she trying to do and why?  What can this approach be used to learn? Student Comment: I'm not really sure how to tie any of this into engagement, other than the fact that the paper confirms that affective values aren't very useful in determining the sentiment of sentences. They claimed that he is a stellar PhD student although they correctly acknowledged not having any proof. Student idea: What I think might be interesting to do with Engagement, is maybe to somehow apply this good-bad distinction to it. I'm not sure it would be useful, but it could be interesting to see if contracting really is "worse" than expanding (wrt affective value, polarity). I guess I don't understand how social causality is different, in the way they're presenting it, from ordinary causality in sentences. Does it just require that both targeted objects in each event are people? They make the claim that studying causality is novel in computational linguistics. This is just wrong on the face of it.

4 Student Objection I'm also confused about how the author thinks this analysis has predictive power for people's actions. I guess if we assume that events, as reported by others, are not biased and are not overly reported (likely due to social influences), then we could use something like this to predict others' reactions?...I'm not sure. My something- fishy-is-happening alarm is going off. Or am I missing some important part of the analysis? What is the implication that event mentioned generalize over tens and modality? What would it do with something like “Those Pakis [they] should be boiled in hot oil for what they did.” from the Indian blog. What is the function of reported events in a blog?

5 Student Comment I think Iris is right that the problem here is that it doesn't actually do much to identify facts about gender. Rather, it identifies ascribed behavior in reported events, and not even particularly generic reported events - only those events which would be reported in a very strictly structured way that only includes pronouns (and proper nouns?

6 Student Question I'll buy that their way of identifying relations (using pronoun templates) works. It seems like a very fast algorithm with high precision (they suggest 97%, though probably with extraordinarily low recall). Also, what does it mean to "represent" 56% of the data?

7 Student Question On page 68, I'm not quite sure how they got from pronouns to clauses and how they merged parts 2 and 4 in with the rest (if they did at all; this seems slightly different from their approach on page 69, but I might just be wildly confused). Clever!! P 69 is the next step – once you have the pair of clauses, you decide whether they are reciprocal or not. Any ideas on how this could be extended to the case where you use arbitrary named entities or arbitrary noun phrases rather than just pronouns?

8 Reciprocity Reciprocal context: if a temporal order is indicated within C1 or C2 using discourse markers like “still”, “then”, etc., then it can’t be symmetric Type of eventuality [state versus event]: eventive version might sound symmetric, but it are not – “They chased each other” implies a sequencing Modality: would, should could – markers of events Temporal order of eventualities: Part 3 might indicate an ordering

9 Student Question I'm also slightly confused about their symmetry results on the top of page 71; it seems that 84-37 = 47, not 78. Reduction in error rate: 100 - 78.33 = 21.67 100 – 84.33 = 15.67 21.67-15.67 = 6 6 / 21.67 = 27.7 But: 6/15.67 is around 37, so maybe the divided by the wrong thing? I'm not all that enamored of their machine learning for symmetry. 84% accuracy with 78/22 split is probably around a 0.2-0.3 kappa, which is... okay? I guess? It's awkward that the features they describe were used, especially F1, since they're only using examples from the ambiguous patterns (I think). Similarly, 51% on a 3- way classification task doesn't look that great. I'm not sure what the set of classes Z is supposed to represent in their HMM. Then on the intentionality section they don't even attempt to automate it - presumably because they tried, and performance was even worse than the other two classifiers. I don't at all like that the different dimensions are being treated with such enormous differences between them.

10 Student Question I also don't understand how a HMM is useful for their purposes and also how the affective value works more than what I implemented for assignment 3. Eric: explain how this connects with what you did for assignment 3. Intuition: based on what you know from the sentiment lexicon, you can notice tendencies in ordering of positive and negative sentiment in eventualities, and you can use those regularities to learn sentiment associations for words where the sentiment is not known. Note that it was not very accurate

11 Student question The author's final paragraph failed to convince me that social causality has much use if any at all because there is no way to normalize the affective aspects of causality. I could see potential use in suicide prevention--trying to understand blogs of people who have attempted or are at risk of attempting suicide before the blog's respective author is thrown overboard.

12 Questions?


Download ppt "Computational Models of Discourse Analysis Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute."

Similar presentations


Ads by Google