![[personal profile]](https://www.dreamwidth.org/img/silk/identity/user.png){kind=small}
guslAt 3:30pm, I went to the first hour of Kearns's "crazy" talk about coordination games that people play on networks. Wean5409 was overflowing and "everybody" was there (why don't they reserve Wean7500 for this sort of thing?): I spoke to Alon and Cosma. He mentioned 4 different games on different kinds of random graphs, under different conditions.
In one condition (complex, rich-get-richer graphs), showing people the graph structure decreased the collective utility. He attributed this to HCI issues (e.g. this knowledge only confused people). But unless there was a condition in which only very few players knew the structure, he can't rule out the possibility that this is like a prisoner's dilemma. The real question is: do individuals profit from knowing the network structure?
Anyway, I missed the end of the talk. Was there a punchline? Any interesting consequences for economics, or computer science?
---
Then I ran off to a seminar talk on rhetorical design strategies, by Ishizaki and Kaufer, from CMU's English department. Their project is about helping writers using text statistics.
Their system gives writers feedback in terms of statistics on language action types (LATs), e.g.:
* commonplace authority: "it is widely believed that..."
* time shift
* positive
* negative
* motion
* etc
This is based on a dictionary of n-grams, annotated by Kaufer himself.
Since writing teachers lack a set of concepts with which to criticize student writing, one of their goals is to develop a meta-language to make rhetorical design strategies explicit.
Their system, Docuscope, has also been used to analyze Shakespeare. (This paper has example texts that are highlighted by Docuscope).
In the after-talk chatter, I learned that RST is used for grading the GRE Writing section.
Now that the English department has its hands dirty with Machine Learning, nothing at CMU can surprise me anymore.
In one condition (complex, rich-get-richer graphs), showing people the graph structure decreased the collective utility. He attributed this to HCI issues (e.g. this knowledge only confused people). But unless there was a condition in which only very few players knew the structure, he can't rule out the possibility that this is like a prisoner's dilemma. The real question is: do individuals profit from knowing the network structure?
Anyway, I missed the end of the talk. Was there a punchline? Any interesting consequences for economics, or computer science?
---
Then I ran off to a seminar talk on rhetorical design strategies, by Ishizaki and Kaufer, from CMU's English department. Their project is about helping writers using text statistics.
Their system gives writers feedback in terms of statistics on language action types (LATs), e.g.:
* commonplace authority: "it is widely believed that..."
* time shift
* positive
* negative
* motion
* etc
This is based on a dictionary of n-grams, annotated by Kaufer himself.
Since writing teachers lack a set of concepts with which to criticize student writing, one of their goals is to develop a meta-language to make rhetorical design strategies explicit.
Their system, Docuscope, has also been used to analyze Shakespeare. (This paper has example texts that are highlighted by Docuscope).
In the after-talk chatter, I learned that RST is used for grading the GRE Writing section.
Now that the English department has its hands dirty with Machine Learning, nothing at CMU can surprise me anymore.
![[identity profile]](https://www.dreamwidth.org/img/silk/identity/openid.png){kind=small}
(no subject)
Date: 2007-10-17 12:00 am (UTC)