3. Discovering Higher-Level Patterns - [E3] 3. Barron, Alexander TJ, Jenny Huang, Rebecca L. Spang, and Simon DeDeo. 2018.

[JunsolKim]

Post questions here for this week's exemplary readings: 3. Barron, Alexander TJ, Jenny Huang, Rebecca L. Spang, and Simon DeDeo. 2018. “Individuals, institutions, and innovation in the debates of the French Revolution.” Proceedings of the National Academy of Sciences 115(18): 4607-4612. 

[pranathiiyer]

Very interesting paper!

1. I was not completely clear on the idea of resonance being novelty minus transience. As I understand, novelty of a speech is the KLD compared to the previous speech and transience is just novelty under time reversal. But I don't think I completely understand how their difference could mean resonance.
2. How might looking at these constructs change in the case of a dynamic topic model? Thanks!

[Hongkai040]

This is a very interest research! It provides a quantified way to trace the flow of ideas. Just have a small question about shifts in committee function. The authors said they used change-point detection method. Is it necessary to do this? The authors told us what they did, but didn't tell us why. I'm quite skeptical about why they only chose scale=27 to explain committee function...

[Hongkai040]

Very interesting paper!

1. I was not completely clear on the idea of resonance being novelty minus transience. As I understand, novelty of a speech is the KLD compared to the previous speech and transience is just novelty under time reversal. But I don't think I completely understand how their difference could mean resonance.
2. How might looking at these constructs change in the case of a topic model? Thanks!

Hi Pranathi, my understanding is: low transience means the speech can easily fade, and high novelty means impressive. High resonance means a speech is both impressive and memorable. So, it's logically reasonable to construct such an equation. As for the second question, I think the authors said they used LDA model to identify those topics!

[pranathiiyer]

Very interesting paper!

1. I was not completely clear on the idea of resonance being novelty minus transience. As I understand, novelty of a speech is the KLD compared to the previous speech and transience is just novelty under time reversal. But I don't think I completely understand how their difference could mean resonance.
2. How might looking at these constructs change in the case of a topic model? Thanks!

Hi Pranathi, my understanding is: low transience means the speech can easily fade, and high novelty means impressive. High resonance means a speech is both impressive and memorable. So, it's logically reasonable to construct such an equation. As for the second question, I think the authors said they used LDA model to identify those topics!

Thanks hongkai! I actually meant dynamic topic model, forgot to type it properly. And thanks, I think that makes sense!

[GabeNicholson]

I wonder if the reason that the committee's opinions were less argued against compared to individuals is that they were well thought out and that more people accepted them once proposed? So their power would then come from superior planning and information dissemination.

[Sirius2713]

This is a very interest research! It provides a quantified way to trace the flow of ideas. Just have a small question about shifts in committee function. The authors said they used change-point detection method. Is it necessary to do this? The authors told us what they did, but didn't tell us why. I'm quite skeptical about why they only chose scale=27 to explain committee function...

Hey Hongkai! I think change-point model here is necessary. Because the authors want to investigate how the role of the committees evolved over time and the shift of their functions. With this model, we can find out when the there was a significant shift of the novelty-transience relationship suggesting the role of the committees changed.

And my question is how we can extend the methods mentioned in this paper to modern settings. For example, can we examine the congress speeches with this method?

[ValAlvernUChic]

The paper was really cool but I couldn't help but think about the paper in its context - legislation and political change. The paper cites heteroglossia tracking changes in speech patterns within a social body allows us to examine cultural evolution: the circulation, selection, and differential propagation of speech patterns in the group as a whole While ideas and information management amongst those seated in the chamber is important, I think it'd be cool to see how the ideas and innovations talked about in the parliament debates translated to the same type of change in the public sphere (though data for that might be rough to collect - thinking newspapers/forum letters as proxies).

[mikepackard415]

Super cool paper, in my opinion. I've been a fan of Simon DeDeo since I watched his ComplexityExplorer tutorial on renormalization.

I think the concepts of novelty and resonance work in this context because for the most part, all the speakers are hearing everything that gets said. I wonder whether they could be adopted to a corpus where that isn't necessarily the case? In other words, can we apply these kinds of methods to contexts where the current author/speaker might not be aware of everything said in the last couple days, but they have a general sense of what has been talked about in the past?

[melody1126]

One foundational idea is that speeches with high novelty (very different from speech patterns before this speech) and low transience (similar to speech patterns after this speech) are influential because they changed the course of the conversation. Why does this combination of high novelty and low transience tell us about speaker's influence in the assembly – would it be possible that the conversation was ripe for a turn at that moment, and the speaker happened to speak upon it?

[MengChenC]

This is really an interesting finding, the applications on different countries may be able to reveal some underlying patterns or different outcomes. Besides that, I am quite confused about how the author combined topic modelling and KL divergence? And what would the metric be like after this change?

[isaduan]

Very cool paper! My question is how accurate new word-use patterns really maps to new semantic patterns? It seems to me that coming up with new words, or new combinations of word, does not necessarily mean innovating! Curious to hear whether there are good validation strategies on this (I quickly skimmed the Supplementary Information and have not found any, but I could miss something!)

[Jasmine97Huang]

It is very interesting the authors are able to use KL Divergence to represent changes in word distribution from past to present and present to future, as well as how they brilliantly conceptualize novelty and transience,. My concern, however, is whether the dataset is balanced across the time frame of interest.

[YileC928]

Liked the paper very much! It is really a clever way to track shifts of language pattern by ‘novelty’ and ‘transience’. My question is in computational content analysis, how to tell when descriptive analysis (presenting associations like in this paper) is enough to produce strong results and when we need robust inference.

[hshi420]

The authors conceptualized and constructed their own measures of some concepts. I was wondering do these measures have to be based on some previous studies? If they can be constructed from scratch, what are the methods for validating the measures to make sure they have good accuracy and precision?

[zixu12]

This is a super interesting work! Nowadays, many patterns can be detected from historical texts. I especially like the resonance and novelty part, which I think can be applied to broader texts and topics. I am a bit curious that is this "resonance and novelty" already a common way to analyze in nlp? what are the other common approaches?

[Qiuyu-Li]

This is a very interesting paper and inspiring in many aspects. My question is about entending the topic: Would the extent to which the speech of officials towards novelty versus transitivity be an effective measure of the development/stability of the political institutions of the country?

[LuZhang0128]

As I'm also doing research in cultural evolution, I believe this article gives me a good insight into what I can do in my own research. As there may be a clear boundary in the article, if I'm generally studying the novel idea or emerging culture, it would be challenging for me to determine the boundary of study. Also, for this kind of research based on historical events, I often wonder if there's any way for us to test the validity of the result?

[chentian418]

This paper give a great a great illustration for me to extract measurements to explain a social science question! However, I am still confused about how are Novelty, transience, and resonance constructed from text. I understand that the author combines the KLDivergence with topic mixtures, but I am not sure how does this method capture their true social science meanings?

[chuqingzhao]

I really enjoy reading this paper! I am much interested in the novelty measurement in languages. The paper mentions the KLD which captures the contextual novelty at present and transience in the future. I wonder whether the novelty can also be related to the languages in the past? For example, a speaker's novel comment might be "disruptive", that is a brand new point and nobody talks about similar points in the past. Or the comment might be "developing", that is the speaker's comment could be surprising at present but developed from a previous opinion.