5. Classifying Meanings & Documents - [E3] 3. Dodds, Peter Sheriden et al. 2015.

[JunsolKim]

Post questions here for this week's exemplary readings: 3. Dodds, Peter Sheriden et al. 2015. “Human language reveals a universal positivity bias.” Proceedings of the National Academy of Sciences 1112(8):2389–2394, doi: 10.1073/pnas.1411678112

[GabeNicholson]

Interesting report but I am still left confused as to what this shows exactly and how this jibes with other literature on this topic. For example, linguists have found that there are around twice as many words with a negative connotation compared to positive connotations. And we know from Behavioral Economics that losses hurt twice as much as gains (loss aversion). Taken together, I would have expected a negative bias since negativity tends to sell well in media and in literature where a storyline is never complete without a challenge or potential loss. Perhaps I misunderstood the reading, but is there an explanation that can reconcile my two points above and the findings in the paper?

[isaduan]

Interesting report but I am still left confused as to what this shows exactly and how this jibes with other literature on this topic. For example, linguists have found that there are around twice as many words with a negative connotation compared to positive connotations. And we know from Behavioral Economics that losses hurt twice as much as gains (loss aversion). Taken together, I would have expected a negative bias since negativity tends to sell well in media and in literature where a storyline is never complete without a challenge or potential loss. Perhaps I misunderstood the reading, but is there an explanation that can reconcile my two points above and the findings in the paper?

I think the paper wants to argue that human use of language exhibits a positive bias. So it wants to make a case about language usage, about humans display positive moods more often in communication. On the other hand, "twice as many words with a negative connotation compared to positive connotations" could mean that humans tend to have more variance of sadness, or higher magnitudes of sadness, while loss aversion means that humans dislike sadness than they dislike absence of happiness. None of the two points you raised is necessarily about how frequent humans display sadness vs. happiness!

My question is about the human coding of how happy a word is - if the coding is done by speakers from different cultures, how can we ensure the coding is consistent?

[pranathiiyer]

Interesting report but I am still left confused as to what this shows exactly and how this jibes with other literature on this topic. For example, linguists have found that there are around twice as many words with a negative connotation compared to positive connotations. And we know from Behavioral Economics that losses hurt twice as much as gains (loss aversion). Taken together, I would have expected a negative bias since negativity tends to sell well in media and in literature where a storyline is never complete without a challenge or potential loss. Perhaps I misunderstood the reading, but is there an explanation that can reconcile my two points above and the findings in the paper?

I think the paper wants to argue that human use of language exhibits a positive bias. So it wants to make a case about language usage, about humans display positive moods more often in communication. On the other hand, "twice as many words with a negative connotation compared to positive connotations" could mean that humans tend to have more variance of sadness, or higher magnitudes of sadness, while loss aversion means that humans dislike sadness than they dislike absence of happiness. None of the two points you raised is necessarily about how frequent humans display sadness vs. happiness!

My question is about the human coding of how happy a word is - if the coding is done by speakers from different cultures, how can we ensure the coding is consistent?

I second this! To add on, I guess I would also like to know more about how people think research like this could help understand more about the linguistic nuances of languages themselves, and what some of the theoretical and non-theoretical implications of research like this could be?

[melody1126]

Following the comments from my three classmates above, what is the external validity of this paper? The authors mentioned that they "paid native speakers to rate how they felt in response to individual words on a nine-point scale, with 1 corresponding to most negative or saddest, 5 to neutral, and 9 to most positive or happiest" (pg. 2390) I am a bit confused how this connects to the theoretical questions they are raising..

[Sirius2713]

Interesting report but I am still left confused as to what this shows exactly and how this jibes with other literature on this topic. For example, linguists have found that there are around twice as many words with a negative connotation compared to positive connotations. And we know from Behavioral Economics that losses hurt twice as much as gains (loss aversion). Taken together, I would have expected a negative bias since negativity tends to sell well in media and in literature where a storyline is never complete without a challenge or potential loss. Perhaps I misunderstood the reading, but is there an explanation that can reconcile my two points above and the findings in the paper?

I think the paper wants to argue that human use of language exhibits a positive bias. So it wants to make a case about language usage, about humans display positive moods more often in communication. On the other hand, "twice as many words with a negative connotation compared to positive connotations" could mean that humans tend to have more variance of sadness, or higher magnitudes of sadness, while loss aversion means that humans dislike sadness than they dislike absence of happiness. None of the two points you raised is necessarily about how frequent humans display sadness vs. happiness!

My question is about the human coding of how happy a word is - if the coding is done by speakers from different cultures, how can we ensure the coding is consistent?

Adding on this question, my question is as mentioned in the paper, translation may bring inaccuracy to the study, for example "lying" in English and "acostado" in Spanish. How can this paper ensure happiness scores obtained by translation will be serviceable for purposes where the effects of many different words are incorporated?

[Hongkai040]

The main finding of this paper is not surprising to me. But I'm kind of skeptical about the ranking and validity stuff. Those corpus are by no means random samples of texts people from different cultures ever generated. So the ranking of the positivity may not be that meaningful ? Second, how did they deal with polysemic word usages? "Thank you a lot, really" Here thank you could be really negative! Another thing is I think the positivity of language could be temporal. If we use corpus collected during World War II, will we get a different conclusion?

[mikepackard415]

I tend to bristle whenever authors claim to have found a 'human universal', even when using enormously big data like in this paper. Like Hongkai above, I wonder about temporal differences (human language has probably been around 200,000 years - are we really ready to claim universality based on this sample?), and I also wonder about indigenous languages and other non-dominant human cultures.

I do find the result pretty interesting. I guess I wonder whether we might make an evolutionary argument here - negative situations might require faster action, so communication is limited to fewer words for efficiency's sake. Meanwhile we're happy to spend more time chatting during positive situations, allowing a wider diversity of positive words to evolve. I am not a linguist or a strong evolutionary theorist so I'm sure this is far too simplistic a suggestion, but it is fun to toss ideas around.

[kelseywu99]

I hold similar skepticism as Hongkai's concern, which is how the positivity and negativity of a word are measured. For instance, fig.1 fig.2 measures the mean of word happiness in corpora spanning across 10 languages with Spanish language texts having the most happiness expressed in their text and Chinese finishing the last. But when I looked over the Chinese corpora I found the most neutral word that embodies vastly different meanings being classified into the negativity category, i.e. the same verb sharing the same character refers to a) grasp b)arrest. So my concern would be how each word in each corpus is classified into its own class.

[LuZhang0128]

I wonder if there's any bias in this paper. Based on my personal experience, people tend to share a better version of their life on social media and tend to digest negative emotions by themselves or with friends. I wonder if this will positively bias the result, meaning that the author is observing more than the true effect. I propose that they can compare languages by the time of the posts. One hypothesis is: language in the post at night is more negative than that during the day.

[hshi420]

In this kind of cross-linguistic study, should we weight different medias in different languages differently? For example, the use of social media or song lyrics can be very different from culture to culture, which are closely related to the languages. Should we take these differences into consideration when analyzing them?

[chuqingzhao]

The paper provides a cross-culture translation measure of word shift and its happiness score. The results of cross culture distribution of happiness resonate with culture difference (e.g. Chinese speakers are more conservative in using positive words). The major implication, from my perspective is that it integrates the culture dimension in to interlingual translation. Beside of computational method, I am left wondering how to explain the casual mechanism behind the positivity bias?

[YileC928]

I am wondering what are the broader implications of this paper - besides presenting the descriptive results of a large-scale dataset? Also, I am especially curious about how they create and evaluate the quasi-ranked list when aggregating multiple corpora for one language.

[chentian418]

In terms of studies of the positivity of isolated words and word stems, what are some possible explanation of the conflicting results? Does any attribute of isolated words/word stems contribute to the positivity bias/negative bias?

[Emily-fyeh]

I would really wonder if there is any word in the experiment that is neither "happy" nor "sad". I would also want to know if the antonym of happy is sad in every language in this paper.

[ttsujikawa]

It was very interesting. However, I wonder if people become more indirect and implicit when they share negative feelings and this tendency might bear bias in this research.