Gender stereotypes
molliem opened this issue · comments
Letters for women are more likely to include gender stereotypes (she is compassionate vs he is a leader) and emotion-focused words.
Goal: Develop code that can read text for words and gender stereotypes and highlight them (Word List). Return a summary statement that directs the author to review the highlighted words and evaluate whether they are relevant for the recommendation or evaluation. If the emotion-focused words are relevant to the letter or evaluation, suggest that the author include additional statements that balance them out and highlight other relevant areas like skills and accomplishments.
Note on the word list: some of the words are incomplete, for example "shar" is included so that it captures sharing or shared or share.
Hey @molliem I loved your project and I'd like to help!
I don't know if it's a problem, but I'm a begginer so I might need a little more guidance on understanding what it is to be done.
First, I read what you said on #3 so, it's okay to use Python, right?
What I'd have to do in this issue is receive a letter and run an analyses over it using the list you'll post (I can help with that too, if you need it) and then return the letter with the gender stereotypes highlighted. I also should return a summary, where should I return it? On the letter itself?
Thank you!
Here's an example that might help: Gender Bias Calculator
Ideally, I'd love to highlight the words in the text box to make them easier to find.
Found a word list and added it here.
I have an idea of how to analyze the letters using python and a list of words. But I have no idea on how to put that on a website. I can write the code that will analyze the data and return the answer on the command line and then you can evaluate to see if it's good or not and then we can think of how to put that on a website.
I believe that in order to make a summary... We should decide what structures a good recommendation letter should have and then create conditions on the code that will test if the letter attend or not to these structures. If they don't we print messages to the user explaining how they can improve their letters. What do you think? Is that a good approach?
I was able to figure out the website communication part, so that is set!
I agree with your thinking about the approach. I tried to identify structures of a good recommendation in the readme and then base the issues around it with information about what to return to the users under the Goals of each issue. Does that match what you're thinking? I want to be certain that we are on the same page.
Hi @marimeireles! You can poke me too if you have any questions about how to get started :D
@j6k4m8 Thanks!
@molliem I see that there is a bullet "Summarizes changes that would reduce bias for the writer" on readme, but I don't know how to do that.
The picture that you sent me from text.io shows a detailed summary with things like: needs more "we" statments, contains too many questions, etc.. And I didn't find anything on readme that guides me through that.
Also the link to the word list isn't working anymore.
Sorry for asking so many questions.
@marimeireles I divided the word list into separate male and female word lists without headers to make it easier to scan. You can find them on the main page.
The summary will pull together the results from all of the different signals. For example, gender stereotypes is one signal and nouns is another signal. For this issue, the summary just needs to return information about the presence of gender stereotypes and what to do. I thought we could direct the author to review the highlighted words and evaluate whether they are relevant for the recommendation or evaluation. If the female-focused words are relevant to the letter or evaluation, then we would suggest that the author include additional statements that balance them out and highlight other relevant areas like skills and accomplishments. These could come from the male word list, although I haven't pulled out specific words yet. Each issue has a description about what to include in the summary statement as part of the goal.
We figured out a way to return the summary statements to the website, so you don't need to worry about that part.
Feel free to ask questions!
@marimeireles this tutorial might be helpful! (although perhaps a bit out of date based upon @neiljp's last PR, #34)
Let me know if I can clarify anything :)
@marimeireles I'm going to start going through the articles on the README to identify words and phrases in the different categories this weekend. If you want to divide up articles and help pull words and phrases into documents for the detectors, let me know!
@marimeireles I divided the word list into separate male and female word lists without headers to make it easier to scan. You can find them on the main page.
Ok, thanks!
I'm going to start going through the articles on the README to identify words and phrases in the different categories this weekend. If you want to divide up articles and help pull words and phrases into documents for the detectors, let me know!
I'd rather focus on writing this code now because I've already took too long :p
@j6k4m8 Those are great docs! I'm going to try to use them as a guide.
@marimeireles Sounds good!