UNPOPULAR OPINION. 99% of my feed is lashing out at this post, but hear me out. The ACTUAL paper is presented as an attempt to replicate known concepts with large-scale quantitative methods. Indeed, the paper addresses its limitations vs. historical linguistics research! 1/8
The article does not claim Groundbreaking Novelty Whoa This Has Never Been Done Before(TM). The tweet and the @physorg_com article do. @kennysmithed here explains it perfectly. 2/8
What went wrong? @physorg_com should not have claimed that this is the "first large-scale, data-driven study" in semantic alignment. It is the first APPLICATION of #MachineLearning to this field. Is it supremely cool? YES. Is it groundbreaking? NO. 3/8
Now, there are a few takeaways we can learn from this mess. 4/8
Takeaway no. 1: #scicomm MUST DO BETTER. It must disengage from the obsession with novelty. Progress is incremental, based on interdisciplinary cooperation, and rarely fueled by random eurekas. It's complex. Guess what? We, the readers, can take complex. We LOVE it. 5/8
Takeaway no. 2: Humanists must make the effort to read scientific papers instead of publicity. We preach the importance of primary sources, don't we? It is our duty, more than anyone else's, to make that extra step. And to direct criticism precisely where it is needed. 6/8
Takeaway no. 3: We all need to stop fostering toxic dynamics of us vs. them. Quantitative & qualitative can coexist. The future is interdisciplinary. We must all work together to make such projects happen and keep learning from & with each other. 7/8
@thai101 supported this statement with a wonderful parallel!
