larlet-fr-david-cache/cache/2024/1cd6127ccec88387f4804f0c3cf1011a/index.md

title: Why wordfreq will not be updated url: https://github.com/rspeer/wordfreq/blob/master/SUNSET.md#i-dont-want-to-be-part-of-this-scene-anymore hash_url: 1cd6127cce archive_date: 2024-09-30 og_image: https://opengraph.githubassets.com/23640beea69cc6e481c29b906530677a890056e4755f495489cf39654858b963/rspeer/wordfreq description: Access a database of word frequencies, in various natural languages. - rspeer/wordfreq favicon: https://github.githubassets.com/favicons/favicon.png language: en_US

This documentation page has gotten a lot of attention recently! I think most of the people who find it understand where I'm coming from. I'd like to highlight a couple of things, now that people are linking to this page from all sorts of contexts.

• I still work on open-source libraries. Here's ftfy, the popular multi-purpose Unicode fixer.

• You could see this freezing of wordfreq data as a good thing. Many people have found wordfreq useful, and the latest version isn't going away. The conclusion that I'm documenting here is that updating it would make it worse, so instead, I'm not updating it. It'll become outdated over time, but it won't get actively worse. That's a pretty okay fate for something on the Internet!

Why wordfreq will not be updated

The wordfreq data is a snapshot of language that could be found in various online sources up through 2021. There are several reasons why it will not be updated anymore.

Generative AI has polluted the data

I don't think anyone has reliable information about post-2021 language usage by humans.

The open Web (via OSCAR) was one of wordfreq's data sources. Now the Web at large is full of slop generated by large language models, written by no one to communicate nothing. Including this slop in the data skews the word frequencies.

Sure, there was spam in the wordfreq data sources, but it was manageable and often identifiable. Large language models generate text that masquerades as real language with intention behind it, even though there is none, and their output crops up everywhere.

As one example, Philip Shapira reports that ChatGPT (OpenAI's popular brand of generative language model circa 2024) is obsessed with the word "delve" in a way that people never have been, and caused its overall frequency to increase by an order of magnitude.

Information that used to be free became expensive

Before I wrote this page, I'd been looking at how I would run the tool that updates wordfreq's data sources.

wordfreq is not just concerned with formal printed words. It collected more conversational language usage from two sources in particular: Twitter and Reddit.

The Twitter data was always built on sand. Even when Twitter allowed free access to a portion of their "firehose", the terms of use did not allow me to distribute that data outside of the company where I collected it (Luminoso). wordfreq has the frequencies that were built with that data as input, but the collected data didn't belong to me and I don't have it anymore.

Now Twitter is gone anyway, its public APIs have shut down, and the site has been replaced with an oligarch's plaything, a spam-infested right-wing cesspool called X. Even if X made its raw data feed available (which it doesn't), there would be no valuable information to be found there.

Reddit also stopped providing public data archives, and now they sell their archives at a price that only OpenAI will pay.

I don't want to be part of this scene anymore

wordfreq used to be at the intersection of my interests. I was doing corpus linguistics in a way that could also benefit natural language processing tools.

The field I know as "natural language processing" is hard to find these days. It's all being devoured by generative AI. Other techniques still exist but generative AI sucks up all the air in the room and gets all the money. It's rare to see NLP research that doesn't have a dependency on closed data controlled by OpenAI and Google, two companies that I already despise.

wordfreq was built by collecting a whole lot of text in a lot of languages. That used to be a pretty reasonable thing to do, and not the kind of thing someone would be likely to object to. Now, the text-slurping tools are mostly used for training generative AI, and people are quite rightly on the defensive. If someone is collecting all the text from your books, articles, Web site, or public posts, it's very likely because they are creating a plagiarism machine that will claim your words as its own.

So I don't want to work on anything that could be confused with generative AI, or that could benefit generative AI.

OpenAI and Google can collect their own damn data, and I hope they have to pay a very high price for it. They made this mess themselves.

— Robyn Speer