Why am I bad at Scrabble?

And, more interestingly, why are large language models?

If I had to put myself on the shape rotator vs. wordcel spectrum, I’d put myself pretty firmly as a wordcel. Not just shape rotation, but anything visual is difficult for me. In middle school, I got near perfect grades in every class except for art, where I got a C+. This shocked my older brothers, who were unaware that it was even possible to not get an A in that class¹. Even my parents, normally strict about grades, let me off the hook for that one because they found it more funny than disappointing.

My wordcellitude has continued my entire life. My career is mostly word-focused (lots of emails and documents with a few spreadsheets); my free time is spent on word-focused things (like this essay and the books I read); I spend my time with my friends doing word-focused things (often just chatting/bullshitting). It’s a virtuous cycle: I spend time on words, I’m good at words, so I get better at words. So: why am I not good at Scrabble?

I mean, it’s not like I’m exceptionally bad at Scrabble. I do okay. But you’d think I’d be great at Scrabble. I won vocabulary and spelling championships in my youth. Even now, I would say I have a voluminous vocabulary² and way above average spelling ability. It seems like my skills and knowledge should be huge boosts for my Scrabble ability.

This is the sole remnant of my brief reign as Connecticut state vocabulary champion. Before me, boundless and bare, the lone and level sands stretch far away.

But they’re not. For some reason, scramble letters up and ask me to form words in columns or rows and my mind goes blank. I find myself relying way too often on 1 point letters and a basic vocabulary. It takes a lot of effort for me to come up with polysyllabic words. It makes me feel like a failure of a wordcel.

As a result, I don’t play Scrabble often. This likely has something to do with why I’m not great at Scrabble. But, you know, I’d still expect some crossover between my wordcel skills elsewhere and Scrabble. I mean, my tennis skills have helped me in every racquet sport that I’ve ever tried. My breakdance skills have helped me in every dance I’ve ever tried to learn. Why wouldn’t a large vocabulary and good spelling help me in Scrabble?

My best answer for this is that Scrabble is not actually a game of wordplay, at least not in the conventional sense. It’s actually possible to be very good at Scrabble without really knowing how to speak at all, like the famous example of the guy who won both the French and Spanish Scrabble championships by simply studying the respective dictionaries for those languages. It’s impossible to imagine someone winning a more traditional wordplay-focused competition, like a crossword competition or a rap battle, in the same way.

One obvious reason for this is that Scrabble asks for words in a different way than you’d see them in any other context. Searching for the right word, spelling words correctly, and making meaningful sentences is a universal feature of languages. Trying to make words out of their component parts is not. In that sense, experience with the language that Scrabble is in does not contribute to one’s skill in Scrabble.

But, that doesn’t explain why the same innate characteristics that make me a natural wordcel don’t make me talented at Scrabble. I have always found reading and writing easy. My language skills have always soared beyond my math skills. Every humanities class came easily to me in school.

So, I want to propose a different reason why I’m not good at Scrabble: unlike most language games, it’s a bounded system. There are a limited number of words that you can make on your turn, bounded by both the letters available and the Scrabble dictionary itself. This is entirely separate from most language games.

As a result, the associative thinking that I normally use to great effect in language games, combining words and ideas to form clever juxtapositions, runs into problems. Instead, I have to rely on much more bounded, linear thinking: these are the letters in my hand, these are the letters on the board, these are the words in the Scrabble dictionary. Creativity is severely limited.

And, if I were to extend further, I think this sort of linear thinking is what I, and other wordcels, struggle with in general. We are great at associative thinking. We draw together thoughts and ideas that you wouldn’t expect would go together, and we’re able to associate them logically in interesting ways. These ways can have bearing on the real world, such as in my scientific endeavors. But we are bad at thinking step-by-step within a ruleset. Math is just one example of this. We are just as bad at board games, chess, many forms of engineering, or, of course, shape rotation.

In that sense, we are sort of like LLMs, who are also great at associative thinking, including some that has bearing on the real world, but struggle with straightforward, linear thinking, including math or Scrabble. The question would then be this: LLMs are supposedly bad at logical thinking because of the nature of how they are trained and structured³. However, the difference between me and a Fields Medal winner is not the structure of our brain, or the way we’re trained as children, when a future Fields Medal winner presumably starts showing the mathematical aptitude that I lacked. It’s something about relative sizes of unique parts of our brain, and perhaps the density of certain connections.

So, what does that mean for us wordcels, me and ChatGPT alike? How close are we, biologically/mechanically to being good at Scrabble? Let me know, because I have some long-held board game grudges to avenge.

I can’t put screenshots in footnotes, so I had to put it here. I’m just making a point that LLMs aren’t bad at Scrabble because of tokenization, because apparently they have no problem understanding misspelled words.

1. Some big part of my poor grade in this class was because I got very sick of drawing the same apple every day, and traced leaves instead. In my defense, the art teacher probably should have actually taught us something about how to draw, instead of just making us draw. Looking at you, Ms. Zweir.

2. Yes, this sesquipedalian description is intentional.

3. And specifically bad at games like Scrabble, because of their tokenization system. However, I’ve found LLMs are remarkably good at understanding misspelled words in a way that doesn’t fit with my understanding of how tokenization is supposed to work. So I don’t buy this explanation. See the screenshot at the end of the article.