You may have seen tweets, received texts, and noticed social media posts. One daily word. Five letters. Six guesses.
Wordle took the world by storm at the beginning of 2022. If you aren’t familiar, Wordle challenges people to guess a five-letter word in six guesses, with a new word refreshing every single day. There are 2,315 secret words with 12,972 possible words as a guess. The game lets you know how you’re doing after each guess by color coding the letters in the word you guessed. A green letter = the letter and position are correct, yellow = correct letter but incorrect position, and gray = letter is not in that word.
Once the game grew in popularity, friendly competition began. “I got today’s word in 5 guesses!” I told my partner excitedly. “Hah, I got it in 4,” he smugly replied.
As you can imagine, there is some pride in guessing the Wordle word in as few tries as possible. There are several strategies to employ, but often the first word you choose is one of the most important considerations. Some people have determined good start words through personal experience, but if you’re just starting out there are some common tips: Use lots of vowels, make different guesses, and try to guess more commonly used words. In addition, mathematical formulas have suggested several best start words, with top words including “LATER,” “RAISE,” and “AROSE.”
One recent take: What can machine learning tell us about an optimal start word? Recently, some scientists in Wisconsin investigated this.
Generally, machine learning lets software applications use some sort of data to predict outcomes without being explicitly programmed to do so. There are many different kinds of machine learning, and they vary based on what kinds of outcomes the scientist wants to predict. The Wisconsin scientists used reinforcement learning, where the model continues to learn when certain desired behaviors (like, a green tile for instance) occur and also learns from undesired ones (like, a gray tile perhaps). It’s sort of like trial and error. It gets a little complicated, but here is a basic rundown of how authors Anderson and Meyer did it with Wordle: First, they chose the data. They downloaded the 12,972 words from the Wordle website, selecting 2,315 as possible secret words. Second, a software program guessed multiple words over and over again and the algorithm gained information from each guess. The computer can choose any word in the list, eventually using ‘rewards’ (like, 5 points for a green tile) and ‘punishments’ (like –15 points if they lose) to figure out which guesses are ‘smart’ ones to make. The cool part is this sort of reflects how we as humans pick strategies as well. We do something, notice the outcome, then update our information and try again.
So, how does this article suggest we win at Wordle? Let’s start with one approach: choosing certain letters in the right spot. In their results, they report the letter S is the most common starting letter while E is the most common ending letter. This means one beneficial strategy is to choose a word that begins with S and ends in E, like “SAREE” or “SEINE”.
What about the optimal word choice? First, guess a word that is most likely to hit a green letter: BOWNE, SLATY, PRICK, FAUGH, or MEVED. If you get a yellow letter, guess words that will exclude more letters. If you get a green letter, limit your words to include those letters. Interestingly, it seems excluding letters should be a priority.
So, next time you’re opening up the Wordle game and wanting to find at least 1 green letter, give BOWNE, SLATY, PRICK, FAUGH, or MEVED a try. While there may be many other advantageous options, this may be a good place to start.
How do YOU play Wordle? Let us know in the comments.
References:
Anderson, B. J., & Meyer, J. G. (2022). Finding the optimal human strategy for Wordle using maximum correct letter probabilities and reinforcement learning. arXiv preprint arXiv:2202.00557.
Edited by Chloe Holden and Evan Leake
Leave a Reply