Friday, Square Enix announced a new, free port of its deeply influential, 40-year-old point-and-click adventure, The Portopia Serial Murder Case. This new version will have dialogue powered by AI-driven Natural Language Processing (NLP), where players can type or say whatever they want to NPCs.
I have a few reasons why i think this is not going to do what people want but I feel like getting into it will take too many words so instead im just gonna say fuckin sigh
I guess Iâm not entirely sure how on Earth you get that to work. Part of the reason so many player choice-centric games get criticized for being so narrow is that your decision tree needs to collapse to a singular (or one of a few singular) point so your main story beats donât get disrupted and so you arenât writing likeâŚinfinite dialogue branches.
Besides, the lovely people at BioWare are much better at flirting than I am - I would much rather leave my options in their hands lol
At best, this feels like itâll be a minor improvement on the near-dead genre of text adventure where instead of the game going âI donât understand SHUTâ if I type âSHUT DOORâ, the AI will be able to say âin this context, SHUT is a synonym of CLOSEâ and know I meant the accepted verb CLOSE for DOOR.
At worst, I can only imagine this is going to take the classic ambiguity issue of âGlass him.â from The Wolf Among Us and multiply it potentially infinitely as you attempt to tell the game that youâd like to FEED MOUSE and itâll look for the closest accepted action⌠perhaps interpret context as food-related and you will EAT MOUSE instead.
But I suppose as long as the responses were all written by a person, itâs not the worst use of AI weâll be seeing in video games in the upcoming years. I donât know if itâll improve the jank of the old VERB NOUN system, honestly, given where the tech is at, but with the right tweaks, it probably wonât be worse⌠though as with all uses of AI, I have to ask âso who wasnât paid to make this a cost-effective decision for Square Enix?â
Okay, so I played this. And itâs real bad. The system apparently works by comparing your inputted phrase with a permitted dictionary of inputs. I actually think this application of language parsing has a lot of potential now that I see what theyâre intending to do; I was worried that there was going to be an attempt to utilize transformers and try to be an LLM generated game, but it seems to exclusively be focused on parsing, which would make some text adventure design a lot more dynamic. Itâs not really accurate to say that Portopia is a visual novel; it was and is a graphical text adventure. (Also, unrelatedly, this game is moderately fucked; beating the shit out of suspects is part of the critical path.) The problem is though that the model theyâve trained is garbage. Seriously, I donât know what theyâre doing wrong here. The file was over 10gb so Iâm wondering if they installed a mini LLM on my computer just to do it.
To give you a sense of how bad the model is, I told the game, âgo to studyâ, which it would not accept. It only accepted it when I said âgo to the studyâ. No joke. The game is filled with absurd issues like this, where you might even know what the game wants you to do, but you have no idea how to convince the game to do it. Now that I know what theyâre trying to do, I think it has some potential, but this demo is so bad they probably shouldnât have even released it. I just wonder if the Japanese version is better.
this is so weird. i can see the tech being used for more natural interactions with VNs or text games but this implementation isnât what i was expecting. surprised bl*ckchain stuff didnât find its way here.
This sounds like such a strange implementation. I used to play tons of INFORM-based text adventures and, either through the planning of individual designers who added custom language to anticipate player actions, or through expansion of INFORM language itself in later versions, they got pretty good at interpreting what you wanted to do as long as you were roughly familiar with how those games worked. And then to strap all of this to something that was never a text adventure to begin with is even stranger.
Ditto. Is there anywhere to play that? At the very least it should avoid issues like needing âtheâ as such articles donât appear in Japanese, but then I wonder how it deals with particles. (Iâm not fluent in Japanese but can often muddle through games. One of my prouder moments was playing the Binary Domain demo in Japanese and getting the voice commands to work properly).
Not sure. Maybe thereâs some region or VPN stuff you can do?
Also, worth mentioning Portopia was actually originally a text adventure! I think versions on the NES or whatever opted for point-and-click design, though.
Oh, my bad! I read the article summary and saw âpoint and clickâ and then apparently skipped over the section in the article where it does actually talk about the '80s text parser version XD
EDIT: So I worked out you could get the Steam version running in Japanese, but I donât have a Japanese language keyboard - I usually use the Windows Japanese input which the game doesnât seem to support. It does have spoken language input, but I could only get it to run in low quality mode (the higher ones resulted in a VRAM warning, which is a surprise given my hardware).
Quickly gave up trying to get it working though â there appears to be no way to control the speed with which text auto-progresses (leaving me pondering whether Iâd read that fast in English, let alone Japanese) and then while I managed to get it to correctly parse what I was saying a couple of times I couldnât get a useful result from your NPC partner⌠and most of the time it interpreted what I was saying as complete garbage. Now, my Japanese may not be fluent, but itâs not that bad! In Binary Domain on PS3 - the last game I played with Japanese language voice input - it could understand me giving orders and saying âthanksâ or ânice shotâ to party members in Japanese and have them respond accordingly. Here? Not so much.
Ugh, I forgot they said that. I canât tell if this is bad news or good news but itâs probably bad. I hope they drop this idea like a rock, because not only do we have all the issues with how it relates to labor, but also because as weâve discussed, this weak ass demo does not bode well in the slightest. The fire for this demo is hot so maybe Square will pick up on it ⌠and just wait another decade before trying again.