tcarr

I'd second @atheistpreacher's recommendation of the Farseer trilogy. Also, it's not exactly new, but I'm two thirds of the way through the Imperial Radch trilogy and it's been excellent so far.

@zombiepie:

So on the issue with NLG, Square Enix bought the license. They paid for some NGL-based software, and while it might not be "live" it seems they did use it during the development of the game. Because the license for the software for the game is included in the credits, that seems to suggest they ran something through an NGL-based AI and instead of keeping it online or active, they took the results and inputted them into the game.

They say that the game originally included NLG functionality and they're looking to add it back in a future build, which is consistent with them having licensed software. As you say, they also might have generated some text and added it to the game. However, given what's written on the Steam page, I don't see why you would go in expecting there to be NLG functionality in the game.

Also, I would push back that running a game's script through and AI is the same as machine translation. A neural machine translation (NMT) system, which is also credited in the game, means that words and phrases are being translated by a neural AI on the basis of the likelihood of a sequence of words and then translating entire sentences in a single integrated model instead of attempting to translate the game literally as-written word by word. Also, from the sounds of it, this game is equally frustrating in Japanese.

I realise that, in this context, the meaning of "script" is a bit ambiguous (i.e. it could mean a block of code that is executed when the game runs). To be clear, by script I meant only the text that is displayed on screen during the game. I didn't mean any of the code that governs the game logic or the text parser. In addition, by "machine translation" I just meant translation performed by computer, rather than a human. I didn't have a specific method in mind, and I would include any AI based method in that.

If Square Enix was trying to make some easy money, I can see why they would try localising a game by machine translating the script, leaving the rest of the code untouched and adding an emulation layer to get it to run on modern hardware. I agree with you that this would be bad, but I don't think they're doing a test run here. This is because they've also swapped out the text parser for the natural language understanding based method. If all they were doing was to try localising the game as cheaply as possible, I don't see why they would do this. The original Japanese text parser didn't require particles, just the verb and a noun, so they can just swap out the Japanese words for English equivalents (I guess this is what the fan translations do?). Alternatively, localise the Famicom version and avoid having to deal with text parsing altogether. The fact that the Japanese reviewers are also complaining about the game not understanding their inputs is entirely consistent with this. The Japanese script is presumably the same as the original game, the change is in how text inputs are interpreted, and I don't see what that change has to do with localisation.

Basically, I do think they were trying a new approach to interpreting inputs, rather than trying a new approach to cheaping out on localisation costs.

Finally, here's the issue I and others take with calling this an free experiment and then letting things rest. There's no alternative to this if you want to legally play Portopia in English or any other language. There's not a huge groundswell asking for a proper port of Portopia, but with the game being one of the most foundational works in the visual novel genre, there's a big fear that Square Enix is going to look at this and say "Well... our work here is done."

I completely agree that there should be a legal way to play this in English (and other non-Japanese languages). I guess where I differ is that I don't think this experiment makes a proper release less likely. I feel like it could go either way. Probably the best outcome is a bunch of people try this experiment on Steam, then write reviews saying the natural language understanding based input interpreter is terrible and they should just release a modern update of the Famicom version. I'll happily do so ;-)

I agree with you that this game deserves a rerelease similar to Famicom Detective Club, but I think you're being too harsh on something that was released for free as an experiment.

Square Enix essentially used Portopia, a game with a short and simple script, as an experiment to see if they could cut out traditional localization third-party sources by opting for AI language systems.

Here's my theory. I think Square Enix took the entire script for Portopia, which is not that much, and then ran that script through an AI language algorithm, proofread what the AI spat out once or twice, and then published the result to see if such an "easy" turnaround could suffice as a game release for an older visual novel.

This doesn't add up to me. If this was all to see if they could cut corners by using machine translation, why not start with the Famicom version and not have to worry about parsing text at all? Alternatively, machine translation wouldn't preclude them from keeping the original text parser. If the objective was seeing if they could get away with machine translation, changing the text parser seems completely unnecessary.

On the Steam page and in the previous press releases, Square Enix said that the AI they used should allow for natural conversations beyond the original script.

From the Steam Page

The release notes from April 24th say the same thing, so it seems like that functionality was taken out before release.

First, let's talk about the art assets in this game. The backgrounds come from, wait for it, the "Kobe Tourism Bureau!" That's right, the re-release uses stock photo backgrounds licensed from the Japanese equivalent of The Chamber of Commerce!

I'm not going to say this is great, but Square Enix is giving this away for free, so I imagine the budget was pretty limited.

I do agree with a lot of your points though. I think you hit the nail on the head when you brought up SCUMM. It seems like they've kept the underlying game systems unchanged, so the game logic defines the same set of actions you can do at each point as the original version. In this context, the best natural language understanding can do is broaden the range of text inputs that trigger a given action. To go to the port, you can type "Let's see what's happening at the port" instead of "go to port". However, ultimately the set of possible actions is what matters, and it's much less hassle to drop the text parser and just make that directly visible.

Also, the part about using data to improve the game is a bit weird. To make it work, they'd have to be collecting data on what action people wanted to trigger with a given text input as well as the text input itself, and it doesn't seem like they're doing that right now.

My charitable take is this. This is what they say it is: an experiment. They wanted to see if natural language processing could fix problems with text parsing. It definitely hasn't worked, but not all experiments do.

I played through Crowns and Pawns when it came out last year and basically had exactly the same experience as you, right down to randomly combining items to solve the sacramental wine puzzle! There were a couple of timing based puzzles that didn't feel great (one with a balloon and one with scooters), but I saw that there was a patch that rebalanced some puzzles so maybe those got tweaked after I played through. However, in general, I thought the puzzles made sense, and I was never stuck on anything for too long. I also really liked the art and the music. It was pretty short, but I enjoyed my time with it, and I hope they make a sequel. 4/5 feels spot on to me.