*** This is only half my comment. The site will not allow me to post the rest. ***
@Tankov PART ONE
Hi Tankov. Thanks for your excellent reply. I read both articles you linked to. I'll address your points:
"These LLM aren't being used by competent people who want to speed things up. They're being mostly used by people with no expertise to fill the gap. They don't know how to code around the game? They use AI. They don't know Japanese? They use AI."
This is an important one. AI is a tool unlike any we've invented previously in that it is a hammer that doesn't just help you build a house, but if held a certain way will not only build the house by itself but also design the house it's building. Its ease of use combined with its incredible power and ubiquity with regard to its wide availablility for little or no money makes it ripe for abuse by those who don't want to work but have the tool do all the work for them. It's the wet dream of tools from a corporate perspective. You don't even need to "use" it, it does everything for you.
As with delegating responsibility to another person, this comes with risks. "If you want it done right, do it yourself." The problem is that god is dead, and in his place corporations offer up convenience, the ultimate ambrosia, and the new god replacing all others. We don't live forever. We all want to be loved. And the tool that is AI is promised as a way to get hard things done faster, and often better, so that you can have both for the very low price of free to next-to-free. It's the most disruptive tech we've invented since fire, the wheel, guns, electricity, and the computer. AI doesn't just exist to make the things you already know how to do easier, it exists to extended what "you can do" to things so far beyond what you know and even your cognitive or temporal ability to know that it is essentially a life drug of the highest caliber. It's dangerous. But like with any tool, it's virtue lies within how people choose to use it. So while we can (and should!) cast blame upon its purveyors, ultimately the responsibility lies upon the end user. It is also an emerging technology with a lot of corporate fanfare behind it, and hasn't even come close to "settling down" yet / maturing, both as a tool and in how the public perceives it and its "responsible" use.
"Then you add the fact how context-dependent Japanese literature is and how LLM doesn't understand context, then everything become a mess. Putting a person at the end of such a process isn't QC. You can only polish a turd so much."
All language as far as I know is like this. Slang. Colloquialisms. Context dependency. Authorial intent based on previous works and overall vibe and life philosophy. Translation is difficult. I remember reading the liner notes of AnimEigo laserdiscs I bought explaining the philosophy of their careful translations with respect to the cultural differences between western and Japanese audiences. AnimEigo taught me how to pronounce "anime" (huge fail these days). They also taught me that translation isn't just a technical exercise, it's an art. If there's an "inside joke" that only Japanese would understand, do you translate it literally leaving western audiences baffled? Or do you find a western equivalent so they get the vibe thus preserving the emotional and intellectual intent of the authors? AnimEigo chose the latter. All of this could be addressed with proper AI wrangling, hypothetically.
I've been playing video games since at least 1982. Many human translated localizations in the 80's and 90's were absolute trash, with famous examples even becoming memes. A lot of the voice acting in the CD era was absolute trash. So my question is, why is AI apparently a bad translation tool, all things considered? It's not replacing anyone's job for a fan translation, and other efforts languished for years. I have plenty of AI rage, but this? Really? It seems the perfect tool for the job, and in the end, it is just a tool.
Comments 2
Re: Sega CD Shadowrun's English Translation Was Developed Using AI
*** This is only half my comment. The site will not allow me to post the rest. ***
@Tankov PART ONE
Hi Tankov. Thanks for your excellent reply. I read both articles you linked to. I'll address your points:
"These LLM aren't being used by competent people who want to speed things up. They're being mostly used by people with no expertise to fill the gap. They don't know how to code around the game? They use AI. They don't know Japanese? They use AI."
This is an important one. AI is a tool unlike any we've invented previously in that it is a hammer that doesn't just help you build a house, but if held a certain way will not only build the house by itself but also design the house it's building. Its ease of use combined with its incredible power and ubiquity with regard to its wide availablility for little or no money makes it ripe for abuse by those who don't want to work but have the tool do all the work for them. It's the wet dream of tools from a corporate perspective. You don't even need to "use" it, it does everything for you.
As with delegating responsibility to another person, this comes with risks. "If you want it done right, do it yourself." The problem is that god is dead, and in his place corporations offer up convenience, the ultimate ambrosia, and the new god replacing all others. We don't live forever. We all want to be loved. And the tool that is AI is promised as a way to get hard things done faster, and often better, so that you can have both for the very low price of free to next-to-free. It's the most disruptive tech we've invented since fire, the wheel, guns, electricity, and the computer. AI doesn't just exist to make the things you already know how to do easier, it exists to extended what "you can do" to things so far beyond what you know and even your cognitive or temporal ability to know that it is essentially a life drug of the highest caliber. It's dangerous. But like with any tool, it's virtue lies within how people choose to use it. So while we can (and should!) cast blame upon its purveyors, ultimately the responsibility lies upon the end user. It is also an emerging technology with a lot of corporate fanfare behind it, and hasn't even come close to "settling down" yet / maturing, both as a tool and in how the public perceives it and its "responsible" use.
"Then you add the fact how context-dependent Japanese literature is and how LLM doesn't understand context, then everything become a mess. Putting a person at the end of such a process isn't QC. You can only polish a turd so much."
All language as far as I know is like this. Slang. Colloquialisms. Context dependency. Authorial intent based on previous works and overall vibe and life philosophy. Translation is difficult. I remember reading the liner notes of AnimEigo laserdiscs I bought explaining the philosophy of their careful translations with respect to the cultural differences between western and Japanese audiences. AnimEigo taught me how to pronounce "anime" (huge fail these days). They also taught me that translation isn't just a technical exercise, it's an art. If there's an "inside joke" that only Japanese would understand, do you translate it literally leaving western audiences baffled? Or do you find a western equivalent so they get the vibe thus preserving the emotional and intellectual intent of the authors? AnimEigo chose the latter. All of this could be addressed with proper AI wrangling, hypothetically.
Re: Sega CD Shadowrun's English Translation Was Developed Using AI
I've been playing video games since at least 1982. Many human translated localizations in the 80's and 90's were absolute trash, with famous examples even becoming memes. A lot of the voice acting in the CD era was absolute trash. So my question is, why is AI apparently a bad translation tool, all things considered? It's not replacing anyone's job for a fan translation, and other efforts languished for years. I have plenty of AI rage, but this? Really? It seems the perfect tool for the job, and in the end, it is just a tool.