January 2011 - GameSpite Interview with Richard Honeywood
Original source is this webpage.
Q: FFVII was huge in 1997. What was the localization team like at Square when you joined, and how was it different by the time you started work on Xenogears? More people? New methods?
A: (I kind of answered this above.) I joined just as FFVII was about to be released in America. Apart from Michael Baskett in the US Squaresoft office, it was just Aiko Ito and myself in Tokyo HQ. (Aiko had worked as a general assistant on the old Chrono Trigger and Super Mario games and, because she spoke English, was assigned to the foreign versions as well.
The original idea back then as that we took all the code off of the dev teams’ hands and did everything for them so they could move on to their next title.
On my first day I was given the source code to the original PS Chocobo’s Dungeon and told to both program and translate it all as a one-man team. Within a few hours I realized that the source code was incomplete and was not in any state to be compiled, let alone translated. But the dev team had been dissolved and no one had backed up the source code that existed on their individual machines. So we gave up on Chocobo’s Dungeon and learned our first lesson: to make the dev teams back up their source code so we could even begin to work on their titles! (In the end, after several similar cases, we gave up on the idea of taking over and doing all the work for the dev teams. Instead we asked their programmers to incorporate the translation back into the text and make any programming for us while their planners were in pre-production of their next title.)
The next major title was Xenogears, which I was assigned as Localization Producer on. We hired some new translators in our US office. However within a few months, Michael had decided to leave Squaresoft and the two new translators were left lost not knowing what was happening on the translation. Having a lot of religious content (some of which was considered controversial), as well as a lot of scientific jargon and psychological concepts, it was above the level of the then newbie translators and they soon asked to be taken off the title. I had to ask the dev team for an extension of several months as I took over the remaining translation myself. They couldn’t spare the staff to help with the title at the time, as they were moving onto Chrono Cross, so they taught me to create QA builds and master up the game myself. I slept at the office as I tried to get the translation out within marketing’s timeframe…but the quality suffered. This was the catalyst to hire editors to review translator’s writing, and to make sure translators cross-check each other’s work. This led to us to establish best practices such as planning properly for the worst case, allowing for familiarization and glossary creation periods, and so on, that later become standard. We also began to create our own tools to help with the translation and file conversion and slowly increased our headcount to match the growing needs of our titles.
Q: What were the biggest challenges (either technological or within the company) you faced as you started working in localization at Square? Did those challenges lessen over time?
A: When I first joined the dev teams were not even thinking of the foreign versions and were just focusing on the Japanese market. They didn’t want to put any effort into the foreign versions and their programming practices made localization a headache.
I mentioned the whole “type all text in double-byte (zenkaku) characters” system for FF games (up till FFIX), but there were worse programming follies going on. For instance, Tobal 2 couldn’t be reprogrammed to allow English text into its memory…which was kind of embarrassing seeing as Square had some of the greatest programmers in the industry for graphics and special effects, but couldn’t fit English text in their games! (Most of it was caused because their programmers quickly moved on to other titles, and no one else knew how to correct their code. So the loc team had to makes sure to schedule/lock-in the programmers’ time going forward to avoid repeats of this kind of thing.)
We had to fight for simple things like proportional fonts. (In fact if you look carefully, Xenogears and Chrono Cross used the same font. The dev team told me it was impossible to implement a proportional font, so I went and used the same font and quickly re-programmed the display routine to make it proportional in order to fit more text on screen. The director wasn’t too happy that I messed with his source code, but seeing as it was in there and was working, he allowed me to keep it.) I often had to call the teams out when they claimed our requests were impossible, or had to present them with design docs and workarounds until they learned what was expected for the foreign versions.
We lost some battles on the way, but I guess we eventually came to win the war. The dev teams slowly began to trust us with their babies, and even came to confer with us to plan ahead. (By the time the PS2 came along, the dev teams were planning for localization at the start. For instance we recorded The Bouncer’s voice overs in English so we could make the lip movements right in English, then dub it back into Japanese, as Japanese voice recording studios and actors were more familiar with dubbing to English than the reverse in America at the time.)
At the end, we had directors like Ishii-san and Tanaka-san working closely with us to name their titles, characters, and plan their game system. (For instance FFXI was planned to use English character names in both the Japanese and English versions from the beginning. We had to rework all the spell names in both English and Japanese to have a compromise that worked in both languages. I sat next to the Japanese name planner for 4 years and we worked together to name everything rather than simply translating what the Japanese team came up with.) But we had a major reset when Enix bought out Square and the companies merged to become Square Enix. Suddenly everything we achieved with Square had to be re-earned with the new Enix side. After 4 years, I was taken off FFXI, and assigned to the Dragon Quest team in order to bring the best practices we established at Square to the DQ dev teams.
Q: Did you have any involvement in the dialogue system employed in Chrono Cross, which would take a line of dialogue and spit it out differently based on which character was in the party? What was it like writing for such a diverse cast?
A: I actually created the English auto-accent generation system and programmed it as a stand-alone tool (for checking purposes) before handing the code over to the CC dev team to implement in the game itself.
The Japanese version had over 40 different accents that were mainly created by changing the verb endings and adding a few different ways of referring to the main character. It is relatively easy to pull off in Japanese. But in English, we had a serious problem. Either translate the game with no variation between characters at all, or somehow cram in 40+ different characterizations. We didn’t have the time or the PS memory to retranslate the same line up to 40 times for each character, so I took a stab at a system that would try to change the text real time into differing accents or speech patterns to match the colorful array of characters that made up the cast of the game.
I composed an initial characterization guide for all of the characters. Then I looked at what patterns of speech we would need to accomplish those accents and broke it down to the syllable or phoneme level. I made a list of each phoneme in English that would get changed according to different accents or speech-impediments, so that the translators and I could replace those phonemes with tokens in our text. At run-time, these tokens would be changed to match the speaker when that text was displayed on screen.
I made a rudimentary search-and-replace tool that replaced all those syllables or phonemes with the corresponding tokens. We translators would tweak these by hand, then we would add those tweaks to a dictionary file so the system gradually improved as we taught it better implementations. It was a real head-ache that I doubt many loc teams would ever be able to implement. (Thanks to Yutaka, Sammy, and Rich for putting up with all the extra workload I added to them!)
I flew over and worked in the US office for over a month (while the dev team was on vacation recovering from the Japanese master up). This allowed me to work side-by-side with my fellow translators while I also had the QA team help me on one of the greatest testing efforts ever. I created a tool that expanded out the tokens as if it was shown on the game screen, using the actual font and window widths. So the testers could run through every single line of text for every single character’s way of speaking and confirm that it read correctly (or at least within the parameters of the style guide, as of course this system did have its limitations) as well as to ensure it didn’t go over the line length limits. The QA testers were only a few weeks behind our translation, so we would be implementing their changes into both our text and the search-and-replace tool. After all that work, the testers then would have to recheck all that text in-game, to make sure the context was also correct and that everything was fine in the final product. So it wasn’t just the dev team and translators that needed to be sold on the idea, but the QA dept. I really don’t think most companies would allow someone to try to pull off such an ambitious system nowadays. Looking back, it was a risky move. While I agree some of the accents seem over-the-top, the game definitely benefited from having variety in the speech patterns to match the characters. (Seriously, would a pirate, a voodoo doll, a mermaid, an effeminate wanna-be-French Muskateer, a Mexican wrestler, a glam rocker, a pink fluffy dog, a skeleton clown and an alien, to name a few, all speak in the same manner?)
I still love the look on people’s face that see the name they enter for their character become all manner of nick-names depending on who is speaking to them. That was a simple trick that still surprises people to this day.
On a side note, the knowledge and courage gained from this auto-accent system also paved the way for a range of grammatical tokens that made latter games (FFXI, DQ series etc) more natural in English and European languages by adjusting the text real-time to handle capitalization, articles, singular/plural, masculine/feminine/neutral, and even German grammatical cases.