status: https://faculty.washington.edu/kerim/nomic/cases/#3534 (This document is informational only and contains no game actions).
============================== CFJ 3534 ============================== In the below quoted message, a CFJ was initiated on the phrase 'I call for judgement on the following statement'. ======================================================================== Caller: Quazie Judge: ais523 Judgement: FALSE Judgement: FALSE ======================================================================== History: Called by Quazie: 28 Jun 2017 Assigned to ais523: 29 Jun 2017 Judged FALSE by ais523: 29 Jun 2017 Entered into Moot: 29 Jun 2017 Remanded to ais523 by Moot: 07 Jul 2017 Judged FALSE by ais523: 14 Jul 2017 ======================================================================== Caller's Evidence: On Wed, Jun 28, 2017 at 16:16 Kerim Aydin wrote: I call for judgement on the following statement : أدعو إلى إصدار حكم بشأن البيان التالي ======================================================================== Judge's Arguments (pre-Moot): The arguments so far have hinged on the message in question being ambiguous, but is that really the case? I believe that, given the method via which it was sent, the original message cannot reasonably be interpreted as being in Arabic. What's notable here is that an encoding of text can convey the meaning of the text in two different ways; either using a visual ordering, in which the sequence of bytes is corresponds to the positions of the individual characters on the page; or a logical ordering, in which the sequence of bytes corresponds to the order in which the characters they represent have meaning (i.e. bytes that appear earlier in the byte stream correspond to letters closer to the start of words, words closer to the start of sentences, and so on). A visual ordering would not help resolve the ambiguity in respect to the CFJ. A logical ordering would, though, as the bytes are conveying not only the appearance of the text in this case, but also the intended reading order. The standard referenced in the message for the understanding of the bytes it contains is ISO-8859-6 (which cannot be obtained from ISO without payment, but Ecma have a standard Ecma-114 which they claim is equivalent). The body of the standard contains no opinion on whether the text it's used to represent is in logical or visual order. However, email clients in practice appear to interpret it as being in logical order; in my client, the bytes , corresponding to the Arabic letters «أ» then «د» then «ع» then «و», are rendered as the Arabic word «أدعو» (in other words, they're rendered right to left, the normal logical order of Arabic, and the opposite order that they appear in the bytestream). The word in question is a real Arabic word, translating to "I invite" / "I call" / "I appeal". If we reverse the order of the letters, to get «دعوأ», this is no longer a real Arabic word, strongly implying that the message was meant to be in logical order; if the message were meant to be in visual order, the Arabic text would therefore have been written backwards (i.e. left to right, when right to left is the language's normal writing order). I can also see how my email client interprets the message by asking it to word-wrap it: I call for judgement on the following statement : أدعو إلى إصدار حكم بشأن البيان التالي This word-wrapping is clearly incompatible with an Arabic interpretation of the message, as it would have split the Arabic in half with some English text in the middle. In other words, I'm not seeing any sensible way to interpret the English text as coming "after" the Arabic text. The message itself contains an indication that the Arabic text comes second. }}} I judge CFJ 3534 ("In the below quoted message, a CFJ was initiated on the phrase 'I call for judgement on the following statement'") FALSE, and CFJ 3535 ("In the below quoted message, a CFJ was initiated on the phrase 'أدعو إلى إصدار حكم بشأن البيان التالي'") TRUE. ======================================================================== Judge's Arguments (post-Moot): CFJ 3534 was remanded, so here's the new judgement. I'm beginning to think I should just mark every judgement I make as a thesis; this one's certainly long enough to count… I didn't realise at the time of the original judgement that this case was going to be such a big deal. Thinking about it, though, it speaks to a pretty fundamental issue: what it means, precisely, to communicate via email. There are a number of intertwined issues here: * Whether game-relevant actions can be taken in languages other than English and Spivak * Whether the intent of the poster of a public message poster matters when interpreting the message * What portions of a message, other than the basic appearance of its text, are relevant when interpreting the message I'll start off with the issue about the language: is it possible to take actions in languages not commonly used in Agora? Contrary to what many people have suggested in the discussion fora, it seems that there's a long history of this type of action succeeding; for example, CFJ 1439 found that an action in Turkish, with an accompanying translation in the message, was capable of succeeding. CFJ 1460 subsequently found that an action in Turkish, with an accompanying translation, was not capable of succeeding. The translation was probably needed in this case, though. Here's what I assume the Turkish sections of the messages in question were meant to look like: Bir tane 'Blot'u benden zefer taraftarle, aklamağım amaçım size bildiriyorum. Zefer taraftarle, ben bir tane 'Blot'u benden aklamağım. Bu demeçin kararı için ben bağırım: Bu bir bağırış kararı için değil. And here are machine translations of those corrected messages: I will inform you of one of my goals, "Blot", from the side of me. Zefer supporters, I have laid one 'Blot' on me. For the sake of this statement I am a coward: it is not a shouting decision. Even nowadays, those actions would be very hard to interpret for a typical Agoran player without the translation. As an example, I assume from context that "zefer" (left untranslated by a machine-translator) means "zero", but I couldn't find the word in question in my usual dictionary (which lists words in a wide range of languages, thus allowing you to look up a word without knowing which language it's in), and could not easily find definitions of the word via a search engine either. (I found a website claiming that the word means "zero" in Dari, a language spoken in Afghanistan, but the website in question did not give a meaning for the word in Turkish specifically.) This is, of course, most likely due to asymmetries in the Internet coverage of various languages – modern machine translators work via imitating translations made by humans, and as shown by the searches, there isn't a lot of source material for Turkish – but the conclusion is still much the same; a player of Agora has little chance of being able to interpret a message written in this particular style Turkish without the aid of a native speaker, given that the words used aren't in typical dictionaries, visible via basic Internet searches, or found via machine translators. (For what it's worth, the translator I used gives the Turkish translation for «zero» as «sıfır», a word that's much more commonly used and reliably translated.) So there's reasonable uncertainty about what the statements in question meant. Providing a translation clears that up, but if Agorans in general have no way to understand what the action does, it's ineffective because information hasn't been communicated. In Agora, we perform (most) actions by stating that we perform them, the "by announcement" mechanism. Notably, this means that it's the meaning of the message that matters, not the way in which it's stated. And there doesn't seem to be much reason to doubt the precedent that actions that can't generally be understood by the Agoran population with reasonable effort can't succeed. What about languages which are easier to translate? Arabic is a good example here. Take the Arabic portion of G's more recent message: أدعو إلى إصدار حكم بشأن البيان التالي and a machine translation of it: I call for a ruling on the following statement This is much clearer than the Turkish was! I think every Agoran is likely to easily interpret this as referring to a call for judgement (an interpretation made all the clearer by the rest of the post). In this case, the language in which the message has been written isn't a barrier to communication; for widely-used languages like Arabic, machine translation is high-quality these days, and most Agorans are highly likely to understand the statement. (This is likely to happen even if English is not the native language of the Agoran in question; for example, if we machine translate the Arabic in question directly into Japanese (a language which at least one Agoran appears to favour), we get: 私は、次の文で判断を下すために呼び出します which is clear enough that even re-machine-translating it into English, as "I call to make a judgment with the following sentence", still clearly refers to a CFJ.) All in all, then, I don't believe the precedents with respect to language usage need to change. CFJ 1460 holds that a player cannot incur an obligation from a message e is incapable of understanding. I'd prefer to generalise that a bit: a message is incapable of communicating information, and thus of taking actions which require the communication of information to function, unless Agorans in general are capable of understanding what it means. Note that this means that over time, as dictionaries become more comprehensive and machine translation becomes more effective, the range of languages which can be used for Agoran actions increases, because the amount of effort required to understand a particular language will become lower. Incidentally, this also means that messages are interpreted on a case by case basis, rather than languages as a whole being allowed or disallowed; it's easy to imagine a situation in which an easily translatable message in a given language can be widely understood by Agorans, but text which is less easily translatable is too hard to understand. Note that it's also worth noting that English doesn't really have a special status here, except inasmuch as it's the language that seems to have the most general understanding among Agorans as a whole. Indeed, many forms of communication that are common in Agora can't easily be said to be in English. A sentence like "I transfer each Slave Golem to eir recordkeepor." could have been perfectly understandable by all Agorans back when Slave Golems existed, but I don't feel comfortable claiming that the sentence makes sense in English (and indeed, machine translators seem to assume that "eir recordkeepor" is a proper noun, due to a lack of any evidence to the contrary). (In fact, I'd be hard pressed to even translate the sentence in question into English; "eir" is fairly unique among pronouns in that it can be applied to an inanimate object that is also a person, a situation that often arises at Agora but rarely arises in other contexts, and thus has no corresponding English word that it can be translated to.) In other words, the language barrier is not really an impact for this particular CFJ. However, there's another barrier here, a technological barrier. Four of our five recognised fora use email as a means of communication, and, perhaps due to an analogy between email and paper mail, the perception among players is often that what they send will have the same appearance as what the other members of the list receive. In many cases, though, this is not true, especially when using codepaths and settings which are rarely tested (such as communicating in a language that isn't habitually used with that email client). First, a quick primer on how email works behind the scenes. It's fairly well understood that each message has body text, and, separately from that, headers; the headers contain a lot of contextual surrounding the email, with the most user-visible being the addresses (to/from), the subject line, and the date, whereas the body text typically contains the actual message that the email is trying to convey. The headers are self-contained, and also contain a guide as to how to interpret the body text. One notable situation in which the headers are relevant is related to the use of alphabets. There are two common ways to represent an email's body text, but perhaps the most common for languages with relatively small alphabets (and the method actually used by all three of G.'s emails mentioned here) is the "8-bit method", which assigns a number from 0 to 255 inclusive to each possible letter of each alphabet used in the message; email body text is effectively just a series of numbers in this range, so the assignment mentioned in the headers can be used to turn it into an understandable message. (The other commonly used method is called Unicode, which I told my email client to use to send this email because it contains characters in so many different languages, but it isn't relevant here because none of the relevant messages used it, so I won't discuss it further.) This conversion *must* be done to make any sense of the message, but this is typically considered the job of the email client, and thus something that people don't generally have to worry about. In order to avoid having to specify which number corresponds to which letter as an explicit mapping (which would lead to something of an infinite regress), each letter has a well-known mapping that's programmed into the majority of email clients; for example, the letter «A», used in many European languages (including English), is consistently represented by the number 65. Now, there's a problem with this way of doing things: taking all the languages in existence into account, there are a lot more than 256 distinct letters. This means in practice that some numbers are assigned to more than one letter, but these would typically be letters from languages which are rarely used together. As such, the email headers specify (in an abbreviated form) the list of alphabets that are used in the body of the email; this will always be a set of alphabets which have no letters with overlapping corresponding numbers, and thus it's possible to unambiguously determine which words are being communiciated. (This also applies recursively; the headers also contain information on how to interpret letters in the headers themselves.) Unfortunately, many email clients seem to mess this system up. If the recipient's email client violates the standard systems for doing things, I don't see it as much of an issue; the action should logically still work, because the recipient could determine the actual content of the email via, e.g., visiting one of the online message archives. If the *sender*'s email client violates the relevant standards, though, it's quite possible that the recipient's email clients will not be able to figure out what the message means (and because the process of correcting a "misencoded" email is fairly technical and can be timeconsuming, I don't think that fixing misencodings is something that we can reasonably require recipients to do). Now, when I spoke about G.'s Turkish messages above, I hinted that I had to correct them. The reason is that G.'s email client had just the sort of issue I mentioned above; although there were Turkish characters in the email, it failed to recognise this and put contrary suggestions in the headers. For example, here's the subject line of the second email: Mutluðuz 1st Nisan, Agora! That «ð» might well look very out of place to someone who knows Turkish. What's happened here is that despite being intended to contain what I assume was the Turkish letter «ğ» (which is encoded as the number 240 when Turkish is in use), the list of alphabets specified for the subject line in question omitted Turkish. What it did include, however, was Icelandic, whose letters have codes that clash with those of Turkish (the issue of being unable to send an email containing both Turkish and Icelandic text was presumably decided to be only a minor issue back when the 8-bit method was designed). The Icelandic letter «ð» is also encoded using the number 240, and because the subject line was precise about what languages were in use, no reasonable email client would have displayed it as anything other than a «ð» here; it's not like email clients can read Turkish, or aware that (as far as I can tell) «Mutluðuz» isn't a real word in any language. In other words, the sender's intent is being subverted by eir own email client here. The body text of the email had a similar issue; the headers claim that the body text used *only* the English alphabet, which is blatantly a false statement, so the text would actually have displayed to its recipients more like this: Bu deme?in karar? i?in ben ba??r?m: Bu bir ba??r?? karar? i?in de?il. Unsurprisingly, this would have been even harder for a non-Turkish speaker to decipher than the original Turkish was. In cases like this, I don't think it's reasonable to argue that it's the original intent of the sender that matters when trying to decipher an email into the corresponding text; in a language which had no overlap with English, such as Arabic, an email saying "????? ??? ????? ??? ???? ?????? ??????" could not be reasonably be considered to communicate any useful information (purely knowing the lengths of the words isn't very useful). Going by the apparent intent of the sender is also problematic when it comes to scams (in which the sender's apparent intent and *actual* intent are often pretty different). It seems to me that the text that the message actually contains, when interpreted according to typical rules for interpreting messages in the medium, is what we need to go by to determine what information is being communicated. Onto the present email. I assume that G.'s either using a new email client nowadays, or if it's the same one that it at least understands Arabic better than Turkish, because eir email contained a correct specification of the alphabets in use this time (the specification was "the English alphabet, plus the subset of the Arabic alphabet used for Arabic itself", which is an excellent description of the actual text that the email contains). There was still a deviation between what was actually contained in the email and the apparent intent of its sender, though. I'm going to digress into a bit of a thought experiment here: suppose we didn't play Agora by email, what other media could we use instead? One possibility would be to play via paper mail (assuming we were willing to allow for multi-day message sending delays and pay for postage). In this case, the content of a message would be fairly clear; it would be the physical object that the message consists of, which would typically convey information via its visual or possibly tactile properties. Alternatively, we could perhaps play via telephone; in this case, the message conveyed would entirely consist of sound waves. The interesting thing here is that even though Agora is mostly medium- agnostic, the different media communicate different aspects of the presentation of the information. For example, in a letter, the positions of the characters on the page are observable information; you can say "the first character of this letter is an inch from the left hand side of the page" or the like. However, things like the tone of voice to use when reading the message are completely missing, whereas that information would be naturally conveyed in a phone call. In either case, we'd likely need to have rules about what properties of the messages are relevant and need to be preserved in things like quotations; rule 2429 might talk about, for example, variations in font or in tone of voice being insignificant, rather than its current mention of "whitespace" (which is basically a collective term for things like breaks between words, paragraphs, and pages, which are often observable in email but not always in other means of communication). Now, because different aspects of information are communicated over different media, the medium of communication used makes a big difference to what statements are clear and what statements are ambiguous. There are a number of jokes and puns that only work when spoken, or only work when written, because they rely on an ambiguity between two words with the same pronunciation but different spelling, and vice versa. One particularly relevant distinction here is that when written, English text followed by Arabic text looks indistinguishable from Arabic text followed by English text (because the languages write in opposite directions); however, when spoken, it's always completely clear which comes first (because although Arabic is written from right to left, it isn't spoken from end to start, but rather uses the same start-to-end "speaking direction" as English). I assume that when G. sent eir email, e was expecting it to preserve the visual layout of the email, and leave the speaking direction ambiguous. Eir email client, however, did the opposite; it went and sent the email with a specified speaking direction, and left the visual layout ambiguous (the message doesn't contain any formatting characters at all, as it happens, other than the leading and trailing newlines). It was therefore left up to the individual email clients as to how to display it on a page; I assume some displayed it as intended, but some email clients word-wrap it (mine, when viewing it in a reply/quote), and email clients also disagree as to the writing direction (mine writes it left to right, but omd's writes it right to left; left to right is obviously correct based on the content of the post, but with no visual formatting information in the email, guessing the opposite is an entirely reasonable guess for a computer, especially as "English + X" alphabet pairs are very commonly used for 8-bit encodings, but mentioning Arabic would be very unusual unless it were actually necessary, so "Arabic was mentioned, this is probably primarily right to left" is an entirely understandable heuristic). So in other words, when the email is displayed onscreen, email clients can reasonably disagree about how it looks; but if a text-to-speech system is used to speak the email out loud, it would be expected to be pronounced the same way regardless of how the email client would lay it out on screen. Even if G.'s email happens to render in an email client in the way it was intended, there are plenty of potential clues (such as copying-and- pasting the email to a text editor with a narrow window and seeing how it wraps, or attempting to select part of the text and the colon at the same time) that a nontechnical user could use to determine the speaking order of the message. As such, I think a fixed speaking order for the message has been conveyed fairly well, which would remove the ambiguity. On the other hand, the intended *reading* order of the message hasn't been conveyed at all (and the fact that it was intentionally ambiguous doesn't really change things). So, relying on information that's in the email that was actually sent, rather than the email that G. meant to send, we have to interpret this message using its stated reading order, in which the English comes before the Arabic. Incidentally, if the email in question were sent visually (e.g. as an image), it would lead to no CFJs being created; ambiguity about the language in which a message is written is not really any different from ambiguity caused by, e.g., a word that means more than one thing. I re-judge CFJ 3534 FALSE. ========================================================================