Scribo

Be relevant! Or at least interesting.

The author of an entangled bank notes in the comments that my virtual world colour ascriptions don't really tell against relevance theory. NW writes:

I was hoping you were going to make a good point against Relevance Theory; but I'm afraid the information, though highly relevant in the 'usual' sense, is not so in the Sperber and Wilson sense. Although it would have a large amount of cognitive effect if you could use it, the fact that so much effort is required to process it puts it low down on the scale of relevance: so the most relevant input is some other factor that gives you less effect but for much less effort.

In all honesty I wasn't really trying to argue against relevance theory per se. (More like thinking outloud than anything else.) As S&W; point out numerous times, their theory gives us absolutely no basis for comparing relevance between contexts, only within a single context, so its not clear that any fact about cross-context comparisons could serve as evidence against their theory. Furthermore, their notion of relevance is purely comparative, and I was definitely using relevant in the ordinary sense in which something can fail to be relevant at all.

That said, I do think the fact that S&W; eshew giving any special weight to a notion of conversational goals is a bad thing, even in the absence of any knock down counterexamples along these lines.

One important aspect of relevance theory is the notion that hearers expand and modify context in order to maximise the relevance of an utterance -- hearers assume that an utterance is relevant, and they try to pick a context to process the utterance in that will validate that assumption.

To continue with the example fo virtual world colour ascriptions and con colours, if a SWG player were to tell me "That storm trooper is blue", there is a context available to me in which knowing the consider colour of the storm trooper to the speaker has tons of contextual implications. That context will include a large number of statements of the form: If the storm trooper cons blue to the speaker and P then I should Q. (e.g. If the storm trooper cons blue to the speaker and I am considerably weaker than the speaker then I should not move any closer to the storm trooper.) The contextual effect of the utterance will be to produce a large number of implications of the form If P then I should Q.

Now, standing outside relevance theory its quite clear to us that the reason we aren't interested in this context in SWG is that information that would cash out the P statements is very difficult to obtain, and the "If P then Q" statements are not useful to us unless we can determine which of the P statements is true. In contrast, the very similar context in other MMOGs is taken into account because the P information is obtainable. Its important to stress here that the other MMOG situations I am comparing are not situations in which you already have the P type information, but situations in which you don't have that information.

However, from the point of view of relevance theory, there seems to be nothing they can say that would explain why we ignore this context in SWG and yet pay attention to a very similar context in other MMOGs. In SWG, saying "that storm trooper is blue" will communicate that the storm trooper is wearing blue armour. Blue storm trooper armour is surprising, but its in general an isolated fact.

That doesn't make it irrelevant. S&W; claim that any utterance comes with a presumption of optimal relevance as follows:

(a) The ostensive stimulus is relevant enough for it to be worth the addresse's effort to process it.
(b) The ostensive stimulus is the most relevant one compatible with the communicator's abilities and preferences. (Relevance, 2nd Ed. 270)

I'm willing to stipulate that the surprising appearance of blue storm troopers is even as an isolated fact worth the processing time, though I note in passing that S&W; seem to have no substantive account whatsoever of what relevant enough amounts to. The problem is why the equally surprising appearance of blue orcs is not sufficiently relevant in other games. Why in other games does the default assumption that a blue orc is blue skinned get systematically trumped by con colour?

Its not that relevance theory says anything incompatible with this. Its that they offer no explanation of the phenomena. Relevance theory claims that we settle (as it were) on the first accessible interpretation of an utterance consistent with the presumption of relevance. What happens in MMOGs generally, but not in SWG, is a change in which interpretation is the first assessible one. S&W; argue that the notion of the point of a conversation, which Grice gives so much prominence to, only comes to bear when a shared goal is part of the context of utterance. When it comes to recognizing conversational implicatures they may even turn out to be right. However, it seems to me that cases like the virtual world colour attributions suggest that shared goals may need to be given more prominance, albeit at a slightly different level of discussion than that which S&W; are engaged in. Relevance theory has to my knowledge nothing in particular to say about what makes one interpretation more accessible than another, and in this case thats largely whats at issue. Its not that this case is one which shows relevance theory to be wrong. Its that relevance theory doesn't seem to particularly help us understand what is going on.

On having a point.

More on virtual world colour attributions -- I promise this post has a point, unlike the last one on this theme.

Ok, so I made a sweepingly general claim about mmog colour attributions here to the effect that "In general, in the context of mmogs, a red N is a N who cons red, regardless of what the usual criteria for being a red N are." This is true in many games with a consider (con) system. However it is not true of Star Wars Galaxies (SWG), and I think the reasons why tell us something about pragmatic processes.

Most mmog con systems are pretty simple -- they take into account exactly one factor -- character level. They don't take into account whether your character is a fighter, or a crafter, or a healer, or what equipment you have, or whether you have friends with you. This means that in order to decide whether you can succeed against that creature you need to take those factors into account yourself.

In contrast, the con colour of a creature in SWG changes depending on what weapons you have, how proficient you are with that weapon, how many people you have with you, and what other skills your character has.

On the face of it, you would think this would be vastly more useful than the simplistic con system used in other games. But interestingly, SWG players almost never communicate information about con colour to each other, and they don't use the mmog colour attribution scheme described above, even though con is colour coded in SWG.

Why not? Well, the simple reason is its too responsive. In a game like Everquest, if I know the character level of the person I am talking to, then I know everything I need to know about them to make use of the information encoded in "that orc is green". Indeed, given con information from a variety of characters of different levels I can triangulate in on the difficultly of the creature, and then use what I know about my characters abilities to make a decision about how difficult it would be for me.

In contrast, if a creature cons red to my friend in SWG, I need to know a huge amount of information about their characters skills, equipment, and who they are with in order to interpret that fact, and I need to be able to compare their skills with mine, something also quite difficult in SWG. Furthermore, because SWG con colour takes into account who you are with, if we are hunting together every creature will con the same to us until we go our seperate ways, so the point of sharing information during a hunt is greatly reduced.

There is a lesson there for game developers -- sometimes more complicated systems are less useful.

But I'm more interested in linguistic lessons.

When Paul Grice wrote about pragmatic processes in utterance interpretation, he gave center stage to the notion that conversations have points. He suggested that we are guided in our interpretations by the assumption that the participants will "make [their] conversational contribution such as is required at the stage at which it occurs, by the accepted purpose or direction of the talk exchange in which [they] are engaged" (Grice, H. P. 1989. Studies in the way of words. Harvard UP, Cambridge, Mass. 26. emphasis mine).

But many later linguists and philosophers of language seem to have lost sight of the point, as it were. Sperber and Wilson talk about relevance, which they analyse in terms of deductive productivity -- the more things you can conclude from a proposed interpretation of an utterance, the more relevant it is. Salience is another frequently harped on feature -- what has been mentioned recently, or what is patently obvious in the context.

All these things may be important. But while con colour is very salient in SWG -- no special command is required to get the information for example (its analogous to the cat sitting prominently on the knee of one of the speakers and the phrase "the cat") -- and certainly, it is potentially highly relevant in both the usual and the Sperber and Wilson sense, the flexibility of the system combined with the inaccessability of the information needed to make real use of it means that it doesn't help SWG players get to the point.

Let me expand. In a game with a simple con system, the information that a creature is red (or green, or blue) can quickly be marshalled to make the decision to either get out the vorpal axe of slaying or run screaming like a little girl. And its making that decision that is the point of the situation. Since the colours of creatures under normal attribution rules is not relevant to that decision, mmog colour attributions trump in almost all cases.

In SWG, the con colour might be helpful if the point of the discussion was to engage in a detailed analysis of how difficult a certain creature was theoretically (though even here there are so many factors that it isn't that helpful). But in a fight or flee situation, the point is to make a decision quickly. Con colours in SWG are so complicated and require so much additional background information that they can't be effectively used for that task. Thats why SWG players don't use mmog colour attributions. In other games, choosing fight or flee is so often the point and con colour so useful in making that decision that the con colour trumps other potential colour properties almost all the time. In SWG, fight or flee is still mostly the point, but con colour is so rarely useful that non-mmog colour attribution rules stay in place.

I think Grice's notion of the point of a conversational exchange has been sadly underrated and underemphasized. Thats my point.

That orc is green!

Some readers (and apparently I have readers now, but thats for another post) may be familiar with Ran Lahav's discussions of adjectives. Very briefly, Lahav has a lot to say about the fact that the criteria for being a red N vary wildly depending on what N is, and it is impossible to specify a rule that will tell you what the criteria will be. So for example, red apples are red on the outside, red crystals are red all the way through, watermelon is red because its red on the inside, red pens are those with red ink, and so on. Similarly for other adjectives.

Now, about the orc. One of my recreational pleasures is playing MMOGs (Massively Multiplayer Online Games). Now, in some of these virtual worlds one can find orcs. Literally of course one finds pixels and programing code, but in a large scale application of Kendall Walton's ideas about makebelieve, we say we find orcs. And orcs, unsurprisingly, come in the greened skin variety. Interestingly, this is usually not what I mean when I tell a fellow player that the orc is green.

One more mmog fact. In the killing and slashing varieties of mmogs, the creatures (mobs) are catergorized into groups depending on how difficult it will be for your character to cold bloodedly murder them with your vorpal sword of showiness. Frequently this categorization is represented to the player by colour coding -- the creature is surrounded by a coloured circle, or described by text in a certain colour, etc. Of course this classification is relative to my character -- as she gets meaner and nastier, mobs will change colour. The slang for this is that the mob cons red (or blue or green ...) -- con being short for consider, which is the command by which you get this information in Everquest, one of the biggest mmogs.

In most games, creatures that are so easy for your character to kill that she will not advance her skills by killing him are assigned the colour green.

"That orc is green!" means that the orc is in this category for my character. In general, in the context of mmogs, a red N is a N who cons red, regardless of what the usual criteria for being a red N are. To say that an orc has green skin or a dragon red scales, I have to do exactly that.

Its interesting (well, perhaps only to me) how quickly and smoothly new players adapt to this system. On anecdotal evidence alone, it seems that as soon as the con system is explained to them, they immediately shift to the mmog criteria for colour attributions. It rather supports Lahav's assertion that if we found ourselves in a world in which rats changed colour, and did so at different speeds depending on the rat, we would immediately know what a gradual rat was.

There is no point to this post really. I guess I should have warned you about that upfront.

Interweb?

I lost geek points on wednesday night.

There I was, happily watching Corner Gas (a Canadian show based on the implausible premise that Saskatchewan is funny -- surprisingly it works), when at the end we were told "Visit us on the interweb at www.cornergas.com". I turned to the SO in surprise. "The what?"

A quick google search later and I learn that, as well as being a common part of web design and internet service provider names, and a key word on a number of "adult sites", interweb is:

A sarcastic term for the internet. Often used in the context of parody regarding an inexperience, unskilled, or incoherent user.

Hay guys, I'm using teh interweb!!!

(Courtesy Urban Dictionary).

A new word -- a new sarcastic word at that -- and I missed it.

Language Log: Clairvoyance? No, just utterance processing

Geoffrey K. Pullum reports on spiffy utterance processing skills at Language Log today, and in doing so prompts this question from your humble host: Is there really any point to the continued debate over whether various pragmatically provided bits of content should count as part of "what is said" (e.g. Recanati) or impliciture (e.g. Bach)?

Both sides agree that there the information conveyed by an utterance includes pragmatically provided information that does not fit into the classical Gricean category of implicature (conversationally or otherwise). Of course, they disagree about some of the details of how this gets done, but their primary quarrel is over whether it should be counted as part of the explicit content of the utterance - i.e. what is said - or should be seen as merely implicit. This looks like a bigger disagreement than it really is, since Bach agrees that speakers commonly count the impliciture of a utterance as relevant to the truth conditions of that utterance, and Recanati allows that there is theoretical use to Bach's more restricted notion of what is said, though he would want to call it something else.

Its rather getting down to you say potato I say potato now -- I wonder if Professors Recanati and Bach would like to bury the philosophical hatchet on this one.

the five envelopes

I recently recieved a mismatched pair of comments from anonymous referees. One referee described the argument of the paper as "clear, succinct, and persuasive". The other commented that the paper was full of "vague statements and unsupported arguments".

The wonderful thing about these comments arriving together is that since both of these can't be true, I feel perfectly free to believe the former, and reject the later comment as misguided. Into the recycling bin with you, while I bask in my persuasiveness.

Sadly the journal editor did not see it my way, and so I will have to send the paper (which I am now quite bored with) off to another journal.

Which brings me to the five envelopes.

A much beloved advisor of mine once told me of a colleague of his (I wish I could work a sister of a friend in there too, but alas it is not to be) who having finished a paper, prepared five envelopes addressed to his five prefered journals, wrote five cover letters, and made five copies of the paper. If the first choice journal rejected the paper, he mailed off the second envelope, and if the second also rejected it, the third, and so on. Only if all five journals rejected it did he read any of the comments he recieved from referees.

On hearing of the recent pair of comments, the beloved advisor said "Thats why you use the five envelope method." I must admit the idea is growing on me.

If only because I do not want to think about this paper for one hour longer.