psygnisfive

I decided to start writing a devblog for the project I'm working on (a conversational AI toolkit). Lemme know if you have any questions or comments! :)

Check out the anarchist socialist experiments like Catalonia during the Spanish civil war. They tended to collapse because of inability to fight external forces, not because of bureaucratic bloat, etc.

Get a tinier belly with this one weird old hack!

I will only accept this as an accurate replication of Philip K Dick if it ends up doing meth and going crazy.

I'm not sure if Feynman intended it or not, and I wouldn't be surprised if he had, but this sounds like a sarcastic commentary on how even "great men" could be subject to appeals to authority (ie Compton).

For those interested, the traditional names for the directions in 4 dimensions (equivalent to "up" and "down", etc.) are "ana" and "kata".

The horrendously misleading part of this is that the experiment says to attach the candle to the wall. The lexical semantics of "attach" do not permit that you "attach the candle to the wall" by tacking a box to the wall and placing the candle in it. The verb attach simply doesn't mean that -- this demonstrates nothing.

Further, numerous studies have been done regarding lexical and conceptual semantics and the role of direct objects in semantic interpretation, and they demonstrate quite simply that whatever the direct object is of a sentence, this tends to be an affected item. Supporting a candle with a box is not affecting the candle in the ways relevant to human cognition, while tacking it to the wall, or melting it, is. So the phrasing again was misleading.

While I'm sure functional fixedness is a real phenomena, I would bet that it isn't even remotely as simple as this particular experiment suggested, and that a wide variety of linguistic techniques can affect how well people are at this. In fact, on the main page for the Candle Problem it's point out that a) using "box and tacks" instead of "box of tacks" in the instructions improves performance, while b) offering a monetary incentive decreases performance. So before you get all hyped up about how you want to think outside the box because my gosh imagine what awesomeness functional fixedness is preventing you from seeing, stop to consider that noone's got any clue what's really going on inside the mind, and you're probably just as likely to screw yourself by trying to manipulate yourself this way as you are to benefit yourself.

Captain, I don't see the relevance of your question.

\\//,

Unfortunately Ruby is not wrong. Ruby is simply being strictly correct from a logic point of view, where < is R^2 -> {T, F}, as opposed to being strictly correct from a math-notation point of view, where < is treated syncategorematically. Neither is correct, sans qualification; given that Ruby is not aimed at mathematicians, it's not unsurprising to find that it's not going to go through the extra parsing effort to use math notation.

While Lojban is "logical", human language is not. Or rather, it is, but it follows its own set of (well defined) rules. If Lojban does not follow these rule's, it's not truly learnable, and thus irrelevant. If Lojban _does_ follow these rules, and is genuinely learnable, then it offers little advantage except perhaps a pared down vocabulary that avoids ambiguity and redundancy, and a less flexible morphology and syntax. Assuming the latter, human languages change naturally, in order to facilitate learning and communication, and Lojban would change too, if actually used. Metaphor, ambiguity, imprecision, etc. all are the _products_ of language adapting itself to suit our needs better. So for all the "flaws" of natural languages, they work, and work well, and if Lojban were adopted by people, it would end up looking like a natural language, with all their "flaws". And that's a good thing.

Synonymy is by far one of the simplest kinds of tasks that are relevant to making computers understand language, despite its awesome usefulness. The Wordnet database (http://wordnet.princeton.edu/) actually has a wealth of synonymy and other word-word relations (probably significantly more extensive than Google's). In the future I can imagine that companies like Google with start to use things like syntactically related phenomena (e.g. syntactic "synonymy" between sentences like "The dog bit John" and "John was bitten by the dog"), in place of the simple word-word relations, and probably eventually even tackle things like answering questions based on the semantic content of the query + websites. There's actually some interesting work done by Phil Resnik (http://www.umiacs.umd.edu/~resnik/) here are UMD trying to do "sentiment analysis", whereby you essentially can detect spin/bias in a document by mapping grammatical structures to semantic features, and then analyzing that. Quite an interesting future. :D

Dennett, Pinker, and Ramachandrans books of the last 20 years.

The idea that any language lacks recursion is firstly very poorly supported — the only person who actually thinks this is true and demonstrated by any real language is Daniel Everett who thinks Pirahã lacks recursion, but there's a wealth of argumentation over whether this is true or not (for starters see Nevins, Pesetsky, and Rodrigues's "Piraha Exceptionality: a Reassessment" for a good discussion on this). Secondly, it's not crucial to any contemporary, wide-spread notions of Universal Grammar that languages have recursion in their structure — the theory of Syntactic Structures does not suppose, or even depend upon, expressions having recursive tree structures, it merely hypothesizes that the grammatical processes of a language can only employ structural relationships (and syntactic types) to determine when they can or cannot apply, but this idea has long since fallen into the dustbin of history as ever more non-structural phenomena (features, binding, etc.) became relevant to the grammatical theory.

Second, there has yet to be a demonstration from any known language that the language actually does not have an underlying tree structure to it — even the most free word-order languages such as Warlpiri display evidence of tree structure. NLP people may have a major bone to pick with core Chomskyan theory because of the incredibly difficult in parsing using it, but Chomskyan linguistics isn't about designing computationally efficient means of parsing sentences, it's about developing a good scientific theory about the nature of the constraints on human grammars. And I say core Chomskyan theory above because NLP people actually love various less core Chomskyan theory (Head-driven Phrase Structure Grammar, for instance, which has plenty of work done on it in the parsing world; or Combinatory Categorial Grammar which has been looked at quite a lot in the computational literature). The fact that NLP people prefer other methods such as dependency grammars doesn't say anything about the theoretical framework Chomsky introduced any more than their preference for numerical methods of differentiation and integration in mathematics bears on the theoretical importance or validity of precise symbolic differentiation and integration. Or to rephrase that more concisely — just because your computer isn't powerful enough to calculate these equations doesn't make the equations wrong.

I don't know whether or not Chomsky should be #1 on this list, but I do know that you have no clue what you're talking about with regards to contemporary linguistic theory.

It doesn't hurt that the gyroscopic behavior of the wheels also makes it unnecessary to do steering corrections at high enough speeds. Once you're going fast enough, you can hop off the bike and it'll still go upright for a while.

You're one of the first. ;)

I like that he's going to introduce quantifiers into his language (e.g. some(X) > every(Y)). What I'm concerned about tho is that this will have some very peculiar "errors" for some people. It seems like he's using what we linguists would call de re scope, which is to say the first quantifier has higher scope than the second one.

For example, using English, and translating into first order logic with restricted quantifiers, as well as Ruby:

  Some dog chased every cat
  ∃x:dog(x)[ ∀y:cat(y) [chased(x,y) ]]
  dogs.some? { |x| cats.all? { |y| chased x, y }}

Unfortunately, the English sentence is known to have another reading, one where each cat is chased by some or other dog, maybe not the same one for each cat:

  ∀y:cat(y)[ ∃x:dog(x)[ chased(x,y) ]]
  cats.all? { |y| dogs.some? { |x| chased x, y }}

Which means some poor coder might end up typing some(dog).chased(every(cat)) in his language, thinking he's getting the ∀ > ∃ reading (the second one above) and bamph, weird behavior. Even if you're familiar with how quantifiers work in this language, the similarity to english in this regard might easily lead someone to absentmindedly make this error.

That's not to say that he shouldn't have quantifiers -- I think we need some interesting natural-language like things like quantifiers in experimental languages -- but it's definitely something he should watch out for.

It's true that news headlines obey their own little grammatical rules. Language Log (http://languagelog.ldc.upenn.edu/nll/) has mentioned this a number of times. Tho noone in the field really cares about it, because its not all the relevant to language on the whole.

OH IRONY

I like how you call yourself a grammar nazi, cite the old (and wrong) definition of a sentence as something with a subject and verb, but then say that "it is pancakes" is not a sentence. Why not? It's got a subject -- it -- and it's got a verb -- is.

Further, English has subject-verb agreement not subject-object agreement or verb-object agreement, hence "it is pancakes" is completely fine. Sure there maybe be some semantic anomalies, what with the singularity of the subject, the plurality of the object, and the copula which provides some sense of equivalence between them, but that's entirely a semantic issue and not a grammatical one.

Further further, setting off non-sentences by periods is common practice. You may not like it, but everyone else accepts it. You're just an idiot.

Further further further, "it's" is completely acceptable English, and isn't an official contraction only because there ARE no official contractions -- unlike French, with its (wholly impotent) French Academy, there is no authority on English, and thus no such thing as official English. The best we can do (as if we should care) is to reference the great authors of English literature (whoever they may be) and if you go by them, then you're most certainly wrong.

Further further further further, it's not your language, it's everyones, and I think the headline is fine; and they're not abusing it, they're using it in a completely standard fashion.

</actual linguist>

You could make the whole 'a > 0 => increment' thing even cleaner than the first example:

  let a = (if a > 0 then a + 1 else a)
      in #something or other

There'll be CHAOS!

Think of all the boyscouts that will get lost without the use of a working compass! LOST I tell you!

The most obviously impactful part of the pole shift will be all the changes in migratory patterns as birds start migrating to the middle of the ocean or steppe.

Speaking of comparatives, a fellow grad student here at the University of Maryland is working on a phenomena called the comparative illusion, which involve sentences like the following:

More people have been to Russia than I have.

Think about that one for a bit. LL did a post on it, btw: http://itre.cis.upenn.edu/~myl/languagelog/archives/000860.h...

Pay no mind to the people who commented before me.

The phrase that follows the comparative preposition "than" is not necessarily a reduced clause. It's actually almost always a simple noun phrase, taking the default accusative case.

"I am taller than him" is Standard American. If you were to say "I am taller than he" you'll sound like a twat.

Doesn't make sense? So what. Language isn't about making sense, it's about getting points across. There's plenty more in English that's even more confusing (exceptional case marking? oh yeah, baby, figure that one out).

Hell, even if it WERE a reduced relative clause there are plausible reasons why it could still be "him", namely, "nominative" and "accusative" are structural case, not semantic case, so they get assigned depending on the assignee's position in the sentence structure, and you could easily reduce a clause and, in doing so, put its subject into precisely the structure necessary for accusative case.

Indeed, that's precisely the point. When findings are unreplicable _despite identical procedure_ it's usually considered a disconfirmation of the results of the initial experiment. Maybe not of the phenomena, but of the experiment. And since pretty much _all_ psi experiments have been disconfirmed, this effectively (tho not technically) disconfirms the phenomena.

This repeatedly comes up, as if Eliezer Yudkowsky convincing someone to say "Ok you can go free" is somehow indicative of how difficult or easy it would be for a real AI to get let out.

Let me put it this way: If it were that easy, out in the real world, I'm fairly certain prisons would be impossible.

Keep in mind that the gatekeeper is not obliged to sit and listen to the AI-in-a-box. The gatekeeper is not even obliged to pretend it's a real AI-in-a-box scenario and make the decisions he would make in that situation; you can sit there thinking, nay, SAYING "man this is gonna be the easiest $10 i ever made" for the whole two hours and tough shit for the would-be-AI.

I continue to refuse to accept the possibility of Eliezer having won those two rounds without being paired with a completely retarded gatekeeper.

This position is incredibly difficult to actually maintain, given that when _non_ parapsychologists (that is, people who don't presuppose the telepathy exists) reproduce the experiments, they consistently get contradictory results. Infact, its only when the believers in psi phenomena do these experiments that they discover positive results. Is this a problem with science as a whole, or is this a problem with parapsychologists?

I think this is sort of correct but not entirely correct. Natural language, for instance, affords us a great many mechanisms which are very much like a functional programming language. For instance, rather than something like so:

  for (x in men):
    if john saw x:
      the_man = x
      break

mary knows the_man

we can do something like

  mary knows men.select { |x| john saw x }

namely, relative clauses: "mary knows the man that john saw".

But relative clauses are difficult to process when they're not tail recursive (think of The House that Jack Built in English vs. Japanese, let's say).

So what it might suggest is that the brain is a functional tail-call optimized system, rather than generically imperative or generically functional. This might explain the apparent preference for imperative programming.

Strunk and White - 50 Years of Really Bad Advice on Grammar

Sexist, homophobic. I'm surprised he didn't also include some comment on how "Adama's a spic when before he was white this is a catastrophy!". I'll take my captains latino, my fighter pilots chicks, and my immortal secret agents omnisexual, thanks.

Bigotted asshole.

there's that word again, "heavy". why are things so heavy in the future? is there a problem with the earths gravitational pull?