It's useful when someone asks things like "will machines ever be able to think?"
And you might say "well, define 'think' ". In other words, that's not that interesting a question, since it is more about semantics than about reality.
Or you say it as Djykstra did, which I think is pretty clever.
And yes, your example about how words differ in different languages does a good job of explaining how this is just semantics. If a question becomes moot in a different language, it is a good indicator that the question is less about reality than it is about semantics.
Proponents of linguistic relativity (or "Whorfianism") have come up with some fantastically absurd claims over the years about how thoughts are constrained by language. And in the popular imagination, these claims would inevitably entangle with stereotypes about speakers of different languages. So I don't blame people for dismissing any discussion about the relationship between language and cognition as fringe theory.
But once you move away from the deterministic end of the spectrum, it is clear that there is some relationship after all between language and thought. The key point to understand is, as the linguist Roman Jakobson put it, that “languages differ essentially in what they _must_ convey and not in what they _may_ convey.”
An oft-repeated trope is that if a language does not have a word for some concept, then its speaker cannot understand this concept. This is plainly rubbish. Language does not constrain thought in that way. People can understand concepts without having the words to express them, and if need be, can always find new ways to express them in words, coining new terms if necessary.
Reality is much more subtle. Language only determines what we have to pay attention to. Take a look at the example of the distinction between cups and glasses in English as opposed to that between "chashka" and "stakan" in Russian. Suppose a Russian-English bilingual was presented with four objects—a cup-"chashka", a glass-"chashka", a cup-"stakan", and a glass-"stakan"—and instructed to group them into pairs of similar objects. What would the response be? Would it depend on the language that the instruction was given in? These are the sorts of interesting questions that we can ask about language and cognition. Speakers of different languages are capable of performing all the same cognitive tasks, but different languages may privilege different pathways to the solution.
The relationship between language and thought is unclear. To be polarising, psychologists believe that language influences thought/culture and linguists believe that thought/culture influences language. There are evidence to support both conclusions, but most of the debate in popular media focuses on examples that are severely misinterpreted. For example, if Language A does not distinguish green and blue, it does not necessarily mean that the speakers of Language A cannot distinguish them, but only that Language A does not distinguish them linguistically. A more concrete example: many languages, such as Chinese, lack linguistic tense (e.g. past, present, future)—using instead aspectual markers (e.g. perfective, imperfective)—, but this does not mean that Chinese speakers cannot understand temporal relations (nor that they could not still express such notions linguistically). Similarly, Russian has separate words for light blue and dark blue, but English speakers can still express these two distinct shades despite not having two separate, distinct words. In short, one has to be extremely cautious about making such broad generalisations without fully understanding the empirical data and, ideally, linguistic theory. See the debate in the Economist between Boroditsky and Liberman for more on this topic: http://www.economist.com/debate/overview/190
> This happens in software, you are limited in what you can create and do by what concepts you can easily express
Because the computer isn't really thinking. It's just mechanically following what the language says. So of course what you can do is limited by what the language can express, because that's all there is.
Human minds don't seem to be the same. They can have ideas without necessarily needing to start with language to build the idea on.
For one thing, we've all experienced having a thought but being unable to think of the word for it. Despite being unable to articulate it, you can look in the dictionary at possible words and tell from their definitions whether they match.
You could say that this is just our brains thinking in terms of that word without being able to recall its concrete form. (The essential meaning of a word and the spelling/sound of it might be handled separately by the brain.) Maybe that's why sometimes, but it's also possible to have a thought and not know that there is a word for it. You might tell a friend about someone who has an annoying habit of rigidly following and enforcing the rules even when that serves no constructive purpose, and your friend might tell you that's called being legalistic.
Still, you could argue that's still language-based thought because all you did was compose together several pieces of language ("rigidly following", "constructive purpose", etc.), and that your new vocabulary word is really just a shorthand for that composition. Maybe that is true for some words, but it can't be true for all of them. If words can only be introduced by reference to language, then there's no way language could have formed in the first place. There must have been a first word.
Another approach is a thought experiment: does a feral child (https://en.wikipedia.org/wiki/Feral_child) have thoughts? If a human grows up with no exposure to language, that will have profound effects on them, but I don't think those effects go as far as making it impossible to have thoughts.
Having said all that, do I think that language heavily influences thinking? Definitely. For one thing, exposure (and non-exposure) to certain ideas has a big influence on thinking. Words convey ideas, and when you learn a word, you learn its idea. It may also be true that ideas are easier to think about (internally) if they have a corresponding word. And it's certainly easier to discuss ideas if there are words for them, so those ideas will be discussed more often.
> Using language i.e. a scene graph representaiton to manipulate ideas and thoughts is far more efficient and faster.
This sounds like language thinker bias, or your definition of language is more abstract (like in Computer Science) than human languages. I can "construct a scene graph" just fine, and while I have to use words to describe the scenes to you, no words played a part in coming up with these examples or thinking about them:
- imagine a rope running through several wooden blocks on a table and pick up various blocks, how do the rest of them behave?
- imagine wooden discs of various sizes with holes, stack them on a pole, does it happen to be a valid solution for Tower of Hanoi?
- imagine a toy car, start lifting one end, at what point does it tip over?
I for sure am not using a human language for that.
> Different languages based on primitives might construct different scene graphs.
This is probably very much true in the same style as your native language biases your perception of color boundaries.
For example: Finnish has just one word for "leg", not separate words for "leg" and "foot", so I view the abstract concept of the whole limb as one thing, not separated into two parts at the ankle. In using language, I may accidentally say in English "I hurt my leg" after kicking my foot into something; they're the same.
For the reverse, Finnish has separate words for the front and back parts of the neck -- and I don't mean just throat, which is just the trachea area, I mean the whole frontal part of it. So, the thought of e.g. dirt on the back of my neck is separate from the though of dirt on the front of my neck. (No, I don't use the words in the thought.)
I do think differently in every language I know, even swearing mentally has distinct hue. It's a good question how useful is that, but there is no question languages add certain perspective.
This is silly, and off-topic, and you're right that the concepts in a language are what matter. Incidentally, that's also what this article is about, so let's focus on that.
We immediately start tripping over philosophical issues though. How is it possible to be confident that a language model is different from thought?
It seems quite possible that a model of thought could be equivalent to a language model and there is circumstantial evidence the two things are equivalent. A lot of people appear to think by reasoning something out with words. People are absolutely persuaded by words that pass a language model check, even if the words don't align to reality.
But what about how humans think? We make use of universals in our language all the time. And sometimes that fits a problem domain well, like making a GUI or running a simulation of things.
Because language is being treated as a thing complete in itself, as opposed to being related to an external world?
One of the issues in the 'Limitations' section was a difficulty with "common-sense physics", such as with the question "if I put cheese into the fridge, will it melt?"
To answer that question, you have to ask the right questions, such as "what is a fridge?" "what is a fridge for?" "What does it mean for cheese to melt?" "what is the cause of cheese melting?" Then one should consider the follow-on questions, such as "what are typical fridge temperatures?" "what are typical cheese melting points?" "what temperature is the cheese likely to be at initially?" (at which point, it helps to introduce the concept of room temperature, and note that it typically falls between the other two.) From facts such as the answers to these questions, one can deduce the probable outcome of putting cheese in a refrigerator, but none of the answers so far explicitly state it.
Is it plausible that any learning, solely from the structure of and correlations between examples of language use, could develop the sort of analytical/modeling approach that I have just outlined? Instinctively, I don't find it very plausible, but I am not very certain in that view.
IMO, one of the fundamental problems with AI research is the idea that thought is important. When a chess program keeps track of the best moves nobody really cares how that thought is stored or retrieved. What's interesting is how to measure what thoughts are useful (for a given problem) and how to generate useful thoughts efficiently (better than random.) But, when people focus on thought implying something that people do, but earthworms don't do they get into a tailspin.
Language as a portable method of conveying thought is a great subject of research. But, attempting to attack a human language in all it's glory is a huge pitfall. Nouns, verbs, additives, adverbs, intonation, tense, and a 29 years of knowledge and I frequently have no idea what someone is saying. But, linking the word ball, with the concept ball, with the sensory perception of ball might be possible today. Link that to some simple nouns like roll, toss, catch, etc and we can start a meaningful integration of language and actions.
Words are always semantic clouds. Not getting that is what sent philosophy down many a dead end, and the source of plenty of arguments regular people get into all the time. One doesn't need to try mapping between two languages - it's enough problem trying to map within the language itself.
>The feature of language that seems the most surprisingly powerful is placeholder words. The words like "someone", "somewhere", "somehow". >I call these the entropy words because they are the words for the information you don't have. They represent sets of possible things rather than a literal specific thing.
As an aside, you might know this, but that's how questions are formed in Chinese. You basically use an enunciative phrase with a placeholder word in the information you're missing ("this pencil is whose", "the dog is where", etc).
Going back to the point, I'm not sure whether the limitations of the language are actually a limitation of the brain that uses it. Sure, if you get into your mind that you want to ask a question, and the opposing party is able to understand that, even gestures will do with a little bit of ingenuity. But that has a lot of prerequisites: you need understanding that some information can be useful to you, understanding that other living beings own that information, some abstract thinking to express an idea that is not literal...
It's not even about questions I think, there's a jump between "tiger here now" and "tiger potentially here at some point, watch".
But still, questions particularly seem to be a clear cut "non existent outside of humans", even in weird emergent forms. There must be a reason they're not there, right?
I have found that, if you are careful to state the subject and object of a sentence, then it's really hard to pose any question that doesn't have a trivial and obvious answer.
TRY IT!
So could many of these things just be an artifact of language?
> In practice, it seems to me that usage and semantics are bent to a language-independent (but imperfectly perceived) reality rather than the other way around.
It seems to me that recognizing that usage and semantics have to respect a language-independent reality is why people will object if you claim that the sky is blue but not colored: because they recognize the language-independent reality that blue is a color. Plus the language-independent reality that if your intention is to communicate information, what you say has to be comprehensible to the listener.
In other words, "what is permissible or obligatory to say or believe" is driven by language-independent realities about facts in the world and requirements for communication to be possible.
This is an old program of the philosophers. Can you, even in principle, find the language at the back of reality? Can you eliminate all ambiguities when you refer to ordinary objects? Are we all talking about more or less the same things? How could we talk about exactly the same things?
The picture gets a little muddled when you think of the choices and distinctions each language makes. Some of them will contradict each other.
Suppose Language 1 has 150 words for snow, but only one for love, while Language 2 has 150 words for love, but only two for snow. Would the uber language preserve the distinctions for each word? Maybe you're losing something by learning all this detail.
You can do the same thing with the intensional meanings of words (connotations). I think one of the original Sapir-Whorf examples traces the meaning of a word for corn in Language A to words in Language B that mean "enemy food". Would the uber language remove connotations? It would probably lose the word from Language B.
But this gets into an even deeper point, which is that when you use a language, words aren't isolated. They frame a certain outlook on the world that hangs together. There were probably a bunch of words in Language B that talk about foreign or "enemy" things. Can you really say those words and mean them without taking on their perspective?
There are many more examples. We have this kind of thing in English too. We have "kill" and "murder". We have the Anglo-Saxon words and the Norman words.
And finally and most importantly, what perspective on the world does the super-language take? Objective? Involved but fair? Polemical? Propagandizing?
And you might say "well, define 'think' ". In other words, that's not that interesting a question, since it is more about semantics than about reality.
Or you say it as Djykstra did, which I think is pretty clever.
And yes, your example about how words differ in different languages does a good job of explaining how this is just semantics. If a question becomes moot in a different language, it is a good indicator that the question is less about reality than it is about semantics.
reply