[HN Gopher] Recent Advances in Natural Language Processing
___________________________________________________________________
Recent Advances in Natural Language Processing
Author : saadalem
Score : 139 points
Date : 2020-08-16 18:03 UTC (4 hours ago)
(HTM) web link (deponysum.com)
(TXT) w3m dump (deponysum.com)
| YeGoblynQueenne wrote:
| Separate comment because separate point.
|
| To my mind, all this benchmark-beating is just pointless faffing
| about and not a little frustrating to see that this is where the
| current state of the art is at. NLP systems keep getting better
| and better at beating benchmarks and yet we still don't have
| artificial agents that can hold an intelligent conversation, even
| in a restricted context.
|
| Or rather, we had such programs since the late 1960's, with no
| one else's than Terry Winograd's own doctoral thesis system,
| SHRDLU [1], but it seems that in the race to build powerful
| classifiers, we forgot how to build systems that can interact
| intelligently with a human.
|
| Here's a small excerpt from SHRDLU (taken from the wikipedia page
| on SHRDLU, originally in an appendix in Terry Winograd's thesis):
| Person: Is there a large block behind a pyramid?
| Computer: YES, THREE OF THEM: A LARGE RED ONE, A LARGE GREEN
| CUBE, AND THE BLUE ONE. Person: Put a small one onto the
| green cube which supports a pyramid. Computer: OK.
| Person: Put the littlest pyramid on top of it. Computer:
| OK. Person: Does the shortest thing the tallest pyramid's
| support supports support anything green? Computer: YES,
| THE GREEN PYRAMID. Person: What color is it?
| Computer: BY "IT", I ASSUME YOU MEAN THE SHORTEST THING THE
| TALLEST PYRAMID'S SUPPORT SUPPORTS. Computer: RED
|
| Note that this is not just mindless text generation. The
| conversation is held in the context of a "blocks world" where a
| robot with a single hand and eye ("Computer" in the excerpt
| above) moves blocks of various shapes and colours around, as
| directed by a human user in free-form natural language. When the
| Computer says "OK" after it's directed to "put the littlest
| pyramid on top of it" it's because it really has grabbed the
| smallest pyramid in the blocks world and placed it on top of the
| small block in an earlier sentence, as the Person asked. The
| program has a memory module to keep track of what ellipses like
| "it", "one" etc refer to throughout the conversation.
|
| SHRDLU was a traditional program hand-crafted by a single PhD
| student- no machine learning, no statistical techniques. It
| included, among other things, a context-free grammar (!) of
| natural English and a planner (to control the robot's hand) all
| written in Lisp and PLANNER. In its limited domain, it was
| smarter than anything ever created with statisical NLP methods.
|
| ______________________
|
| [1] https://en.wikipedia.org/wiki/SHRDLU
| quotemstr wrote:
| Why is it surprising that a CFG can approximate a subset of
| English grammar?
| trott wrote:
| > Why is it surprising that a CFG can approximate a subset of
| English grammar?
|
| "Colorless green ideas sleep furiously" is a famous example
| of a sentence that is grammatical, but meaningless. The goal
| of SHRDLU was far more ambitious than approximating English
| grammar.
| liuliu wrote:
| We knew hand-crafted program in limited domains can work for
| NLP, computer vision and voice recognition long time ago. The
| challenge is always, that the limited domain can be extremely
| limited, and to get anything practically interesting requires a
| lot of human involvement to encode the world (expert system).
|
| Statistical methods traded that. With data, some labelled, some
| unlabelled and some weakly-labelled, we can generate these
| models with much more efficient human involvement (refine the
| statistical models and labelling data).
|
| I honestly don't see the frustration. Yes, current NLP model
| may not be the "intelligent agent" everyone looking for yet to
| any extent. But claiming it is all faffing and no better than
| 1960s is quite a stretch.
| joe_the_user wrote:
| The thing that qualifies as "faffing" in my opinion isn't the
| statistical NLP programs, which "are what they are" but
| _claims of progress based primarily on benchmarks_ , as
| YeGoblynQueene rightly states.
|
| And "limited domain" is relative. A program that gets many
| aspects of language right talking about a small world might
| be said to have a larger domain than a program that outputs
| stream of semi-plausible, semi-gibberish involving the whole
| of English language. Which again isn't saying modern NLP is
| nothing but rather we should a somewhat better to talk of it
| (and machine learning generally) than "hitting benchmarks".
| YeGoblynQueenne wrote:
| >> We knew hand-crafted program in limited domains can work
| for NLP, computer vision and voice recognition long time ago.
|
| Yes, we did. So- where are the natural language interfaces by
| which we can communicate with artificial agents in such
| limited domains? Where are the applications, today, that
| exhibit behaviour as seemingly intelligent as SHRDLU in the
| '60s? I mean, have you personally seen and interacted with
| one? Can you show me an example of such a modern system?
| Edit: Note again that SHRDLU was created by a single PhD
| student with all the resources of ... a single PhD student.
| It's no stretch to imagine that an entity of the size of
| Google or Facebook could achieve something considerably more
| useful, still in a limited domain. But this has never even
| been attempted.
|
| Yes, it is faffing about. Basically NLP gave up on figuring
| out how language works and switched to a massive attempt to
| model large datasets evaluated by contrived benchmarks that
| serve no other purpose than to show how well modern
| techniques can model large datasets.
| joe_the_user wrote:
| Good question!
|
| As you say SHRDLU was the best program of that time.
|
| I've read Winnograd's book and looked at the SHRDLU source
| code (not that I'm much of a lisp hacker but I got some
| time). It's built-on a parser and planner (logic program,
| pre-prolog). And it's built the old-fashioned-way,
| rewriting the source code with the parser rewriting input
| and then re-rerunning and other harry things. I think this
| achieves the parsing of idiomatic constructions on a high
| level. I believe the "raw lisp" of the day was both
| incredibly powerful since you could do anything and
| incredibly hard to scale ... you could do anything.
|
| Winnograd wrote it himself but I think that's because he
| had to write it himself. In a sense, a programmer is always
| most productive when they are writing something by
| themselves because they don't have to explain anything they
| are doing (until the fix-ups and complexity overwhelm them
| but the base fact remains). And in the case of SHRDLU,
| Winnograd would have had an especially hard time explaining
| what he was doing. I mean, there was a theory behind - I've
| read Winnograd's book. But there was lots and lots of
| brilliant spaghetti/glue-code to actually make it work,
| code that jumped module and function boundaries. And the
| final had a reputation of being very buggy, sometimes it
| worked and sometimes it didn't. And Winnograd was a
| brilliant programmer and widely read in linguistics and
| other fields.
|
| The software industry is an industry. No company wants to
| depend on the brilliance of it's workers. A company needs
| to produce based on some sort of average and a person
| working with just average skills isn't going to do SHRDLU.
|
| So, yeah, I think that's why actual commercial programs
| never reached the level of SHRDLU
| _emacsomancer_ wrote:
| So what you're saying is that we really shouldn't be
| relying on industry for good AI?
| joe_the_user wrote:
| Well, an "industrial model" is a model of a factory and
| it seems unlikely you could product a fully-functional,
| from scratch GFAI program like SHRDLU in something like a
| factory.
|
| Perhaps one could create a different kind of enterprise
| for this but it's kind of an open problem.
| YeGoblynQueenne wrote:
| Winograd made changes to the Lisp assembly to make SHRDLU
| work and he never back-ported them to his SHRDLU code,
| but his original version worked fine and was stable. The
| experience of breaking refers to later versions that were
| expanded by his doctoral students and others and to ports
| to java and I think C. The original code was written in
| 1969 but inevitably suffered from bit rot in the
| intervening years so it's true that there is no stable
| version today that can reliably do what Winograd's
| original code could do... but Winograd's original code
| was rock solid, according to the people who saw it
| working.
|
| There's some information about all that here:
|
| http://maf.directory/misc/shrdlu.html
|
| _[Dave McDonald] (davidmcdonald@alum.mit.edu) was Terry
| Winograd 's first research student at MIT. Dave reports
| rewriting "a lot" of SHRDLU ("a combination of clean up
| and a couple of new ideas") along with Andee Rubin, Stu
| Card, and Jeff Hill. Some of Dave's interesting
| recollections are: "In the rush to get [SHRDLU] ready for
| his thesis defense [Terry] made some direct patches to
| the Lisp assembly code and never back propagated them to
| his Lisp source... We kept around the very program image
| that Terry constructed and used it whenever we could. As
| an image, [SHRDLU] couldn't keep up with the periodic
| changes to the ITS, and gradually more and more bit rot
| set in. One of the last times we used it we only got it
| to display a couple of lines. In the early days... that
| original image ran like a top and never broke. Our
| rewrite was equally so... The version we assembled circa
| 1972/1973 was utterly robust... Certainly a couple of
| dozen [copies of SHRDLU were distributed]. Somewhere in
| my basement is a file with all the request letters...
| I've got hard copy of all of the original that was Lisp
| source and of all our rewrites... SHRDLU was a special
| program. Even today its parser would be competitive as an
| architecture. For a recursive descent algorithm it had
| some clever means of jumping to anticipated alternative
| analyses rather than doing a standard backup. It defined
| the whole notion of procedural semantics (though Bill
| Woods tends to get the credit), and its grammar was the
| first instance of Systemic Functional Linguistics applied
| to language understanding and quite well done." Dave
| believes the hardest part of getting a complete SHRDLU to
| run again will be to fix the code in MicroPlanner since
| "the original MicroPlanner could not be maintained
| because it had hardwired some direct pointers into the
| state of ITS (as actual numbers!) and these 'magic
| numbers' were impossible to recreate circa 1977 when we
| approached Gerry Sussman about rewriting MicroPlanner in
| Conniver." _
|
| Regarding the advantage of a lone programmer- that's
| real, but large teams have built successful software
| projects before, very often. I guess you don't even need
| a big team, just a dozen people who all know what they're
| doing. That shouldn't be hard to put together given FANG-
| level resources. Hell, that shouldn't be hard to do given
| a pool of doctoral students from a top university... but
| nowadays even AI PhD students would have no idea how to
| recreate something like SHRDLU.
|
| Edit: I got interested in SHRDLU recently (hence the
| comments in this thread) and I had a look at Winograd's
| thesis to see if there was any chance to recreate it. The
| article above includes a link to a bunch of flocharts of
| SHRDLU'S CFG but even deciphering those hand-drawn and
| occasionally vague plans would take a month or two of
| doing nothing else, something for which I absolutely do
| not have the time. And that's only the grammar - the rest
| of the program woudl have to be reverse-engineered from
| Winograd's thesis, examples of output from the original
| code or later clones, etc. That's a project for a team of
| digital archeologists, not software developers.
| joe_the_user wrote:
| _"... his original version worked fine and was stable.
| The experience of breaking refers to later versions that
| were expanded by his doctoral students and others and to
| ports to java and I think C. "_
|
| I can believe this but I think your details overall
| reinforce my points above.
|
| _For a recursive descent algorithm it had some clever
| means of jumping to anticipated alternative analyses
| rather than doing a standard backup._
|
| Yeah, fabulous but extremely hard to extend or reproduce.
| The aim of companies was to _scale_ something like this.
| It seems like the fundamental problem was only a few
| really smart people could programs to this level and no
| one could take them beyond it (the saw that a person half
| to be twice as smart to debug a program as to write comes
| in, etc).
| lacker wrote:
| The modern natural language interfaces with limited domains
| are Alexa and Siri. Yes, they're limited. But they are far
| more impressive and useful than SHRDLU.
| YeGoblynQueenne wrote:
| Alexa and Siri (and friends) are competely incapable of
| interacting with a user with the precision of SHRDLU. You
| can ask them to retrieve information from a Google search
| but e.g. they have no memory of the anaphora in earlier
| sentences in the same conversation. If you say "it" a few
| times to refer to different entities they completely lose
| the plot.
|
| They are also completely incapable of reasoning about
| their environment, not least because they don't have any
| concept of an "environment" - which was represented by
| the planner and the PROGRAMMAR language in SHRDLU.
|
| And of course, systems like Siri and Alexa can't do
| anything even remotely like correctly disambiguating the
| "support support supports" show-off sentence in the
| excerpt above. Not even close.
|
| Edit: Sorry, there's a misunderstanding about "limited
| domain" in your comment. Alexa and Siri don't operate in
| a limited domain. A "limited domain" would be something
| like being in charge of your music collection and nothing
| else. Alexa annd Siri etc are supposed to be general-use
| agents. I mean, they are, it's just that they suck at
| it... and would still suck in a limited domain also.
| rvense wrote:
| My impression is that the systems never progressed much
| after SHRDLU even though there were attempts at larger
| scale "expert systems". But adding more advanced rules and
| patterns proved extremely difficult and did not always have
| the expected effect of making the systems more general.
|
| There was the whole AI winter thing, of course, but that
| was as much a result of things not living up to the hype as
| a cause.
| YeGoblynQueenne wrote:
| This doesn't directly address your question, though it
| perhaps can give you some pointers if you want to read
| about the history of AI and the AI winter of the '80s,
| but in a way SHRDLU featured prominently in the AI
| winter, at least in Europe, particularly in the UK.
|
| So, in the UK at least the AI winter was precipitated by
| the Lighthill Report, a report published in 1973,
| compiled by a Sir James Lighthill and commissioned by the
| British Research Council, i.e. the people who held all
| the research money at the time in the UK. The report was
| furiously damning of AI research of the time, mostly
| because of grave misunderstandings e.g. with respect to
| combinatorial explosion and basically accused researchers
| of, well, faffing about and not doing anything useful
| with their grant money. The only exception to this was
| SHRDLU, that Lighthill praised as an example of how AI
| should be done.
|
| Anyway, if you have time, you can watch the televised
| debate between Lighthill and three luminaries of AI, John
| McCarthy (the man who named the field, created Lisp and
| did a few other notable things), Donald Michie (known for
| his MENACE reinforcement-learning program running on...
| matchboxes, and basically setting up AI research in the
| UK) and Richard Gregorie (a cognitive scientist from the
| US for whom I confess I don't know much). The (short)
| wikipedia article on the Lighthill Report has links to
| all the youtube videos:
|
| https://en.wikipedia.org/wiki/Lighthill_report
|
| It's interesting to see in the videos the demonstration
| of the Freddy robot from Edinburgh, that was capable of
| constructing objects by detecting their components with
| early machine vision techniques. In the 1960's.
| Incidentally:
|
| _Even with today 's knowledge, methodology, software
| tools, and so on, getting a robot to do this kind of
| thing would be a fairly complex and ambitious project._
|
| http://www.aiai.ed.ac.uk/project/freddy/
|
| The above was written sometime in the '90s, I reckon but
| it is still true today. Unfortunately, Lighthill's report
| killed the budding robotics research sector in the UK and
| it has literally never recovered since. This is typical
| of the AI winter of the '80s. Promising avenues of
| research were abandoned not because of any scientific
| reasons, as is sometimes assumed ("expert systems didn't
| scale" etc) but, rather, because pencil pushers in charge
| of disbursing public money didn't get the science.
|
| Edit: A couple more pointers. John McCarthy's review of
| the Lighthill Report:
|
| http://www-
| formal.stanford.edu/jmc/reviews/lighthill/lighthi...
|
| An article on the AI winter of the '80s by the editor of
| IEEE Intelligent Systems:
|
| https://www.computer.org/csdl/magazine/ex/2008/02/mex2008
| 020...
| rvense wrote:
| Interesting, thank you for the clarifications.
| ericbarrett wrote:
| I read grandparent's post as saying that despite all the
| research and untold amounts of compute power poured into NLP
| over the decades, its practitioners have yet to address the
| original real-world goals that led us to study it in the
| first place. Missing the forest for the trees, if you will.
|
| (I don't assert that it has hasn't. I know just enough about
| the topic to see how little I know. But it seems, from the
| outside, a valid criticism, and not one unique to the field.)
| tmalsburg2 wrote:
| Since this post is receiving downvotes, I'd like to know
| which part of the HN guidelines it is violating.
| mqus wrote:
| Not a single mention if this is only applicable to english or to
| other natural languages. Afaict this mostly lists advancements in
| ELP (english language processing), Especially the Winograd schema
| (or ar least the given example) seems to be heavily focused on
| english.
|
| Relevant article for this problem:
| https://news.ycombinator.com/item?id=24026511
| MiroF wrote:
| But there's no reason the models are english specific...
| woodson wrote:
| In a certain way, they could be, simply because their
| structure may work better on the type of language that
| English is. It may not work as well for languages that
| exhibit other grammatical patterns (e.g. morphologically rich
| languages, or those that exhibit superficially more flexible
| word order which nonetheless conveys information about
| topic/focus).
| dvduval wrote:
| Some languages like Chinese have a large corpus of
| information that can be studied, and there has been a
| continuous effort to standardize the language. it wouldn't
| surprise me if it turns out to be easier than English. I
| would expect it would be more difficult when you're talking
| about languages that have regional differences in usage of
| words, perhaps northern Vietnamese vs. Southern Vietnamese,
| or the Spanish of Spain versus the Spanish of Mexico.
| east2west wrote:
| Chinese has large regional differences, too. Cantonese
| has different grammar, not just word usage. Modern
| Chinese grammar is somewhat similar to English while
| Cantonese follows ancient Chinese, which places verb to
| the end of sentence.
| dvduval wrote:
| True, and Mandarin would be the better word to use here.
| Mandarin and Cantonese are for all intents and purposes
| separate languages. They are more different than English
| and Spanish.
| yorwba wrote:
| If we're going down that path, then it should more
| specifically be Standard Mandarin, since Standard
| Mandarin and e.g. Southwestern Mandarin are only
| partially mutually intelligible like French and Spanish
| are partially mutually intelligible.
| DonaldFisk wrote:
| Cantonese is, like English and Mandarin, SVO.
|
| E.g. I bought this book: ngoh(S) maaihjo(V) nibun syu(O).
|
| Contrast with Japanese: watashi wa(S) kono hon o(O)
| kaimashita(V).
|
| Differences from Mandarin Grammar are listed here: https:
| //en.wikipedia.org/wiki/Cantonese_grammar#Differences_...
|
| Classical Chinese is also SVO:
| https://en.wikipedia.org/wiki/Classical_Chinese_grammar
| kingkawn wrote:
| All language has significant regional difference if it
| covers enough regions
| akerro wrote:
| I'm 100% sure NLP will do the same mistake with non-English
| language as face recognition did with faces of black people.
|
| I'm taking bets.
| m0zg wrote:
| For some of them, there is. Russian (my mother tongue), for
| example, has the kind of morphology that makes linguists'
| hair stand on end. Also, verbs and nouns are gendered (and
| therefore they must agree, and you must know what everything
| refers to at all times, often pretty far away from where the
| noun was first mentioned), and there are declensions on
| various parts of speech, and they must agree as well. And
| words can be freely formed out of several roots, prefixes,
| and suffixes. For any of this to be understood or generated,
| your model has to be able to model all of that, which is much
| harder than modeling English.
| YeGoblynQueenne wrote:
| Yes, there is. A "model" is a set of parameters optimisted by
| some algorithm or system trained on the data in a specific
| dataset. Thus, a language model trained on a dataset of
| English language examples is only capable of representing
| English language utterances, not, e.g. French, or Greek, or
| Gujarati utterances. Diagrammatically: data
| --> system --> model
|
| What is not necessarily English-specific are the systems used
| to train different language models, at least in theory. In
| practice, systems are typically hand-crafted and fine-tuned
| to specific datasets, to such a degree that most of the work
| has to be done anew to train on a different dataset.
| ascavalcante80 wrote:
| NLP is great for many things, but, from my own experience as a
| NLP developer, machines are not even close to understand human
| language. They can interpret well some kind of written speech,
| but they will struggle to grasp two humans speaking to each
| other. The progress we are make on building chatbots and vocal
| assistants is mainly due to the fact We are learning how to
| speaking to the machines, and not the contrary.
| p1esk wrote:
| Note this is pre-GPT-3. In fact I expect GPT-4 will be where
| interesting things start happening in NLP.
| curiousgal wrote:
| I honestly don't get where the big deal is with NLP. So far the
| most useful application has been customer support chatbots and
| those still don't rise to the level of having an actual human
| that can understand the intricacies of your special request.
| feral wrote:
| I work on such a bot @Intercom.
|
| A lot of support requests aren't actually intricate and
| special: there's almost always loads of simple requests that
| come in again and again.
|
| When you ask a request like that and get the answer
| instantly, that really is ML delivering value.
|
| You mightn't think it, but a _lot_ of people work in customer
| support, and spend a lot of time answering rote questions
| again and again.
|
| People talk about how much hype there is, or the chance of
| another AI winter. Yes there's a ton of hype - but I think
| they aren't considering the real value already being
| delivered this time around.
|
| Everyone is excited about GPT3, but there's already been
| amazing progress in practical NLP already, over the last few
| years.
| ben_w wrote:
| Current NLP is bad. Still useful (Google search increasingly
| feels like it is doing NLP to change what I asked for into
| what it thinks I meant) but bad. A hypothetical future
| "perfect" NLP can demonstrate any skill that a human could
| learn by reading, and computers can read so much more than
| any given human.
| plafl wrote:
| Is reading enough to understand the real world without
| direct experience of the real world? Is there any research
| that tries to answer this question?
| p1esk wrote:
| _Is there any research that tries to answer this
| question?_
|
| That's the whole point of the experiment called GPT-3.
| rvense wrote:
| As of about ten years ago when I received my degree in
| linguistics, I understood there were two schools on this
| issue:
|
| A: Of course not, let's do something else. B: What? You
| put text in the maths and stuff comes out.
| MiroF wrote:
| That's just not even close to the only or most useful
| application.
|
| I use NLP and associates s2s techniques every day. I struggle
| to see how so many people don't see the obvious benefits deep
| inference is bringing to stuff all around them.
| p1esk wrote:
| _those still don 't rise to the level of having an actual
| human_
|
| What if they did? Do you see where the big deal is now?
| claudeganon wrote:
| This past week, I used this new edit distance library to
| identify quasi-duplicates in a large dataset:
|
| https://github.com/Bergvca/string_grouper
|
| Saved me hours of work because all the Levenshtein
| implementations are pretty slow and I'm going to need to
| rerun the analysis as the dataset grows.
|
| I don't know about consumer-facing tools, but NLP stuff has
| helped me solve all kinds of tedious data problems at work.
| thekyle wrote:
| If a chatbot was as good as a human then would you notice it
| was a chatbot?
| narag wrote:
| Has it happened that a "thought experiment" has become a real
| experiment ever?
| exo-pla-net wrote:
| Not sure exactly what you mean, but Einstein developed
| relativity largely through thought experiments. And relativity
| has been verified by real experiments.
| dane-pgp wrote:
| Most historians think that this was actually a thought
| experiment:
|
| https://en.wikipedia.org/wiki/Galileo's_Leaning_Tower_of_Pis...
|
| An equivalent experiment was famously carried out for real in
| 1971 on the surface of the Moon.
| simonh wrote:
| In a sense every experiment starts as a thought experiment.
| FiberBundle wrote:
| I found the science exams results interesting and skimmed the
| paper [1]. They report an accuracy of >90% on the questions. What
| I found puzzling was that they have a section in the experimental
| results part where they test the robustness of the results using
| adverserial answer options, more specifically they used some
| simple heuristic to choose 4 additional answer options from the
| set of other questions which maximized 'confusion' for the model.
| This resulted in a drop of more than 40 percentage points in the
| accuracy of the model. I find this extremely puzzling, what do
| these models actually learn? Clearly they don't actually learn
| any scientific principles.
|
| [1] https://arxiv.org/pdf/1909.01958.pdf
| teej wrote:
| That's the thing. Machine Learning is a misnomer, the models
| don't "learn" anything about the domain they operate in. It's
| just statistical inference.
|
| A dog can learn to turn left or to turn right for treats. But
| they don't understand the concept of "direction", their brain
| isn't wired that way.
|
| Machine learning models perform tricks for treats. The tricks
| they do get more impressive by the day. But don't be deceived,
| they aren't wired to gain knowledge.
| 0-_-0 wrote:
| Doesn't your brain derive knowledge through statistical
| inference? How else? I can imagine the AGI of the future
| (having perfected knowledge gathering through statistical
| inference) some day saying: "Humans perform tricks for
| treats. The tricks they do are impressive. But don't be
| deceived, they aren't wired to gain knowledge."
| teej wrote:
| I didn't claim that the way brains work didn't include
| statistical inference. That's why the dog metaphor works.
| Both the dog and the machine learning model use statistical
| inference to perform tricks.
|
| Dogs haven't been to the moon, however. There's more to the
| brain we don't understand.
| plutonorm wrote:
| My god if I hear this argument one more time I'm going to
| pop. What on earth gives you the idea that you are something
| more than statistical inference?
| teej wrote:
| Yes, it's widely believed that statistical inference is a
| part of how the brain operates. But we have barely
| scratched the surface in our understanding how the human
| brain works.
|
| Do you honestly believe statistical inference completely
| explains a human's ability to learn?
| wrs wrote:
| I would be interested in hearing the results from _humans_
| presented with adversarial answer options. You may say that a
| machine learning correlations between words isn't really
| learning science, but I wonder how many human students aren't
| either, just pretty much learning correlations between words to
| pass tests...
| FiberBundle wrote:
| They do give an example of a question, in which the model
| chose an incorrect answer in the adversarial setting:
|
| "The condition of the air outdoors at a certain time ofday is
| known as (A) friction (B) light (C) force
| (D)weather[correct](Q) joule (R) gradient[selected](S)trench
| (T) add heat"
|
| I assume this might be characteristic for other questions as
| well, although I don't know anything about the Regents
| Science Exam and whether there are multiple questions about
| closely related topics.
| jacobwilliamroy wrote:
| Most multiple choice math problems can be completely
| circumvented by simply finding the digital root of the
| expression in the problem. I was surprised to find this to be
| true, even on college entrance exams.
| flir wrote:
| Seriously? Can you explain the mechanism? 'cos that sounds
| like - well, numerology, to be honest.
| 0-_-0 wrote:
| Adversiarial training means that they specifically search for
| answers that the network would misunderstand. If this only
| leads to a 40 percent loss (the network still answers correctly
| 50% of the time) I still consider that remarkable.
|
| Choosing the best from 8 answers where some of them were
| adversarially derived should be equivalent to choosing the best
| from all possible answers, of which there could be tens of
| thousands. How would a human do in that situation?
|
| Although the kinds of mistakes the network makes seem like a
| mistake you would never do (i.e. you wouldn't call the
| condition of air outdoors _gradient_ ), the opposite could also
| be true, that it would easily answer questions you would have a
| problem with.
| _emacsomancer_ wrote:
| A bit I found rather strange, on the language-side:
|
| > This is to say the patterns in language use mirror the patterns
| of how things are(1).
|
| > (1)- Strictly of course only the patterns in true sentences
| mirror, or are isomorphic to, the arrangement of the world, but
| most sentences people utter are at least approximately true.
|
| Presumably this should really say something like "...but most
| sentences people utter are at least approximately true _of their
| mental representation of the world_. "
| rvense wrote:
| I think the point about language being a model of reality was
| interesting. I have an MA in linguistics including some NLP from
| about a decade ago and was looking at a career in academic NLP. I
| ultimately left to become a programmer because (of life
| circumstances and the fact that) I didn't see much of a future
| for the field, precisely because it was ignoring the (to me)
| obvious issues of written language bias, ignorance of multi-
| modality and situatedness etc. that are brought up in this post.
|
| All of these results are very interesting, but I'm not really
| feeling like we've been proved wrong yet. There is a big question
| of scalability here, at least as far as the goal of AGI goes,
| which the author also admits:
|
| > Of course everyday language stands in a woolier relation to
| sheep, pine cones, desire and quarks than the formal language of
| chess moves stands in relation to chess moves, and the patterns
| are far more complex. Modality, uncertainty, vagueness and other
| complexities enter but the isomorphism between world and language
| is there, even if inexact.
|
| This woolly relation between language and reality is well-known.
| It has been studied in various ways in linguistics and the
| philosophy of language, for instance by Frege and not least
| Foucault and everything after. I also think many modern
| linguistic schools take a very different view of "uncertainty and
| vagueness" than I sense in the author here, but they are
| obviously writing for non-specialist audience and trying not to
| dwell on this subject here.
|
| My point is, when making and evaluating these NLP methods and the
| tools they are used to construct, it is extremely important to
| understand that language models social realities rather than any
| single physical one. It seems to me all too easy, coming from
| formal grammar or pure stats or computer science, to rush into
| these things with naive assumptions about what words are or how
| they mean things to people. I dread to think what will happen if
| we base our future society on tools made in that way.
| skybrian wrote:
| These tools seem to be getting pretty good at fiction. In
| particular, playing around with AI Dungeon, it doesn't believe
| anything, or alternately you could say it believes everything
| at once. It's similar to a library that contains books by
| different authors, some fact, some fiction. Contradictions
| don't matter. Only local consistency matters, and even then,
| not that much.
|
| Unfortunately, many people want to believe that they are being
| understood. But, on the bright side, this stuff seems
| artistically useful? Entertainment is a big business.
| rvense wrote:
| > many people want to believe that they are being understood
|
| Not only that, they also want to understand. I read some
| computer poetry a few years ago that _worked_. The computer
| did not intend any meaning, but I found it anyway.
|
| And of course I did. This is how language works. We assume
| that people (and by extension, the texts they write) follow a
| set of rules which structures how we engage with them in a
| successful manner. Paul Grice formulated this as the
| cooperative principle in his study of everyday language, but
| I believe it is the exact thing that is at play when people
| meet NLP:
|
| https://en.wikipedia.org/wiki/Cooperative_principle
| joe_the_user wrote:
| _My point is, when making and evaluating these NLP methods and
| the tools they are used to construct, it is extremely important
| to understand that language models social realities rather than
| any single physical one._
|
| I'd claim actual involves language both a general shared
| reality and quite concrete and specific discussions of single
| physical and logic facts/models. Some portion of language
| certainl looks mostly like a stream of associations. But within
| it is also are references to physical reality and world model
| and the two are complexly intertwined (making logical sense is
| akin to a "parity check" - you can for a while without it but
| then you have to look at the whole to get it). I believe one
| can see this in a GPT paragraph, where the first two sentences
| seem intelligence, well written but the third sentence
| contradicts the first two sentences sufficiently one's mind
| isn't sure "what's being said" (and even here, our "need for
| logic" might be loose enough that we only notice the
| "senselessness" after the third logical error).
| rvense wrote:
| In human language I think physical reality is always a few
| layers out. Language is social, first and foremost, and
| naming something is not neutral. We can hardly refer to
| single objects directly, we mostly do it through their class
| membership, which always, always include a whole range of
| associations, reductions, metaphor, etc. that are cultural.
| joe_the_user wrote:
| _In human language I think physical reality is always a few
| layers out._
|
| Yes, the point is you can neglect physical (and logic)
| reality for a while in a stream of sentence. But not
| forever and that's current NLP's output has it's limits.
| Just a simple level, a stream of glittering adjectives can
| be added to a thing and just add up to "desirable" unless
| those adjectives "go over threshold" and contradict each
| other and then the description can get tagged by the brain
| as a bit senseless.
| bloaf wrote:
| I know that there are allegedly NLP algorithms for generating
| things like articles about sports games. I assume they have
| something more like the type signature (timeline of events) ->
| (narrative about said events)
|
| What this article is about is more (question/prompt) ->
| (answer/continuation of prompt)
|
| Does anyone know if there is progress in the (timeline of events)
| -> (narrative about said events) space?
| YeGoblynQueenne wrote:
| >> The Winograd schema test was originally intended to be a more
| rigorous replacement for the Turing test, because it seems to
| require deep knowledge of how things fit together in the world,
| and the ability to reason about that knowledge in a linguistic
| context. Recent advances in NLP have allowed computers to achieve
| near human scores:(https://gluebenchmark.com/leaderboard/).
|
| The "Winograd schema" in Glue/SuperGlue refers to the Winograd-
| NLI benchmark which is simplified with respect to the original
| Winograd Schema Challenge [1], on which the state-of-the-art
| still significantly lags human performance:
|
| _The Winograd Schema Challenge is a dataset for common sense
| reasoning. It employs Winograd Schema questions that require the
| resolution of anaphora: the system must identify the antecedent
| of an ambiguous pronoun in a statement. Models are evaluated
| based on accuracy._
|
| _WNLI is a relaxation of the Winograd Schema Challenge proposed
| as part of the GLUE benchmark and a conversion to the natural
| language inference (NLI) format. The task is to predict if the
| sentence with the pronoun substituted is entailed by the original
| sentence. While the training set is balanced between two classes
| (entailment and not entailment), the test set is imbalanced
| between them (35% entailment, 65% not entailment). The majority
| baseline is thus 65%, while for the Winograd Schema Challenge it
| is 50% (Liu et al., 2017). The latter is more challenging._
|
| https://nlpprogress.com/english/common_sense.html
|
| There is also a more recent adversarial version of the Winograd
| Schema Challenge called Winogrande. I can't say I'm on top of the
| various results and so I don't know the state of the art, but
| it's not yet "near human", not without caveats (for example,
| wikipedia reports 70% accuracy on 70 problems manually selected
| from the originoal WSC).
|
| __________
|
| [1] https://www.aaai.org/ocs/index.php/KR/KR12/paper/view/4492
| jvanderbot wrote:
| I'd go one step further: Humans themselves don't understand
| anything, we are just good at constructing logical-sounding
| (plausible, testable) stories about things. These are mental
| models, and it's the only way we can make reasonable predictions
| to within error tolerances of our day-to-day experience, but they
| are flat-out lies and stories we tell ourselves not based on a
| high-fidelity understanding of anything.
|
| Rumination, deep thinking, etc is simply actor-critic learning of
| these mental models for story-telling.
| runT1ME wrote:
| Do current NLP systems understand arithmetic, and can the do it
| with unfamiliar numbers they've never seen on? If not, I'd
| think that your theory is demonstratably false, as a child can
| extrapolate mathematical axioms from just a few example
| problems, whereas NLP models are not able to do so.
| glenstein wrote:
| >Do current NLP systems understand arithmetic, and can the do
| it with unfamiliar numbers they've never seen on?
|
| I don't know if that question is rhetorical or not, but GPT-3
| can do basic math for problems it has not been directly
| trained on, and there's been a fair amount of debate,
| including right here at hn, about what the takeaway is
| supposed to be.
| YeGoblynQueenne wrote:
| GPT-3 can't do arithmetic very well at all. There is a big,
| fat, extraordinary claim that it can in the GPT-3 paper but
| it's only based on perfect accuracy on two-digit addition
| and subtraction, ~90% accuracy on three digit addition and
| subtraction and ... around 20% accuracy on addition and
| substraction between from three to five digits and
| multiplication from between two measly digits. Note: no
| division at all and no arithmetic with more than five
| digits. And very poor testing to ensure that the solved
| problems don't just happen to be in the model's training
| dataset to begin with, which is the simplest explanation of
| the reported results given that the arithmetic problems
| GPT-3 solves correctly are the ones that are most likely to
| be found in a coprus of natural language (i.e. two- and
| three- digit addition and subtraction).
|
| tl;dr, GPT-3 can't do basic math for problems it has not
| been directly trained on.
|
| ____________
|
| [1] https://arxiv.org/abs/2005.14165
|
| See section 3.9.1 and Figure 3.10. There is an additional
| category of problems of combinations of addition,
| subtraction and multiplication between three single-digit
| numbers. Performance is poor.
| avmich wrote:
| >but it's only based on perfect accuracy on two-digit
| addition and subtraction, ~90% accuracy on three digit
| addition and subtraction and ...
|
| Which grade of school it corresponds to? It would be
| astonishing to have an AI system of this capability level
| mere few years ago.
| YeGoblynQueenne wrote:
| Performing arithmetic per se is not impressive. A
| calculator can do it and the rules of arithmetic are not
| so complex that they can't be hand-coded, as they are,
| routinely. The extraordinary claim in the GPT-3 paper is
| that a language model is capable of performing arithmetic
| operations, rather than simply memorising their results
| [1]. Language models compute the probabilities of a token
| to follow from a sequence of tokens and in particular
| have no known ability to perform any arithmetic, so if
| GPT-3, which is a language model, were capable of doing
| something that it is not designed to do, then that would
| be very interesting indeed. Unfortunately, such an
| extraordinary claim is backed up with very poor evidence,
| and so amounts to nothing more than invoking magick.
|
| __________
|
| [1] From the paper linked above:
|
| _In addition, inspection of incorrect answers reveals
| that the model often makes mistakes such as not carrying
| a "1", suggesting it is actually attempting to perform
| the relevant computation rather than memorizing a table._
|
| Now that I re-read this I'm struck by the extent to which
| the authors are willing to pull their results this way
| and that to force their preferred interpretation on them.
| Their model answers two-digit addition problems
| correctly? It's learned addition! Their model is making
| mistakes? It's because it's actually trying to compute
| the answer and failing! The much simpler explanation,
| that their model has memorised solutions to a few
| problems but there are many more it hasn't even seen,
| seems to be assigned a very, very low prior. Who cares
| about such a boring interpration? Language models are few
| shot learners! (once trained with a few billion examples
| that is).
| gwern wrote:
| How ironic you claim that the paper overstates it, when
| you very carefully leave out every single qualifier about
| BPEs and how GPT-3's arithmetic improves massively when
| numbers are reformatted to avoid BPE problems. Pot,
| kettle.
| ximeng wrote:
| "Byte pair encoding": more discussion at
| https://www.gwern.net/GPT-3#bpes
| Davidzheng wrote:
| I question the premise here. Can young children (say <5) who
| don't know arithmetic at all learn the axioms from just a
| bunch of examples? This isn't how children are taught
| arithmetic; they are taught the rules/algorithm not just a
| bunch of data-points. Do you have a source on the claim?
| joe_the_user wrote:
| The funny thing with GPT-3 is that it comes up with sentences
| that seem to make correct deductions using arithmetic or
| other submodels of reality but it will blithely generate
| further sentences that contradict these apparent
| understandings. It's impressive and incoherent all at the
| same time.
| cscurmudgeon wrote:
| But you claim to understand right in this comment how humans
| understand other things.
|
| Isn't that self-contradictory?
| p1esk wrote:
| He's just not very good at constructing logical-sounding
| (plausible, testable) stories about things :)
| xdavidliu wrote:
| no he did not claim so. His claim is just as much a lie as
| all other claims. Lies can be valuable, even if they are not
| strictly true.
| soulofmischief wrote:
| Mental models are not lies.
|
| "The car is red" is not a lie just because I didn't phrase it
| internally as "The car reflects photons of a frequency of
| around 700nm".
|
| We have to be able to simplify and internalize simplified
| models in order to make any sense of anything. It's the same
| reason your eyes only have a focal point in the dead center:
| attention requires vast amounts of processing power.
|
| To reiterate, a simplification is not a lie. Especially not a
| flat-out lie.
| Turing_Machine wrote:
| Yes. Bridges are designed using old-school Newtonian
| mechanics, without (in general) considering relativistic or
| quantum effects at all, but they stand up nonetheless.
|
| That wouldn't be the case if they were built based on "flat-
| out lies".
|
| Edit: the sticking point here is the use of the word "lie", I
| think.
|
| A lie is a falsehood told with intent to deceive.
|
| Thus, novels are not "lies", despite being works of the
| imagination. Everyone knows that the people in Moby-Dick or
| Wuthering Heights were made up.
|
| Neither is a simplified model a "lie" if it is close enough.
|
| All engineers know that the steel and concrete components of
| a bridge have quantum and relativistic effects going on
| inside them, so ignoring those effects isn't a "lie" in any
| meaningful sense.
|
| It just doesn't matter for the purpose at hand.
| bananaface wrote:
| I get what you're saying, but I think this is a side-effect of
| saving thinking in abstractions. We use abstractions to plug
| holes to avoid having to drill down on every concept we're
| exposed to.
|
| That can look like a surface-level linguistic understanding,
| but it's not, it's a surface-level abstraction. It's not
| arbitrary, it has structure, and when you flesh it out you're
| fleshing it out with actual abstract structure, _not_ just
| painting over the gaps with arbitrary language.
| skybrian wrote:
| Darn, based on the title, I was hoping for an overview of recent
| research.
|
| Lots of people are having fun playing with GPT-3 or AI Dungeon,
| myself included, but it seems like there is other interesting
| research going on like the REALM paper [1], [2]. What should I be
| reading? Why aren't people talking about REALM more? I'm no
| expert, but it seems like keeping the knowledge base outside the
| language model has a lot going for it?
|
| [1] https://ai.googleblog.com/2020/08/realm-integrating-
| retrieva... [2] https://arxiv.org/abs/2002.08909
| plutonorm wrote:
| I think the standard responses you see about gpt-3 -
|
| Hur durr, can't have intelligence from statistics. Hurt dur,
| Chinese room. Hur durr, doesn't understand semantics, it's just
| dumb. Ect ect ect ad nausea.
|
| They, go to show not how dumb gpt-3 is but how unthinking most
| people are. Just pulling from a bag of generally agreed upon
| notions, matching them up in simplistic ways and regurgitating.
___________________________________________________________________
(page generated 2020-08-16 23:00 UTC)