[Lingtyp] A brief summary of the discussion on bipolar polysemy
Daniel Ross
djross3 at gmail.com
Sat Jun 9 08:30:09 UTC 2018
Dear Ian,
Just a brief reply after reading through all of the earlier replies myself,
and now thinking about how it would apply to the syntactic construction in
question.
A close analogy may be found in Claude Hagège's reply regarding Classical
Arabic (and other languages).
The following paragraph from that message is most relevant:
"We often find also, rather than *enantiosemy*, the including of two
meanings under one global heading. Languages are able to subsume multiples
under extensible classifications, whose very vagueness allows them to
better capture reality while contributing to the dynamic of vocabularies.
Classical Arabic is known to contain a number of words that express a
relation, however asymmetrical, betwwen two terms : thus, باع bāˁa once
had the meaning « buy » as well as « sell », thus designating, actually,
the operation of exchange, without expressing its asymmetry."
Several others mentioned similar ideas, I believe, such that we shouldn't
think of a verb that can mean "buy or sell" as meaning opposites, but
rather having a general meaning along the lines of "engage in monetary
exchange".
The same explanation should similarly apply to the so-called Caused Motion
Construction. In fact, as you argue in your paper, this would mean that the
Construction does not inherently relate to motion per se. Instead, it
relates to something more general. It is something like the "Caused
Spatiotemporal Status Construction". (Or possibly even looser, maybe in the
right contexts no "cause" at all but some sort of cooperation or general
interaction. It depends on how far we want to stretch the 'Construction'
semantically and where we'd draw the line with particular deviating
examples.)
So was Goldberg "wrong" to call it a "Motion" construction? Well, no,
that's just a label, referring to a dominant interpretation. And "Motion"
is better than "Spatiotemporal Status" as a label. (Note in this case that
the relationship is metonymic, such that motion per se is not obligatorily
expressed but the encompassing category of spatiotemporal status is.) This
does suggest we should be careful about interpreting what constructions
"mean" beyond their labels.
Indeed, there is probably a very abstract but important discussion to be
had along the lines of the earlier discussion of word meaning, related to
how constructions can mean things, and whether they actually have core
meanings, whether those depend on context, and so forth! Actually for
someone proposing a theory of grammar in which everything is a construction
defined by "meaning", this is not a trivial topic at all. (I have
personally wondered how to delimit constructions by meaning and whether
they are necessarily categorical: how do we really know where one
construction begins and another ends? Is it really feasible to enumerate
constructions, or for example write a paper about "a construction"?
Something along the lines of nested constructions with broader to narrower
meanings may lead to a solution, but I'll stop here now.)
Daniel Ross
PhD Candidate in Linguistics
University of Illinois at Urbana-Champaign
On Sat, Jun 9, 2018 at 1:04 AM, Joo Ian <ian.joo at outlook.com> wrote:
> Dear all,
>
>
>
> I would like to express my sincere gratitude to everyone for answering my
> brief, rather simplistic, question on “bipolar polysemy” with such a
> fruitful discussion that truly provided me detailed insights on lexical
> polysemy and ambiguity. Since I am quite busy at the moment, I will just
> briefly summarize the discussion I have followed to the extent my limited
> time and knowledge allows me to.
>
>
>
> First, many have pointed out the cases of auto-antonyms
> <https://en.wikipedia.org/wiki/Auto-antonym>, which are indeed abundant
> across many languages. For example, a word can mean either ‘host’ or
> ‘guest’. Auto-antonyms are very interesting, but they do not necessarily
> fall into the category of bipolar polysemy, because *antonyms* are
> different from *negations*: ‘host’ is an antonym of ‘guest,’ but is not
> exactly the negation of ‘guest’ (‘non-guest’).
>
>
>
> There was a suggestion that English *let* may be a true case of bipolar
> polysemy, as it can mean ‘to allow’ and ‘to prevent’ at the same time. I
> checked this information on Wiktionary
> <https://en.wiktionary.org/wiki/let#English>, and it seems that the two
> meanings have different origins, the former being *leten* and the latter
> *letten*. So this may be a case of homonymy rather than polysemy.
>
>
>
> Next, there are cases where a word can convey bipolar meanings due to
> phonological reductions, such as French *T’inquiète* (`don’t worry’, lit.
> `you worry’), which is derived from the full form *Ne t’inquiète pas* ‘Do
> not worry’. Also, David pointed out the case of Malay *ta(h)u *‘to know’, which
> can mean ‘I don’t know’, but I this could be also a phonological reduction
> of ‘tak/gak tahu’ (don’t know).
>
>
>
> So there remains the question whether there exists a lexical item can
> fully convey a meaning and the negation of it in any context. I may have
> neglected if someone really did come up with such a case; in case my memory
> fails, then I would appreciate it if someone brought it up.
>
>
>
> The discussion was followed by the the philosophical question what a
> ‘meaning’ is. What does it really mean to say that ‘X means Y’? In what
> context? I did not follow this debate in detail, as my time was limited,
> but it did signal to me that my question would have been too simplistic in
> nature, and that I perhaps should have added more details to it to get a
> meaningful answer.
>
>
>
> Lastly, I would like to explain my motivation for asking this question in
> the first place. I read Goldberg (1995)’s paper on Construction Grammar,
> and its description of Caused Motion Construction on particular. This
> Construction, according to Goldberg, means “X causes Y to move Z”, as in
> (1):
>
>
>
> 1. She sneezed the napkin off the table.
>
>
>
> But in some cases of Caused Motion, the Y does not move anywhere, such as
> in (2):
>
>
>
> 1. He kept her at arm’s length.
>
>
>
> In (2), he does not cause her to move, quite the opposite: he causes her
> *not* to move (beyond arm’s length). Goldberg’s explanation for the case
> of (2) is that the Caused Motion Construction is polysemous and that it can
> mean: “X prevents Y from moving comp(Z)” where comp(Z) is the complementary
> opposite of Z.
>
>
>
> Since Goldberg’s claim was (to my understanding) that every linguistic
> unit is a construction, including morphemes, words, etc, it seemed
> unrealistic to me that a linguistic unit can both mean X and not-X. Since
> there is no morpheme that can mean X and not-X, why should there exist a
> Construction that can mean something like that?
>
>
>
> Please find attached a draft of a paper I wrote to tackle this problem. I
> left out the question of bipolar polysemy, because I discovered that it is
> a far more complex issue than I have imagined, but I believe that my
> argument still holds. I would appreciate any critical comments or
> suggestions on it.
>
>
>
> To conclude, thanks again to everyone who answered to my question, I
> really learned a lot from it. And I am sorry that I was not able to include
> every helpful comments in my brief summary.
>
>
>
> From Hong Kong,
>
> Ian Joo
>
> http://ianjoo.academia.edu
>
>
>
> *From: *m.m.jocelyne.fernandez-vest at vjf.cnrs.fr
> *Sent: *Wednesday, June 6, 2018 8:54 PM
> *To: *ebernard at filol.ucm.es; stela.manova at univie.ac.at
> *Cc: *LINGTYP at LISTSERV.LINGUISTLIST.ORG
> <lingtyp at listserv.linguistlist.org>
> *Subject: *Re: [Lingtyp] Does bipolar polysemy exist?
>
>
>
> Right yes: racism is universal, and sexism very frequent.
>
> Yet, after 40 years of career within Language science, I claim that open
> minded linguists also exist: you should try to collaborate with them and
> avoid the sectarist ones.
>
>
>
> Keep being proud of your name, language and origin: even in "Western"
> Europe, women are allowed to be smart.
>
> Good luck, dear Stela!
>
>
>
> M.M.Jocelyne FERNANDEZ-VEST
>
> CNRS & Université Sorbonne Nouvelle
>
>
> Envoyé de mon iPhone
>
>
> Le 6 juin 2018 à 11:41, ENRIQUE BERNARDEZ SANCHIS <ebernard at filol.ucm.es>
> a écrit :
>
> You're fully right, Stela. Racism is rampant. As sexism also is
>
> Enrique
>
>
>
> 2018-06-06 11:16 GMT+02:00 Stela Manova <stela.manova at univie.ac.at>:
>
> Dear all,
>
> In the past few days, in relation to the discussion on bipolar polysemy, I
> exchanged a number of messages with Mattis List on the relevance of
> Google’s research to linguistics. I thank Mattis for his reactions to my
> postings. I was honored: a male scholar born in Germany recipient of a
> highly prestigious ERC Staring Grant replied to me – a female scholar born
> in Bulgaria. I have lived in Vienna for almost 20 years. Yesterday, I was
> reminded one more time that I should be happy that I had the opportunity to
> study for a PhD degree in Austria but I cannot do science because as a
> female Bulgarian, I am supposed to be either a cleaning lady or a
> saleswoman. In other words, I am an Eastern European cockroach that tries
> to invade foreign territories. At least, I felt so the whole day yesterday,
> which postponed the writing of this message and also influenced its content.
>
> Mattis suggested that I prepare something on what we exchanged views here,
> namely Google and linguistics, so that the topic can be further discussed
> in blogs or other Internet platforms. I noticed (I use Google analytics)
> that after my postings on this list many linguists from all over the world
> visited my homepage. I understand it as an indication of interest in the
> topic. I could imagine preparing something on Google, math and linguistics
> but currently I have many problems in Vienna that do not allow me to do it.
> I am sorry for bothering you with this but in my opinion it is telling
> about what is going on in the European linguistics at the moment and these
> problems are not less important for the future of linguistics than Google’s
> research. I am specialized in Slavic and general linguistics, so I give
> examples from these fields. In the part of Europe where I reside, to be
> from Eastern Europe is a fault - even in the Slavic department of a
> university! In the Department of Slavic Studies in Vienna, people of Slavic
> origin are also divided into born in Austria and born in a Slavic country
> and the former group discriminates the latter, to show that they are
> superior. In the Department of Linguistics, the full professor, the natural
> morphologist Wolfgang U. Dressler, retired and was replaced with the formal
> semanticist Daniel Büring who immediately started a generative cleaning:
> only formal linguistics is linguistics, everything is syntax and
> syntax/semantics and all people who do research on other topics have to
> die. I have joint publications with two famous morphologists, Dressler and
> Aronoff, and I claim that there is morphology (not only distributed). If I
> could understand, to some extent, Büring’s incompatibility with Natural
> Morphology, could anybody explain to me why having published with Mark
> Aronoff who has a PhD from MIT makes me a bad linguist? Why are people who
> are not aware of basic principles of math very successful in formal
> grammar? What gives such people the right to impose their theoretical
> beliefs on others? Where does so much hate in linguistics come from?
>
> And now frankly, how many linguists will read an Internet text by a person
> with a Slavic name such as Stela Manova? It seems to me that a discussion
> on Google’s research and its relevance to linguistics, as suggested by
> Mattis, would make sense only if well-established linguists support it. I
> am afraid that I myself will also need some support in order to work on
> this; ideally, to distance from the linguistic absurd in Vienna, spending
> some time at an institution where I will not be treated as a cockroach. I
> am still doing linguistics, somehow; but I am tired of the senseless wars I
> have to war every day to survive in linguistics and as you could guess from
> my posts, I am now focused more on cognitive science and programming than
> on linguistics pure. As for my interest in mathematics and computers, I
> grew up in Bulgaria where at the age of 10, I was discovered as
> mathematically gifted and then received a solid education in mathematics.
> Yes, during the socialism, women were allowed to be smart.
>
> If someone is interested in working with me on what I addressed on this
> list or related issues, please feel free to contact me.
>
> Best,
>
> Stela
>
>
>
> On 04.06.2018, at 11:33, Mattis List <mattis.list at lingpy.org> wrote:
>
>
>
> Dear Stela,
>
> I think the points you brought up are very interesting, but it's
> probably time to stop the discussion at this point. What I would
> encourage you to do is now to write up your arguments in some form of a
> blog post (that you publish online on your preferred venue), as I think
> they are very interesting and important for a broader audience of
> linguists, and this is just a mailing list for typologists. If you do
> so, it would be very interesting, probably not only for me, but also for
> other colleagues from different fields, to jump on the train of this
> very interesting discussion and respond in blogs accordingly. It would
> also put the discussion on a more solid ground, as we could bring in
> quotes from colleagues and the like.
>
> Looking forward to read your arguments in much more detail. I'll try to
> do my best to answer them following all due scientific standards.
>
> All the best,
>
> Mattis
>
> On 2018-06-04 11:00, Stela Manova wrote:
>
> Dear Mattis,
>
> You tend to provide misleading information, this not only about the
> capacity of the human brain, s. Dmitry Nikoleav’s correction message
> below. You are also mistaken about the Google’s goal, which is neither
> “mimicking translation” nor "to get a machine running that replicates a
> human talking". Google’s goal is to have an effective search engine -
> the most effective one; everything else is more or less related to it.
> The Google search algorithm is highly relevant to linguistics - Google
> searches primarily texts. Yet for some reason, you never mention the
> Google search engine when you discuss Google’s research results.
>
> Google's AI beating the professional Go player: Yes, I know the story
> but your version is misleading again. This was done by DeepMind in
> London within their AlphaGo
> program: https://deepmind.com/research/alphago/. Google bought the
> start-up DeepMind - for £400 M! The Go player against the machine
> contest took place afterwards but it is a DeepMind story. Google did not
> pay so much money for a game-developing company. DeepMind was and is
> specialized in visual recognition. So, it is not about playing Go or any
> other computer game better than a human but about training a neural
> network to solve visual recognition tasks and it is good that the
> computer won. Computer vision assists us in many areas of life:
> medicine, healthcare, security and navigation, to mention just a few.
>
> Grammar / non-grammar and Google: I speak of non-grammar because
> Google's method does not have anything to do with linguistics. I refer
> to the method explained in the following
> video: https://blog.ycombinator.com/jeff-deans-lecture-for-yc-ai/. A
> neural network is trained on different types of data, including
> language. I do not see a connection to Chomsky or any other theoretical
> framework in linguistics.
>
> In linguistics, Baayen’s Naive Discriminative Learning is in line with
> Google’s research.
>
> As for linguistic research / fundamental linguistic questions and
> Google’s approach, I do not have Google's resources (human and funding),
> and therefore do morphology in linguistics. My research was not inspired
> by Google but by the Gauss-Jordan
> elimination: http://homepage.univie.ac.at/stela.manova/
> uploads/1/2/2/4/12243901/cognitiveapproachsuff1-suff2.pdf. I
> have investigated suffix combinations in a number of languages and it
> turned out that in all those languages suffix combinations are fixed,
> i.e. if a word has more than one derivational suffix, based on the first
> suffix, one can predict the following suffix because there is only one
> option for a following suffix. I then tested this finding
> psycholinguistically. Native speakers know which suffix combinations
> exist and which do not in their language and they do not need bases
> (roots / stems) to judge whether a suffix combination is a legitimate
> one. All linguistic theories derive morphological structure starting
> with a root (or a stem, depending on the theory). Yet, for some reason,
> native speakers that took part in the experiments, all without
> linguistic education, did not need roots and stems and could do
> something they were not supposed to be able to do. I think that we do
> not know enough about the role of memory in language processing. This is
> how I understand Google’s research. It seems to me that in language
> processing the human brain relies on structures (sequences) of various
> lengths and uses them as ready-made blocks as well as that it also uses
> pieces of structure that linguists do not recognize as linguistic units,
> i.e. that there are not only phonemes, morphemes, etc. in language
> processing but the human brain also operates with structures that are
> neither words no morphemes, neither phrases nor sentences, etc. In the
> case of my research, e.g., suffix combinations are structures between
> morphemes and words.
>
> Best,
> Stela
>
>
> On 03.06.2018, at 12:25, Mattis List <mattis.list at lingpy.org
> <mailto:mattis.list at lingpy.org <mattis.list at lingpy.org>>> wrote:
>
> Dear Stela,
>
> I still don't see how building a machine that is barely understood,
> mimicking translation from French to English or similar can be
> considered as being a scientific alternative to linguistic research. You
> may say, from the perspective of AI research it is scientific, in the
> sense that new knowledge is generated if the machine words better, but
> it does NOT answer linguistic questions, it could optimally provide a
> model of linguistic intuition among humans, but this does not answer
> fundamental questions that linguists ask themselves, and we are
> discussing linguistic implications here, not questions of engineering,
> like how to tweak your neural network most quickly and efficiently.
>
> Have you followed the debate about google's AI beating the Go players?
> The follow-up was: now, that the machine is better than humans, humans
> will have to study the moves by the machine to learn from it. So a
> machine was created that beats humans in a particular game, but the
> knowledge, what creates successful strategies when playing that game was
> NOT created. That IS the difference between engineers working on
> automatic translation and linguists trying to investigate certain
> properties of human languages.
>
> I have discussed this in an earlier blog post (longer time ago) that
> focuses on topics of historical linguistics, where machine learning
> techniques have an even harder time in providing useful solutions.
> There I also point to some literature reflecting that engineers and
> computer scientists are well aware of this problem (maybe not the people
> with google, but maybe even them: it is much easier to enhance a model
> if you understand why it fails):
>
> *
> http://phylonetworks.blogspot.com/2016/11/once-more-on-
> artificial-intelligence.html
>
> I also don't understand why you insist on the non-grammar part of google
> translate? They use sequence models, right? And what is a model that
> creates a sequence other than a grammar, albeit a rather simple one on
> Chomsky's hierarchy. Or am I getting something wrong, and there's a
> definition for grammar I am not aware of? Definitely possible, but I'd
> like to know which one you base your distinction on...
>
> Best,
>
> Mattis
>
>
> On 03.06.2018 11:31, Stela Manova wrote:
>
> Dear Mattis,
>
> You write:
>
>
> but there's a misunderstanding regarding scientific endeavor
> here: google people are engineers, their goal is to get a machine
> running that replicates a human talking. What linguists want to do is
> scientific endeavor, we do not only want to replicate a machine doing
> the same things that we do, but we want to UNDERSTAND what the
> machine does.
>
>
> What is scientific endeavor? Imagine that we have to describe
> scientifically, let me say, a brick. We can say: 1) a brick is building
> material; it is used for building houses; when building houses, we order
> bricks in a specific way to construct walls, etc.; 2) Alternative
> definition: a brick is a/ /parallelepiped with 90 degree angles; bricks
> can differ in form and size; a brick usually contains holes that can
> also differ in form and size; etc. Which definition provides a better
> understanding of what is a brick - the one that focuses on what a brick
> is good for or the one that cares about form, size, and holes?
>
> Like all recent research in AI, Google research is inspired by the
> organization of the human brain (thank you, Dmitry, for the addition).
> Is this scientific endeavor or not? I think it is.
>
> I cannot agree that Google engineers do not understand what a machine
> does. On the contrary, exactly because they understand it very well,
> they managed to optimize the Google Translate algorithm. Btw, what
> should make linguists nervous is not the fact that the algorithm without
> grammar performs faster than that with grammar (in my previous message I
> explained why; it is a matter of mathematics). What is really surprising
> is that the algorithm without grammar translates more precisely than the
> algorithm with grammar.
>
> I am aware how sensitive the grammar-non-grammar issue is for the
> linguistic community, but as Volker Gast wrote here: "At the end of the
> day, the various approaches to linguistics should be judged against the
> value of their results…”
>
> Best,
> Stela
>
>
>
> On 02.06.2018, at 21:11, Dmitry Nikolaev <dsnikolaev at gmail.com
> <mailto:dsnikolaev at gmail.com>> wrote:
>
> Dear Mattis,
>
> a small correction:
>
>
> Furthermore, it is not that trivial as the google-people suggest: they
> use extremely large training corpora for automatic translation which is
> based on stochastic (albeit apparently simple) grammars. A human,
> however, acquires a language with much LESS training material and a
> smaller brain. This questions cannot be solved if we rely on google or
> the engineering part of "computer science".
>
>
> The biggest announced neural networks seem to have on the order of 1
> to 2 hundred billion parameters (weights of connections between
> neurons). Human brain has ~100 billion neurons and on the order of 100
> trillion connections / learnable parameters. Huge NLP endeavours
> probably match and surpass the amount of input humans receive when
> acquiring a language, but computationally human brain is not small, it
> is in another universe.
>
> With kind regards,
> Dmitry
>
>
> On Sat, 2 Jun 2018 at 14:29, Mattis List <mattis.list at lingpy.org
> <mailto:mattis.list at lingpy.org>> wrote:
>
> Dear Stela,
>
> very brief, but there's a misunderstanding regarding scientific
> endeavor
> here: google people are engineers, their goal is to get a machine
> running that replicates a human talking. What linguists want to do is
> scientific endeavor, we do not only want to replicate a machine doing
> the same things that we do, but we want to UNDERSTAND what the
> machine does.
>
> This issue of machine learning approaches which are all very
> black-boxy,
> has now finally gained some intention among scholars, since it is
> also
> dangerous, if we want to use machines to replace human labor in the
> future (look at how badly facebook filters hate-speech). But it is
> also
> fundamentally different as an approach: we NEED to care about
> categories, as we want to look inside the box, not simply create a
> new one.
>
> Furthermore, it is not that trivial as the google-people suggest:
> they
> use extremely large training corpora for automatic translation
> which is
> based on stochastic (albeit apparently simple) grammars. A human,
> however, acquires a language with much LESS training material and a
> smaller brain. This questions cannot be solved if we rely on
> google or
> the engineering part of "computer science".
>
> Best,
>
> Mattis
>
>
>
> On 02.06.2018 11:17, Stela Manova wrote:
>
> Dear Randy,
>
> What you write simply shows that you do not know enough about
>
> numerical
>
> systems and how a computer works. Yes, there exist different
>
> numerical
>
> systems, btw not only the binary and the decimal one, but there are
> special notations for the different systems, so that
>
> mathematicians and
>
> computers know in which system a number is. Additionally, a computer
> works only in binary code. How exactly those things happen in
>
> computer
>
> science is explained, e.g.,
>
> here: http://www.cplusplus.com/doc/hex/.
>
>
> Regarding induction / deduction and Jeff Dean’s method, I will not
> philosophize, there is a clear definition of mathematical
>
> induction. In
>
> math, induction is used in recursive situations to establish the
>
> basic
>
> case. That MIT professor explains induction and recursion very
>
> well: https://www.youtube.com/watch?v=WPSeyjX1-4s&t=0s&list=
> PLUl4u3cNGP63WbdFxL8giv4yhgdMGaZNA&index=23.
>
> Let us leave readers decide of what type is Jeff Dean’s method.
>
> What linguists cannot understand is the fact that in order to apply
> mathematical logic, one needs elements that are of the same
>
> type. If you
>
> assume that there are different types of words (basic elements of a
> system), you cannot describe that system mathematically, at
>
> least not
>
> without preliminary sortings of the elements, which will make the
> analysis more time-consuming = slower computer program.
>
> Therefore, Jeff
>
> Dean claims that using grammar is less efficient than handling
>
> without
>
> grammar. In sum, the difference between the computer scientist
>
> Jeff Dean
>
> and a linguist: Jeff Dean treats all words as units (elements of the
> same type) while linguists philosophize on bipolar polysemy =
>
> Jeff Dean
>
> solves a problem, linguists create an additional one.
>
> Btw, if linguists listen to computer scientists, there would not
>
> be any
>
> research on complexity in linguistics, either. The above MIT
>
> professor
>
> again, part 1
>
> at: https://www.youtube.com/watch?v=o9nW0uBqvEo&list=
> PLUl4u3cNGP63WbdFxL8giv4yhgdMGaZNA&index=36, and
>
> part 2
>
> at: https://www.youtube.com/watch?v=7lQXYl_L28w&list=
> PLUl4u3cNGP63WbdFxL8giv4yhgdMGaZNA&index=37.
>
>
> Best,
> Stela
>
>
> On 02.06.2018, at 08:51, Randy J. LaPolla
>
> <randy.lapolla at gmail.com <mailto:randy.lapolla at gmail.com>
>
> <mailto:randy.lapolla at gmail.com
>
> <mailto:randy.lapolla at gmail.com>>> wrote:
>
>
> Dear Stella,
> The mathematical approach you discussed is very much in the
> Structuralist tradition, and not that much in line with the most
> cutting edge recent AI research. Almost all linguistics (including
> Chomsky), plus most computer science, particularly NLP, is based on
> Structuralist principles (though Interactional Linguistics,
> Usage-based approaches, and Halliday’s approach are not). What you
> said, "in mathematics / computer science, in isolation, a
>
> sequence of
>
> elements always has a single meaning because if it has not, no
> computation is possible”, and you assume it must be true for
>
> language,
>
> is very much the sort of thing I was talking about. Even in
>
> computer
>
> science that is not true, as “10” in a binary system such as
>
> machine
>
> code has a different “meaning” from “10” in a non-binary
>
> situation, so
>
> 1 + 1 = 2 is only true in the context of a non-binary code.
> Mathematics and logic is also tautologies, as Wittgenstein pointed
> out, so quite different from natural language, where even “War
>
> is war”
>
> is not a tautology, and that is why there was the whole Oxford
>
> School
>
> of Ordinary Language Philosophy (Grice, Austin, Searle, etc.),
>
> as they
>
> saw that natural language is quite different from the mathematical
> approach being pushed by the logical positivists and analytic
> philosophers. (Frege and Russel had turned logic into
>
> mathematics, and
>
> tried to apply it to language—the early Wittgenstein went along
>
> with
>
> that initially, but later saw how problematic even his own early
> approach was.)
>
> I am aware of what has been going on in AI, particularly by
>
> Jeff Dean,
>
> in the switch from symbolic (deductive/rule-based) AI to inductive
> approaches, and am quite happy they finally have seen the light in
> that regard, and that has made a big difference in terms of
>
> what the
>
> systems can do. That switch, from rule based deductive
>
> algorithms, is
>
> what Dean means by doing without grammar; what they find using the
> inductive approach is still grammar (as Peirce said “Induction
>
> infers
>
> a rule”), and simply based on symbol manipulation, so a long
>
> way from
>
> modelling actual communication, which is based on meaning, not
> symbols, and so what they are talking about is not really
> “understanding". Induction can only take you so far (Peirce’s
>
> view was
>
> that deduction (which is tautology) and induction do not tell you
> anything new—although abduction is the “weakest” inference, as
>
> he put
>
> it, it is the only one that tells you something new; On the
>
> difference
>
> between the latter two: “. . . the essence of an induction is
>
> that it
>
> infers from one set of facts another set of similar facts, whereas
> hypothesis [abduction—rjl] infers from facts of one kind to
>
> facts of
>
> another.”); the next step is to understand how communication
>
> actually
>
> works (as it isn’t coding/decoding) and try to see if it is
>
> possible
>
> to model abductive inference, which is what real communication is
> based on. I don’t know if that is possible. The problem is they are
> not working with linguists who understand communication, and so
>
> on the
>
> one had assume it is about symbol manipulation, and on the
>
> other end
>
> up often reinventing the wheel. One example is a talk I went to
>
> at our
>
> Complexity Institute, where the speaker talked about how his
>
> algorithm
>
> had shown that some words in English, such as “a little bit" occur
> together more often than others. We linguists of course knew that
> decades ago, but as this person had not talked to any linguists
>
> before
>
> starting a linguistic study, he had no clue about what had been
>
> done
>
> in terms of collocational relationships.
>
> Yes, the abilities and principles related to meaning creation and
> linguistic behaviour are general cognitive mechanisms and
>
> behavioural
>
> principles, not specific to language, and not unique to humans. You
> say, "Linguists believe that linguistics is a module of its own
>
> in the
>
> brain and love re-defining things as something specific for the
> field”, but that statement only applies to an ever-shrinking
>
> minority
>
> of people doing rationalist philosophy rather than empirical
> linguistics, and the ones associated with the now discredited
>
> symbolic
>
> AI.
>
> All the best,
> Randy
> -----
> *Randy J. LaPolla, PhD FAHA* （羅仁地）
> Professor of Linguistics and Chinese, School of Humanities
> Nanyang Technological University
> HSS-03-45, 14 Nanyang Drive | Singapore 637332
> http://randylapolla.net/
> Most recent book:
>
> https://www.routledge.com/The-Sino-Tibetan-Languages-
> 2nd-Edition/LaPolla-Thurgood/p/book/9781138783324
>
>
>
>
>
>
>
>
>
> On 1 Jun 2018, at 4:57 PM, Stela Manova
>
> <stela.manova at univie.ac.at <mailto:stela.manova at univie.ac.at>
>
> <mailto:stela.manova at univie.ac.at
>
> <mailto:stela.manova at univie.ac.at>>> wrote:
>
>
> Dear Randy,
>
> What I wrote does not have anything to do with structuralism
>
> but is
>
> based on recent research in language understanding on which we
>
> rely
>
> every day. I mean research carried out by Google. Intriguingly,
> people who do NLP and LU at Google are not linguists but computer
> scientists and the senior fellow of the Google Brain Team,
>
> Jeff Dean,
>
> claims that language understanding does not need grammar, see his
> slides on Scaling language understanding models
>
> at https://blog.ycombinator.com/jeff-deans-lecture-for-yc-ai/, starts
>
> at 24:54 in the video, as well as the slides on Google Translate -
> 27:52 in the video (the slides are below the video), but
>
> please watch
>
> the whole video if you have time. This is one of Jeff Dean’s many
> talks on Deep Learning, I give this link because I have it in my
> computer but you can google the topic and the speaker. So,
>
> Google’s
>
> LU does not use grammar but is based on combinations /
>
> sequences of
>
> elements and statistics; and ironically, linguists who believe in
> grammar and irony (based on your message below) use Google
>
> products
>
> every day. The wisdom from the Google sequence-to-sequence
>
> model is
>
> that single examples do not count as evidence for the
>
> organization of
>
> a system. Reread now our discussion on what is bipolar
>
> polysemy and
>
> you will understand, why so many linguistics professors from
>
> so many
>
> different countries cannot agree on a definition.
>
> It is not about bipolar polysemy, it is about the future of the
> field. Google guys claim and prove that the same learning logic
> applies to all areas of life; roughly, the same rules operate
> in visual perception, chemistry, language, etc. Linguists believe
> that linguistics is a module of its own in the brain and love
> re-defining things as something specific for the field - there is
> even statistics for linguists which unfortunately differs from
>
> Google
>
> statistics because people who do statistics in Google are
> mathematicians while (most of the) linguistic statisticians
>
> were bad
>
> at math at school and therefore studied languages at the
>
> university, etc.
>
>
> I have a PhD in general linguistics from the University of Vienna
> (and my both PhD supervisors were very bad at math) but I cannot
> agree that this is sufficient evidence that the tip of my nose
>
> is the
>
> end of the horizon. OK, I was also educated in math nine years -
> intensively.
>
> Best,
>
> Stela
>
>
>
> On 01.06.2018, at 06:06, Randy J. LaPolla
>
> <randy.lapolla at gmail.com <mailto:randy.lapolla at gmail.com>
>
> <mailto:randy.lapolla at gmail.com
>
> <mailto:randy.lapolla at gmail.com>>> wrote:
>
>
> Hi All,
> This whole discussion shows how problematic some of the a priori,
> non-empirical assumptions of the Structuralist approach are. The
> assumption that there is a fixed association of sign and
>
> signifier,
>
> and so words have meaning in some abstract universe divorced from
> context, and the assumption that language can be dealt with
> mathematically, and the assumption that communication happens
> through coding and decoding (on the computational model), and
>
> that
>
> the “real” word is the written, abstract, out-of-phonetic-context
> form, and so phonology in context can be ignored, and as there is
> only one “real” meaning to a word, the different uses in
>
> context ,
>
> such as irony, can be simply ignored or treated as deviant. The
> assumption that there is a fixed system that has iron-clad rules,
> and that there are aspects of the system that are necessary for
> communication to occur.
>
> There is much literature showing how problematic these
>
> assumptions
>
> are, but somehow they are still in force in much of
>
> linguistics, as
>
> reflected in some of this discussion.
>
> My own view is that communication involves one person
>
> performing a
>
> communicative act in a particular place and time and to a
>
> particular
>
> addressee, and the addressee abductively inferring that person’s
> reason for performing that act in that particular context to that
> particular person at that particular time. So it is completely
> context dependent, as Nick shows, and there is no minimum
> morphosyntactic structure required, as David Gil has shown.
>
> No part
>
> of the communicative situation or act can be left out in terms of
> understanding the meaning that the addressee creates in inferring
> the communicator’s intention (as Mark shows in including
>
> gesture in
>
> his discussion, though it also includes non-conventionalised
> behaviour, e.g. gaze and body movements; and it is creation of
> meaning, not transfer of meaning, and so subjective and
> non-determinative). Language and other conventionalised
> communicative behaviour (language is behaviour, not a thing, and
> does not differ in nature from other conventionalised behaviour)
> emerges out of the interaction of the people involved.
>
> So the question asked is like a Zen koan: you can’t answer it
>
> yes or
>
> no, as it is based on problematic assumptions.
>
> Randy
>
> -----
> *Randy J. LaPolla, PhD FAHA* （羅仁地）
> Professor of Linguistics and Chinese, School of Humanities
> Nanyang Technological University
> HSS-03-45, 14 Nanyang Drive | Singapore 637332
> http://randylapolla.net/
> Most recent book:
>
> https://www.routledge.com/The-Sino-Tibetan-Languages-
> 2nd-Edition/LaPolla-Thurgood/p/book/9781138783324
>
>
>
>
>
> On 1 Jun 2018, at 7:42 AM, Nick Enfield
>
> <nick.enfield at sydney.edu.au <mailto:nick.enfield at sydney.edu.au>
>
> <mailto:nick.enfield at sydney.edu.au
>
> <mailto:nick.enfield at sydney.edu.au>>> wrote:
>
>
> In Lao:
>
>
> 1. The verb cak2 means ‘know’, and can be negated as in
>
> man2 bòò1
>
> cak2 [3sg neg know] ‘S/he doesn’t know.’ But when used
>
> alone,
>
> with no subject expressed, often with the perfect marker
>
> (as in
>
> cak2 or cak2 lèèw4) it means “I don’t know.”
> 2. The verb faaw4 means ‘to hurry, rush’, and can be
>
> negated as in
>
> man2 bòò1 faaw4 [3sg neg rush] ‘S/he doesn’t hurry/isn’t
> hurrying.’ But when used alone as an imperative, with no
> subject expressed, often repeated, or with an appropriate
> sentence-final particle (as in faaw4 faaw4 or faaw4 dee4) it
> means “Don’t hurry, Stop hurrying, Slow down”.
> 3. Often, both positive and negative readings of verbs are
> available when the irrealis prefix si is used (with
>
> context or
>
> perhaps intonation doing the work); eg khaw3 si kin3
>
> [3pl irr
>
> eat] could mean ‘They will eat it’ or ‘They will
>
> definitely not
>
> eat it’ with a meaning similar to the colloquial English
> expression “As if they would eat it.” The second meaning is
> made more likely by insertion of the directional paj3 ‘go’
> before the verb (khaw3 si paj3 kin3 [3pl irr go eat] ‘As if
> they would eat it.’).
>
>
> Nick
>
> * *
> * *
> * *
> * *
> *N. J. ENFIELD *| FAHA FRSN | Professor of Linguistics
> Head, Post Truth Initiative https://posttruthinitiative.org/
> Director, SSSHARC (Sydney Social Sciences and Humanities
>
> Advanced
>
> Research Centre)
> Faculty of Arts and Social Sciences
> *THE UNIVERSITY OF SYDNEY*
> Rm N364, John Woolley Building A20 | NSW | 2006 | AUSTRALIA
> T +61 2 9351 2391 | M +61 476 239 669
> orcid.org/0000-0003-3891-6973
>
> <http://orcid.org/0000-0003-3891-6973>
> <http://orcid.org/0000-0003-3891-6973>
>
> E nick.enfield at sydney.edu.au <mailto:nick.enfield at sydney.edu.au>
> <mailto:nick.enfield at sydney.edu.au
>
> <mailto:nick.enfield at sydney.edu.au>> | W sydney.edu.au
> <http://sydney.edu.au/>
>
> <http://sydney.edu.au/> nickenfield.org
>
> <http://nickenfield.org/> <http://www.nickenfield.org/>
>
> * *
>
>
> *From: *Lingtyp <lingtyp-bounces at listserv.linguistlist.org
>
> <mailto:lingtyp-bounces at listserv.linguistlist.org>
>
> <mailto:lingtyp-bounces at listserv.linguistlist.org
>
> <mailto:lingtyp-bounces at listserv.linguistlist.org>>> on behalf of
>
> Mark Donohue <mark at donohue.cc <mailto:mark at donohue.cc>
>
> <mailto:mark at donohue.cc <mailto:mark at donohue.cc>>>
>
> *Date: *Friday, 1 June 2018 at 7:13 AM
> *To: *David Gil <gil at shh.mpg.de <mailto:gil at shh.mpg.de>
>
> <mailto:gil at shh.mpg.de <mailto:gil at shh.mpg.de>>>
>
> *Cc: *"LINGTYP at LISTSERV.LINGUISTLIST.ORG
>
> <mailto:LINGTYP at LISTSERV.LINGUISTLIST.ORG>
>
> <mailto:LINGTYP at LISTSERV.LINGUISTLIST.ORG
>
> <mailto:LINGTYP at LISTSERV.LINGUISTLIST.ORG>>"
>
> <lingtyp at listserv.linguistlist.org
>
> <mailto:lingtyp at listserv.linguistlist.org>
>
> <mailto:lingtyp at listserv.linguistlist.org
>
> <mailto:lingtyp at listserv.linguistlist.org>>>
>
> *Subject: *Re: [Lingtyp] Does bipolar polysemy exist?
>
> In Tukang Besi, an Austronesian language of Indonesia, the verb
> 'know' is dahani; verbs are generally prefixed to agree with the
> S,A argument, thus
>
> ku-dahani 'I know'
> 'u-dahani 'you know'
>
> etc.
> In some contexts (imperatives, emphatic generic (TAME-less)
> assertion), the prefix can be omitted.
>
> dahani 'I/you certainly know'
>
> Now, I've heard this (and only this) verb used, in the
>
> absence of
>
> any inflection, with exactly its opposite meaning
>
> Dahani 'I don't know'
>
> in what might be a sarcastic sense. Unlike the antonymic uses of
> many adjectives in many languages, including English, this
>
> use of
>
> dahani is actually a simple (though emphatic) negation of the
> verb's 'normal' meaning.
>
> -Mark
>
> On 1 June 2018 at 04:43, David Gil <gil at shh.mpg.de
>
> <mailto:gil at shh.mpg.de>
>
> <mailto:gil at shh.mpg.de <mailto:gil at shh.mpg.de>>> wrote:
>
> Yes, as Matti points out, negative lexicalization is not
>
> quite as
>
> rare as I was implying. Yet at the same time, I suspect
>
> that it
>
> might not be as common as Matti is suggesting. Looking at the
> examples that he cites in his Handbook chapter, I suspect
>
> that in
>
> some cases, the negative counterpart isn't "just" negative,
>
> but is
>
> also associated with some additional meaning components.
>
> Matti doesn't list "good"/"bad" as being such a pair, though,
> citing work by Ulrike Zeshan on sign languages, he does mention
> other evaluative concepts such as "not right", "not possible",
> "not enough". in English, at least, "bad" is not the
>
> negation of
>
> "good", it is the antonym of "good"; there's all kind of
>
> stuff in
>
> the world which we attach no evaluative content to, and which
> hence is neither good nor bad. (It's true that in English,
>
> in many
>
> contexts, the expression "not good" is understood as meaning
> "bad", which is interesting in and of itself, but still, it
>
> is not
>
> necessarily understood in this way.) While I have no direct
> evidence, I would strongly suspect that in languages that have
> lexicalized expressions for "not right", "not possible",
>
> and "not
>
> enough", the meanings of these expressions will be the
>
> antonyms of
>
> "right", "possible" and "enough", and not their negations.
>
> Under lexicalized negatives in the domain of tense/aspect,
>
> Matti
>
> lists "will not", "did not", "not finished". Well the one case
> that I am familiar with that falls into this category is
>
> that of
>
> the Malay/Indonesian iamative/perfect marker "sudah", which
>
> has a
>
> lexicalized negative counterpart "belum". However, "belum"
>
> isn't
>
> just "not sudah"; it also bears a strong (if not invariant)
> implicature that at some point in the future, the state or
> activity that is not complete will be completed — in fact, just
> like the English expression "not yet". (When people in
>
> Indonesia
>
> ask you if you're married, it's considered impolite to
>
> answer with
>
> a simple negation "tidak"; you're supposed to say "belum"
> precisely because of its implicature that you will, in the
>
> future,
>
> get married. By avoiding this implicature, the simple negation
> "tidak" is viewed as a threat to the natural order of
>
> things, in
>
> which everybody should get married.)
>
> I suspect that many if not all of the cases characterized
>
> by Matti
>
> ...
>
> [Message clipped]
> _______________________________________________
> Lingtyp mailing list
> Lingtyp at listserv.linguistlist.org
> http://listserv.linguistlist.org/mailman/listinfo/lingtyp
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.linguistlist.org/pipermail/lingtyp/attachments/20180609/bc62f1ef/attachment.htm>
More information about the Lingtyp
mailing list