From: IN%"lakoff@cogsci.Berkeley.EDU" 29-JAN-1994 00:15:54.53 To: IN%"twkwan@cuhk.hk" CC: Subj: Received: from decpxg.csc.cuhk.hk (decpxg) by vax.csc.cuhk.hk (PMDF #12160) id <01H88H22Q8SG8WWU60@vax.csc.cuhk.hk>; Sat, 29 Jan 1994 00:14 +0800 Received: by decpxg.csc.cuhk.hk id AA23648 (5.65c/IDA-1.4.4 for b071767@vax.csc.cuhk.hk); Sat, 29 Jan 1994 00:06:42 +0800 Received: from cogsci.Berkeley.EDU by decpxg.csc.cuhk.hk with SMTP id AA23607 (5.65c/IDA-1.4.4 for ); Sat, 29 Jan 1994 00:02:14 +0800 Received: by cogsci.Berkeley.EDU (5.63/1.29) id AA01923; Fri, 28 Jan 94 08:04:51 -0800 Date: Fri, 28 Jan 94 08:04:51 -0800 From: lakoff@cogsci.Berkeley.EDU (George Lakoff) To: twkwan@cuhk.hk Message-id: <9401281604.AA01923@cogsci.Berkeley.EDU> X-Ph: V3.14@decpxg.csc.cuhk.hk From cogling-relay@UCSD.EDU Thu Jan 13 21:03:42 1994 Received: from ucsd.edu by cogsci.Berkeley.EDU (5.63/1.29) id AA06664; Thu, 13 Jan 94 21:03:33 -0800 Received: from localhost by ucsd.edu; id UAA01979 sendmail 8.6.4/UCSD-2.2-sun Thu, 13 Jan 1994 20:58:04 -0800 for cogling-list Received: from cogsci.UCSD.EDU by ucsd.edu; id UAA01972 sendmail 8.6.4/UCSD-2.2-sun via SMTP Thu, 13 Jan 1994 20:58:02 -0800 for Received: by cogsci.UCSD.EDU (4.1/UCSDPSEUDO.4) id AA13194 for cogling@ucsd.edu; Thu, 13 Jan 94 20:58:01 PST Date: Thu, 13 Jan 94 20:58:01 PST From: robinson@cogsci.UCSD.EDU (Ed Robinson) Message-Id: <9401140458.AA13194@cogsci.UCSD.EDU> To: cogling@ucsd.edu Subject: Recent debate with questions Status: R The following brief debate appeared recently on a new brainstorming mailing list we set up in the cognitive science department at UCSD. The debate involved a number of questions which concern people on this list so I thought it might be nice to repost it and see what answers people on this net might offer. The original authors are mostly on this group. Hopefully, this will get our net off the ground. If there are any conferences, papers, or jobs that anyone would like to post, please feel free to do so at cogling@ucsd.edu. Are there any calls for reviews for books for _Cognitive Linguistics_? There are plenty of students in this group. Are there any general questions about vague concepts that people would like to clarify? Please post a question. Is anyone wondering if research has been conducted on topic x? If it has something to do with Cognitive Linguistics, someone here is likely to know. Thank you, Ed Robinson From: snyder@cogsci To: everyone@cogsci Subject: i'll start Ok, just to show I'm serious, I'll start with my own brainstorming ideas. Then hopefully we'll set up the list and move the discussion there. Specifically, after attending Prof. Zipser's WA, which as far as I could tell, essentially blasted neuroscience for not developing deeper, more substantive theory, I thought of discussions I've had with Jean and others re: image-schemas as possible templates for language and even thought. These thoughts are thus based on Jean's work, Gille's work, the Bates language class I took (part. Kemmer's lecture) and more. Specifically, my sense of things is the following. My intuition (you've got to start with something, so we might as well begin with what Norm Anderson calls my "at-oneness" with nature) is that human infants are building from sensory experience abstractions of that experience, and that these abstractions prepare the infant to learn language much more quickly. Why? Because language is itself built largely from these very abstractions! If language developed by people using sensory terms to describe whatever ... for example the expression "cool" or the sentence "That is hot research" then it makes sense to me that the same templates, or abstractions, or image-schemas, that infants build from their experience will be helpful to them in learning their language. Now this is not to say that the original connections between the sensory term and the idea it represents are still clear - they may have been fogged up by history ... most of us don't recognize, or think we recognize, the roots of words, for example. However, that information may be still partially present, and their may remain enough terms which are still quite close to the sense terms they're built from, to make life easier for the infant in terms of learning the language. For example, take the expression "built from". I can see an infant playing with blocks, building structures, and developing some abstraction of what it means to build. Then when the time comes to use the term "built from" I see the infant making use of that abstraction. Another example is "Did you get it?" to refer to "Did you understand this idea?" Infants are constantly getting and giving objects back and forth, back and forth. Maybe, just maybe, they're building some sort of template of that experience, which they make use of to underly the meaning of "Did you get it?" Now the part which really fascinates me is how could we build an image schema - an abstraction of sensory experience - in such a way as to make it useful for conveying a "higher order" meaning? Could we develop computational models to do this? And if these ideas are correct, ought there not to be some type of experiments we can run, with functional MRI, etc. that could potentially test the theory? And if the equipment is not yet good enough for that, ought it be become possible one day? For example, I wonder which parts of the brain are activated by the expression "Did you get it?" used in two different contexts: 1. As a referent to an object - "Did you get it?" meaning "Did you get the milk?" 2. As a referent to an idea - "Did you get it?" meaning "Did you understand this note" - not did you physically receive this note, but did you mentally understand it. Would these two expressions trigger similar brain regions? Where would the difference occur? why? Could it tell us something about how the brain may or may not be building these abstractions of sensory experience for use in understanding the expression in its proper context? Jonathan From cogling-relay@UCSD.EDU Thu Jan 13 21:09:24 1994 Received: from ucsd.edu by cogsci.Berkeley.EDU (5.63/1.29) id AA06702; Thu, 13 Jan 94 21:09:20 -0800 Received: from localhost by ucsd.edu; id VAA02457 sendmail 8.6.4/UCSD-2.2-sun Thu, 13 Jan 1994 21:04:37 -0800 for cogling-list Received: from cogsci.UCSD.EDU by ucsd.edu; id VAA02447 sendmail 8.6.4/UCSD-2.2-sun via SMTP Thu, 13 Jan 1994 21:04:35 -0800 for Received: by cogsci.UCSD.EDU (4.1/UCSDPSEUDO.4) id AA13344 for cogling@ucsd.edu; Thu, 13 Jan 94 21:04:34 PST Date: Thu, 13 Jan 94 21:04:34 PST From: robinson@cogsci.UCSD.EDU (Ed Robinson) Message-Id: <9401140504.AA13344@cogsci.UCSD.EDU> To: cogling@ucsd.edu Subject: debate cont. Status: R From: rgrush@sdcc3 (Rick Grush) Jonathan makes some potentially interesting points, but what would, I think, be of greatest interest as far as the contribution made to linguistic capacity via image schemata is not their contribution to the semantics of various terms (it strikes me as obvious that in at least a large number of cases the semantics of phrases and terms must be so influenced or structured), but rather the potential contribution to the structural or syntactic aspects of language. I read Lakoff at trying, with debateable results, to address this with his so-called invariance hypothesis, which claims that structural relations from the source domain, even those not crucial to the metaphorocal extension at hand, are mapped to the target domain, effectively constraining aspects of the usage of the extended family of terms. Paul Deane's book "Language in the Mind and Brain" is a bold attempts to specifically Address syntactic issues such as c-command, government, constraints on pronomical anaphora, etc., via extensions of structure from part-whole, source-path-goal, etc. schemata to the structure of phrases and sentences. But, though his book is interesting, I think his story is wrong (any opposing opinions?). Langacker makes wonderful headway in explaining 'syntactic' phenomena with, inter alia, networked shemata, but the schemata he employs are not typically image scemata. I guess my question is, has anybody really done any interesting work in showing how Lakoff-style image schemata can do good syntactic work? Rick Grush Philosophy From: Randy Gobbel > Langacker makes wonderful headway in explaining 'syntactic' > phenomena with, inter alia, networked shemata, but the > schemata he employs are not typically image scemata. > I guess my question is, has anybody really done any > interesting work in showing how Lakoff-style image > schemata can do good syntactic work? Yes, Langacker has. Where did you get the idea that the schemas he refers to are not image schemas? I'm certain that he would say they are. -Randy >From coulson@cogsci Thu Jan 6 13:37:35 1994 i don't think langacker would like his scemata id'd w/image-schemas qua mark johnson but more w/ chuck fillmore's frames. maybe he's on this list... From: Randy Gobbel > i don't think langacker would like his scemata id'd w/image-schemas qua > mark johnson but more w/ chuck fillmore's frames. > maybe he's on this list... Yeah, you're probably right. Of course, since nobody really know what an image schema IS, anyway.... -Randy From cogling-relay@UCSD.EDU Thu Jan 13 21:12:42 1994 Received: from ucsd.edu by cogsci.Berkeley.EDU (5.63/1.29) id AA06741; Thu, 13 Jan 94 21:12:39 -0800 Received: from localhost by ucsd.edu; id VAA02741 sendmail 8.6.4/UCSD-2.2-sun Thu, 13 Jan 1994 21:07:32 -0800 for cogling-list Received: from cogsci.UCSD.EDU by ucsd.edu; id VAA02738 sendmail 8.6.4/UCSD-2.2-sun via SMTP Thu, 13 Jan 1994 21:07:31 -0800 for Received: by cogsci.UCSD.EDU (4.1/UCSDPSEUDO.4) id AA13403 for cogling@ucsd.edu; Thu, 13 Jan 94 21:07:30 PST Date: Thu, 13 Jan 94 21:07:30 PST From: robinson@cogsci.UCSD.EDU (Ed Robinson) Message-Id: <9401140507.AA13403@cogsci.UCSD.EDU> To: cogling@ucsd.edu Subject: debate end Status: R >From faucon@cogsci Thu Jan 6 15:50:01 1994 The terminology 'image schema' is slightly unfortunate, because it evokes the notion 'image' more directly than it perhaps should. These schemas are clearly quite abstract, and have frame structure, as Seana points out (for Langacker's). In the cognitive linguistics approach, one key function of syntax is to map general frames/schemas onto more specific ones. The grammar gives important clues, but never total information as to what mappings are to be carried out. Syntax does many other things as well, such as setting up discourse and point of view organization, determining construals, and more generally guiding (without specifying) conceptual operations that must be carried out (e.g. XYZ metaphors or analogical counterfactuals). So the goal (an ancient and venerable one) is to link form and meaning, and almost as a side product we get explanations (Rick Grush's comment) for syntactic phenomena which had been approached on an autonomous formal basis. Interestingly, when source/target mappings are involved (Lakoff&Brugman), the syntactic form may be determined partly by aspects of the source, and partly by aspects of the target. Snyder (Jonathan)'s problem is a variant of Plato's problem: how can abstractions be built at all, whether from sensory experience, or from archetypes (such as block-building, or 'look before you leap' sort of things)? Do schemas and generic frames have to be innately available for this to happen in the first place? If so, what's the evolutionary story. If not, schemas will be emergent - what's a good model? Since the publication of the first work on invariance (of which Mark Turner was an originator), a range of other factors have come to light which enrich, but complicate the story. Turner and i are doing some new work showing that the standard two-space mappings of schemas from source to target are fundamentally insufficient, and we come up with a four space model that seems to apply nicely to superficially different kinds of conceptualization - non-verbal, mathematical, metaphorical, counterfactual, etc. [a short version "Conceptual Projection and Middle Spaces" should be available electronically in 3 or 4 weeks]. gilles From: snyder@cogsci Just for people's info, Arturo Hernandez sent me the following note > Eleanor > Saffran (who was a CRL visitor last quarter) and colleagues will have > a paper in the CRL newsletter on something very relevant to your > thought. It should be out in February. Jonathan From: Mike Cole Gile writes: The terminology 'image schema' is slightly unfortunate, because it evokes the notion 'image' more directly than it perhaps should. ---------- Gilles-- I am willing to make the argument that ALL IMAGES ARE INTERACTIONS. I know the 'flattened' interpretation you are indexing, but is that the conception of image that cognitive scientists use? A snapshot conception of images, without time extent? mike that cognitive scientists have in mind? A "snapshot?" mike From: Randy Gobbel > Gilles-- I am willing to make the argument that ALL IMAGES ARE > INTERACTIONS. I know the 'flattened' interpretation you are > indexing, but is that the conception of image that > cognitive scientists use? A snapshot conception of images, > without time extent? > mike I'm not sure what Langacker or Lakoff would say about that--I think Ron would say that at least *some* images do indeed have time extent. He makes this pretty clear in his discussions of verbs. In my attempt to build a working parser based on Ron's theory, I found it necessary (or at least it certainly simplified things) to have all images be more like movies than snapshots--i.e., all had time extent. "Snapshot" images were just degraded instances of this. -Randy Whoa....... That's a lot of stuff all at once to discuss. Why don't we air all of this over the cogsci network, starting small. These are issues discussed at Berkeley all the time face to face and I am aghast at the gap between the Berkeley and UCSD groups. By the way, the l-zero group at icsi ought tobe in on this. Could you add ``l-zero@icsi.berkeley.edu'' to the subscription list? Some responses: 1. Terry Regier's thesis (for a copy write to regier@icsi) gives some inkling of what image-schemas might be from a neural perspective. His connectionist spatial model of spatial relations computes image-schemas getting the topological properties via models of topographic maps of the visual field and the orientational properties via models of orientation-sensitive cells. In short, I think image-schemas have neural representations and that structured connectionist models like Regier's can represent them. 2. I stand behind the suggestions I made in WF&DT that (A) frames are structured by image-schemas and (B) my spatialization of form hypothesis that syntactic relations are also structured by image-schemas -- including container-schemas, source-path-goal schemas, part-whole schemas and link schemas. For discussion see Edelman's Bright Air, Brilliant Fire -- last section. Also Paul Deane's book. Deane's book is long and complex, and I like some parts and would question others. But I think the general thrust is right-headed. It's an important enough book to merit detailed discussion. Is Deane on this list? If not, put him on it right away. I think the way to proceed is to back up on these discussions. Pick one question. Make sure everybody relevant is in on it. Could Ed send everybody the current list of subscribers so that we can see if anybody who should be in on the discussion has been left off. First, we make the list is right, and then start off with either one issue or a short list of issues. George ` From cogling-relay@UCSD.EDU Fri Jan 14 21:49:37 1994 Received: from ucsd.edu by cogsci.Berkeley.EDU (5.63/1.29) id AA15553; Fri, 14 Jan 94 21:49:33 -0800 Received: from localhost by ucsd.edu; id VAA06106 sendmail 8.6.4/UCSD-2.2-sun Fri, 14 Jan 1994 21:43:23 -0800 for cogling-list Received: from umd5.umd.edu by ucsd.edu; id VAA06102 sendmail 8.6.4/UCSD-2.2-sun via ESMTP Fri, 14 Jan 1994 21:43:21 -0800 for Received: from hamlet.umd.edu by umd5.umd.edu(8.6.4/1Dec93) with ESMTP id AAA16122; Sat, 15 Jan 1994 00:43:13 -0500 From: Mark Turner Received: from localhost by hamlet.umd.edu id AAA18186; Sat, 15 Jan 1994 00:43:12 -0500 Date: Sat, 15 Jan 1994 00:43:12 -0500 Message-Id: <199401150543.AAA18186@hamlet.umd.edu> To: cogling@ucsd.edu Subject: Robinson's debate Status: R As Gilles Fauconnier has mentioned, we will make a technical report available (electronically) in a few weeks adumbrating a model that, among other things, is meant to subsume and extend various analyses of projection from abstract schematic structures to more specific structures, at levels from the conceptual to the syntactic. These schematic structures, including their frame structure, operate from what Fauconnier and I call generic middle spaces. This technical report (a short and place-holding version of a book we are trying to write on the subject) suggests but does not state a revision of the invariance hypothesis that goes back in spirit to the earliest suggestions of that hypothesis (pages 143-148 and 162-166 of Death is the Mother of Beauty), which Lakoff and I generalized and revised considerably into the invariance hypothesis in More than Cool Reason. The revised principle is that conceptual projection shall not result in a clash of schematic structure in the target. One corollary is that in a completed projection, the most abstract generic space, the one which contains just the schematic structure taken to apply to both source and target, shall contain no image-schematic clash. This version differs somewhat from the version of the invariance hypothesis in More Than Cool Reason or in Lakoff's subsequent publications; it acknowledges the many ways of achieving such a completed projection, including constraint and projection from the target, as well as dynamic and on-line reconceptions of source and target in efforts to avoid such a clash. (Some of this discussion of the invariance hypothesis is presented in Reading Minds, pages 54-64 and pages 172-182. Pages 180-182 present some speculation about the correspondence of conceptual and neural topology that I describe in the text as wild, given the current state of knowledge.) There are many ways to avoid such a clash. We have choice in what we project to the source, what we project to the target, and what we project to the generic space and therefore to the target. We can vary all of these choices in order to meet the constraint. For example, "Italian is the mother of Latin" can be interpreted as meeting the constraint so long as we project to the target not the space of historical derivation of Romance languages but rather the space of order of learning languages in a particular school. There is one way to meet the constraint that may not be obvious. Suppose a (crackpot) professor of Romance languages says, "Italian is the mother of Latin," and we respond, "You can't mean what I think you mean, you can't mean that Latin as a language derives from Italian as a language," and the professor responds, "Oh, yes, I do. You see, originally there was a small tribe of Etruscans who developed a new language, and it was Italian. Latin really derived from that language, although nobody recognizes this since the original speakers of Italian couldn't write - they just carved wooden sarcophagi lids all the time, and this took up such a ton of energy every day that they were always too tired to get around to learning the alphabet - so they left no documents. Later on, of course, Latin turned back into Italian, but in fact Italian never stopped being spoken, up in the Apenines; it was just never written down. So you see, everyone thinks Latin is the mother of Italian, but as a matter of historical fact, Italian is the mother of Latin." In this case, we are being directed to meet the constraint of the invariance principle not by changing altogether the spaces projected to source, target, or generic space, but rather by erasing and reforming the image- schematic structure in the mental space "historical derivation of Romance languages" so that when that reformed space is recruited to structure the target (which contains just Italian and Latin), the result will not be a clash with the image-schematic structure we are hoping to be able to project to the target from the source by way of a generic space. In fact, we are unlikely to accept this direction. We are more likely to form a new mental space of this professor's belief of the historical derivation of Romance languages. If we project this belief space to the target, then the invariance principle holds. But we will also certainly tell the professor (though not in these words) that our mental space of our belief of the historical derivation of Romance languages is not identical to our mental space of this professor's belief of the historical derivation of Romance languages, and that we see no reason to unify them. The question of the neurobiology of schematic structure in conceptual projection is, I think, very hard. It is relatively easy to see image-schemas at work in our behavior and our language. If we care to walk in the rain, we must go outside our house- container so we will not be under a roof that stops the rain from falling down onto us, and we must move along a path out of doors. We not only behave according to these image-schemas, we also describe our behavior in language that refers to them. It is harder to locate image-schemas at work in the brain, but there are some early results. The cerebellum, for example, has traditionally been recognized as a specialized part of the brain suited for neuronal group patterns whose activation results in sequences of precisely timed and coordinated movement, like throwing a curve ball or touch- typing a common word or playing a theme on the piano. What we would like to know is how such brain patterns for spatial movement are connected across modalities: when we see someone throw a rock at a window, the visual image-schemas according to which we recognize and understand the event are presumably connected to the kinesthetic image-schemas according to which we would perform the event, the auditory image-schemas that are part of the structure of the event, even the tactile image-schemas of touching the rock. Theories of the connections of such image-schemas have only recently been developed, and remain speculative. Antonio Damasio has proposed a complicated neurobiological model of "convergence zones" that might have something to say about such cross- modal integration. Gerald Edelman has proposed a controversial but tantalizing process - "reentrant mapping" - that provides a model of cross-modal integration, and of integration across submodalities, like form, motion, and color in vision. The most specific evidence of image- schemas in the brain, and of the connection of image-schemas in the brain, that I know of comes from reports of what are known as "orientation tuning" columns. The primary visual cortex responds to moving bars of light in an interesting way: a given neuron will have a preferred "orientation tuning" - it will respond best to a bar at a given angle. Other neurons in the column appear to have the same preferred stimulus, so that the column constitutes a neuronal group of cells that fire together in time in an organized manner to recognize a line at a preferred angle. Different orientation columns prefer different angles. Orientation tuning columns, then, might be neurobiological image-schemas for structuring certain kinds of visual experience, and thus for understanding it. Interestingly, these orientation tuning columns in the primary visual cortex are connected to neuronal groups in another, separate visual map, known as V2. These two separate but connected visual maps respond coherently to the same preferred stimulus - which suggests that image-schemas in primary visual cortex are coordinated with analogous image- schemas in V2. It would be a mistake, I think, to overwork these beginning results or to interpret them too rigidly. It is not at all clear how to connect the evidence for image-schemas in the study of the mind to the evidence for image- schemas in the study of the brain. It may be, for example, that the neurobiological analogue of an image-schema is not one neuronal group pattern but rather the interactive complex of several neuronal group patterns in different sites, all coordinated. The best evidence to date of the specific nature of image-schemas comes from the work they appear to do in language, although Mandler's evidence from "How to Build a Baby" offers intriguing lines of possible research. (See Antonio R. Damasio, "Time-locked multiregional retroactivation: A systems-level proposal for the neural substrates of recall and recognition" Cognition, 33 (1989), pages 25-62. "This article outlines a theoretical framework for the understanding of the neural basis of memory and consciousness, at systems level. It proposes an architecture constituted by: (1) neuron ensembles located in multiple and separate regions of primary and first-order sensory association cortices ('early cortices') and motor cortices; they contain representations of feature fragments inscribed as patterns of activity originally engaged by perceptuomotor interactions; (2) neuron ensembles located downstream from the former throughout single modality cortices (local convergence zones); they inscribe amodal records of the combinatorial arrangement of feature fragments that occurred synchronously during the experience of entities or events in sector (1); (3) neuron ensembles located downstream from the former throughout higher-order association cortices (non-local convergence zones), which inscribe amodal records of the synchronous combinatorial arrangements of local convergence zones during the experience of entities and events in sector (1); (4) feed-forward and feedback projections interlocking reciprocally the neuron ensembles in (1) with those in (2) according to a many-to-one (feed forward) and one-to-many (feedback) principle. I propose that (a) recall of entities and events occurs when the neuron ensembles in (1) are activated in time-locked fashion; (b) the synchronous activations are directed from convergence zones in (2) and (3); and (c) the process of reactivation is trigged from firing in convergence zones and mediated by feedback projections. This proposal rejects a single anatomical site for the integration of memory and motor processes and a single store for the meaning of entities or events. Meaning is reached by time-locked multiregional retroactivation of widespread fragment records. Only the latter records can become contents of consciousness.") (See also Gerald Edelman, Neural Darwinism: The Theory of Neuronal Group Selection [New York: Basic, 1987], passim; and Gerald Edelman, The Remembered Present: A Biological Theory of Consciousness (New York: Basic, 1989), chapter 4, "Reentrant Signaling," pages 64-90; and O. J. Sporns, J. A. Gally, G. N. Reeke, Jr., and G. M. Edelman, "Reentrant Signaling Among Simulated Neuronal Groups Leads to Coherency in Their Oscillatory Activity" Proceedings of the National Academy of Science 86 (1989): 7265-7269.) > Langacker makes wonderful headway in explaining 'syntactic' > phenomena with, inter alia, networked shemata, but the > schemata he employs are not typically image scemata. > Yes, Langacker has. Where did you get the idea that the schemas he > refers to are not image schemas? I'm certain that he would say they > are. > i don't think langacker would like his scemata id'd w/image-schemas qua > mark johnson but more w/ chuck fillmore's frames. Langacker told me in about 1990 that he regarded what Mark Johnson calls "image-schemas" as a subcategory of what he calls "images." Fauconnier and I hope that the model we propose in the technical report captures some of the unity we see running across various treatments of projections of schematic structure to more specific spaces and constructions, as presented in work by Langacker, Lakoff and Johnson, Fauconnier, Talmy, Sweetser, Goldberg, Fillmore, Fillmore and Kay, Mandler, various analogy theorists, Lakoff and Turner, and Turner, including projections that structure "syntactic phenomena." From: George Lakoff Some responses: 1. Terry Regier's thesis (for a copy write to regier@icsi) gives some inkling of what image-schemas might be from a neural perspective. His connectionist spatial model of spatial relations computes image-schemas getting the topological properties via models of topographic maps of the visual field and the orientational properties via models of orientation-sensitive cells. In short, I think image-schemas have neural representations and that structured connectionist models like Regier's can represent them. 2. I stand behind the suggestions I made in WF&DT that (A) frames are structured by image-schemas and (B) my spatialization of form hypothesis that syntactic relations are also structured by image-schemas -- including container-schemas, source-path-goal schemas, part-whole schemas and link schemas. See pp. 290-291. For discussion see Edelman's Bright Air, Brilliant Fire -- last section. Also Paul Deane's book. Deane's book is long and complex, and I like some parts and would question others. But I think the general thrust is right-headed. It's an important enough book to merit detailed discussion. Is Deane on this list? If not, put him on it right away. Response to Rick Grush: Rick wrote: I read Lakoff at trying, with debateable results, to address this with his so-called invariance hypothesis, which claims that structural relations from the source domain, even those not crucial to the metaphorocal extension at hand, are mapped to the target domain, effectively constraining aspects of the usage of the extended family of terms. Comment: I'm not sure what you meant by ``even those not crucial to the metaphorical extension at hand, ...effectively constraining aspects of usage of the extended family of terms.'' That doesn't sound familiar to me. What did you have in mind? Can you give an example or two? It's important, of course, to distinguish between a fixed conventional mapping and the use of the whole system of mappings for some particular example. The invariance hypothesis has two parts: 1. The mapping ``adapts'' to the inherent structure of the target domain. Another way to put it is that the inherent structure of the target domain remains fixed and any part of the source-to-target mapping that would be inconsistent with it is ``overridden'' -- though the term ``overriding'' may be misleading in that it may sound like something is being cancelled out, which is not intended. 2. In the part that gets mapped, image-schema structure is preserved, e.g., interiors of containers get mapped onto interiors, exteriors onto exteriors, etc. There is of course more that limits these mappings: experiential bases (experiential correspondences across domains), and other constraints (like those presnted in Pam Morgan's paper at the October conference at UCSD). In any particular case, additional particular knowledge will be mapped as predicted by the fixed mappings in the conceptual system. Again, I'm not sure what kinds of examples Rick has in mind. Rick, Can you give me some help? About Mark Turner's comments: I read an early version of Mark and Gilles' paper, which wasn't explicit enough for me to make sense of some of Mark's comments. When the report comes out, I hope we can discuss it thoroughly. George From cogling-relay@UCSD.EDU Mon Jan 17 15:58:55 1994 Received: from ucsd.edu by cogsci.Berkeley.EDU (5.63/1.29) id AA29032; Mon, 17 Jan 94 15:58:49 -0800 Received: from localhost by ucsd.edu; id PAA07612 sendmail 8.6.4/UCSD-2.2-sun Mon, 17 Jan 1994 15:51:14 -0800 for cogling-list Received: from cogsci.UCSD.EDU by ucsd.edu; id PAA07608 sendmail 8.6.4/UCSD-2.2-sun via SMTP Mon, 17 Jan 1994 15:51:13 -0800 for Received: by cogsci.UCSD.EDU (4.1/UCSDPSEUDO.4) id AA29523 for cogling@ucsd.edu; Mon, 17 Jan 94 15:51:11 PST From: faucon@cogsci.UCSD.EDU (Gilles Fauconnier) Message-Id: <9401172351.AA29523@cogsci.UCSD.EDU> Date: Mon, 17 Jan 1994 15:49:19 -0800 To: rgrush@sdcc3.UCSD.EDU Subject: Re: invariance hypothesis Cc: cogling@ucsd.edu, faucon@cogsci.UCSD.EDU Status: R Re: Rick Grush Our example on Latin and Italian mentioned by Mark in his posting shows straightforwardly the inadequacy of the way we've all been talking about conceptual domains, as either having inherent structure waiting to be matched, or acquiring new structure through projection. What's going on is more interesting. Recall the two (of many) interpretations: (i) Italian is the daughter of Latin (standard 'derived language' reading) (ii) Latin is the daughter of Italian (learning Italian is a helpful prerequisite for learning Latin). The source (mother/daughter) is an archetype for more abstract notions of causation and generation. Suppose causation/generation is a schema S. (ii) projects this schema to a target space. But it doesn't tell us what that space is. It only constrains it to match the structure: Italian ----S-----> Latin To construct such a space, we need to find a much more elaborate frame (e.g. language learning) in which a relation R (e.g. "learning x causes learning y") fits schema S. Clearly, there is no sense in which the conceptual domain of Latin and Italian is or is not structured by the high-level schema S. The question only makes sense with respect to particular target spaces that we build up. Our conception of Latin and Italian can therefore fit S in one direction with respect to a space of language genealogy, and fit S in the other direction with respect to a space of language learning. The metaphors are not adding or matching static conceptual structure; they are guiding our construction of appropriate spaces and frames by specifying higher schematic constraints that must be met. But of course, it's not just metaphors that do this. It's also a general function of grammar. Grammar provides (among other things) high level schematic constraints, reflected by Langacker's schemas or Fillmore's constructions, which, in order to yield real interpretations, must be mapped onto much more specific frames and spaces that are constructed locally and contextually. [Our folk theory of language tends to mask this, by fostering the illusion that words and sentences are directly associated with full meanings]. gilles fauconnier p.s. incidentally, the very interesting invariance hypothesis will fall out automatically from this account, and so will, in part the creative (rather than simple linguistic convenience) quality of metaphor, although the latter needs the additional notion of blended space. From cogling-relay@UCSD.EDU Mon Jan 17 12:22:16 1994 Received: from ucsd.edu by cogsci.Berkeley.EDU (5.63/1.29) id AA27625; Mon, 17 Jan 94 12:22:09 -0800 Received: from localhost by ucsd.edu; id MAA15112 sendmail 8.6.4/UCSD-2.2-sun Mon, 17 Jan 1994 12:15:02 -0800 for cogling-list Received: from sdcc3.UCSD.EDU by ucsd.edu; id MAA15094 sendmail 8.6.4/UCSD-2.2-sun via SMTP Mon, 17 Jan 1994 12:14:58 -0800 for Received: by sdcc3.UCSD.EDU (4.1/UCSDGENERIC.3) id AA02082 to cogling@ucsd.edu; Mon, 17 Jan 94 12:14:57 PST Date: Mon, 17 Jan 94 12:14:57 PST From: rgrush@sdcc3.UCSD.EDU (Rick Grush) Message-Id: <9401172014.AA02082@sdcc3.UCSD.EDU> To: cogling@ucsd.edu Subject: invariance hypothesis Status: R From: Rick Grush I said that one of the claims of the invariance hypothesis was that structural relations from the source domain get mapped to the target domain, even in cases where the exact area of structural projection is not crucial to the mapping at hand. I am surprised that Lakoff would find this in any way controversial, as it seemed to me to be one of the primary theses of the image schemata/metaphorical mapping contingent. Perhaps I just used unfortunate and unclear language. All I meant is this: Take an instance of a metaphor, 1) Look how far we've come. Spoken by one member of a romantically involved couple (example from Lakoff (1990) "The Invariance Hypothesis: Is abstract reason based on image-schemas?" Cognitive Linguistics 1:1:39-47). This metaphor involves the projection of some structure from a source domain, journeys, to a target domain, romantic relationships. However, the source domain in this case includes many entities and much structure which are not explicitly mapped in this instance of metaphor. Nonetheless (and I understood this to be explicitly embraced by Lakoff) these aspects, which are not essential to the projection at hand, are still mapped to the source domain (perhaps implicitly). As Lakoff says: "Such a conceptual metaphor explains why new and imaginative extensions of the mapping can be understood instantly, given the ontological correspondences and other knowledge about journeys." And as long as I have Lakoff's ear, and maybe some other informed ears as well, I have some questions about image schemas and the invariance hypothesis. First, I just get more and more confused as to exactly what is supposed to count as image-schematic structure. My initial understanding was something like Image-schematic structure is structure which is derived from low- level* sensory, motor, or sensorimotor experience. Specifically, it is distilled from instances of such bodily experience which have sufficient structural commonalities to support some degree of schematization. What confuses me is that the term 'image schematic structure' seems to be used more and more for sorts of structure which seem to have little to do with basic bodily experience (for example, a recent posting to this list, when discussing the extension of mother/daughter relationships to language phylogeny described the source domain, in this case human genetic relationships, as image schematic). This is why I earlier resisted characterizing the schemata Langacker used as typically image schematic, as he helps himself (at least some of the time) to schemata which don't seem to be image schematic in the sense explained above (which is fine because Langacker doesn't seem to make Lakoff's and Johnson's metaphysical project (see below) central to his own). This bleaching of the term 'image schematic structure' troubles me because I thought that one of the purposes of identifying image schematic structure with low-level bodily experience was to avoid problems with what Lakoff called (in WF&DT) objectivist metaphysics. The problem, as I understood it, was in assuming that the objective world has all this intricate structure independent of our knowledge or conceptualization of it, and the solution was supposed to be that the putative objective structure of the world is really no more than metaphorically projected structure from the less-metaphysically-extravagant domain of bodily experience. But if anything can count as image-schematic, then no progress can be made. I describe the source of image schematic structure as 'low-level' bodily experience, and it seems to me that, in order to carry out the interesting metaphysical project that Lakoff, as well as Johnson, advocate, 'low-level' must have some teeth. There is a sense, to be sure, in which genetic relationship trees are things I've had sensorimotor experience of -- I have drawn representations of them on paper, for example. But I've also had similar sensorimotor experience programming computers, solving wave equations, doing predicate calculus, etc., and if these sorts of things (e.g. a schematic approach to solving wave equations) can count as image schematic, then no progress has been made in avoiding the objectivists' metaphysical assumptions. We are _assuming_ structure in the world, such as genetic relationships, and not _explaining_ it. This leads to my final confusion, which is the prima facie incompatibility between the metaphysical project alluded to above and the way in which metaphorical mappings are typically explained. Metaphorical projections are often described as mappings of structure from a source domain to the structure of some target domain, but this seems to presuppose that the target domain has structure independent of the mapping from the source domain (what else could be meant by 'mapping to THE STRUCTURE OF THE TARGET DOMAIN'?) Lakoff makes it explicit: "The invariance hypothesis has two parts: 1. The mapping ``adapts'' to the inherent structure of the target domain. Another way to put it is that the inherent structure of the target domain remains fixed and any part of the source-to-target mapping that would be inconsistent with it is ``overridden'' -- though the term ``overriding'' may be misleading in that it may sound like something is being cancelled out, which is not intended. 2. In the part that gets mapped, image-schema structure is preserved, e.g., interiors of containers get mapped onto interiors, exteriors onto exteriors, etc." Well, if the target domain has its structure inherently, then what's all the excitement about metaphorical projections and image schemata? That is, it seems to me that there are (at least) two possibilities, one which strikes me as very interesting, and one which strikes me as sort of interesting. They are, in order: a) Metaphorical mappings, from 'image-schematic' sources to 'abstract' targets are what underwrite and licence the very structure of (including legitimate inferences and reasoning about) the target domain. b) Metaphorical mappings are ways in which terms and language from one domain of experience can be used to help describe reasoning about some other target domain, but this is just a linguistic convenience in that the target domain is inherently structured and thus supports intrinsic relationships and inference independently of the mapping. Lakoff seems to want to tow the (a) line: "The data covered by the Invariance Hypothesis includes the metaphorical understanding of time, states, events, actions, purposes, means, causes, modalities, linear scales, and categories. Because the source domains of these metaphorical concepts are structured by image schemas, the Invariance Hypothesis suggests that reasoning involving these concepts is FUNDAMENTALLY IMAGE- BASED." (ibid., p. 39, emphasis added). But in what sense is this understanding fundamental if the mapping itself requires that the target have its structure independently and inherently? To summarize (and destroy any remaining patience the reader may have with my ramblings), what I take to be most interesting and most original about the Lakoff and Johnson project of explaining language use and understanding of 'objective' and 'abstract' domains in terms, ultimately, of metaphorical projections of image schemata seems to require (at least) two components: A) Image schemata must be derived from some domain of experience which is somehow less metaphysically suspect than the objective abstract domains which we are trying to account for (otherwise what's the point?), and B) The metaphorical projections which purport to account for this structure should not assume that the target domains have their structure inherently and independent of the mapping (otherwise we're just closet objectivists). A lot of the usage of 'image schema' seems to run afoul of (A), which maybe is ok, -- not everyone, I suppose, shares Lakoff's and Johnson's metaphysical project. And just about everything I've heard about metaphorical projections, including from Lakoff and Johnson, seems to run afoul of (B), which, if true, is less ok. I'll stop babbling now. Rick Grush Departments of Philosophy and Cognitive Science, UCSD rgrush@sdcc3.ucsd.edu From: George Lakoff Re: Reply to Rick Grush First, I want to thank Rick for his note. It contains a number of important misunderstandings of my views -- misunderstandings that are no doubt the result of my not being clear enough. It is only by such queries as Rick's that I can get a better idea of what issues need to be discussed. Rick writes: Metaphorical projections are often described as mappings of structure from a source domain to the structure of some target domain, but this seems to presuppose that the target domain has structure independent of the mapping from the source domain (what else could be meant by 'mapping to THE STRUCTURE OF THE TARGET DOMAIN'?) My Reply: Target domains of metaphorical mappings have some, perhaps very minimal, structure of their own -- BUT THEY ARE NOT COMPLETELY STRUCTURED ON THEIR OWN. Having some inherent structure is not the same as being completely structured. Metaphorical mappings typically add structure to what is inherently there. Take the obvious example of good old LOVE IS A JOURNEY. The targt domain of love has some inherent structure -- it has at the ver least two people, the lovers. It has a relationship between them. It has some notion of difficulty in a love relationship. Love does not inherently have paths, vehicles, impediments to travel, destinations, crossroads, etc. They are added by the metaphorical mapping, along with the inference patterns that goes with them. Or take another classic, the Conduit Metaphor. The target domain has ideas, language, speaking, hearing. It has the built-in knowledge that when you tell someone an idea, you still have it afterwards. That cannot be constradicted by any metaphor. But the mental domain doesn't inherently have words as `containers' that ideas `go in', it doesn't have the `sending' and `receiving' of ideas and all the inferences that go with those notions. It doesn't the `putting' of ideas `into' words or the idea `going over my head'. That structure is added by the metaphor. As Rick says: Lakoff makes it explicit: "The invariance hypothesis has two parts: 1. The mapping ``adapts'' to the inherent structure of the target domain. Another way to put it is that the inherent structure of the target domain remains fixed and any part of the source-to-target mapping that would be inconsistent with it is ``overridden'' -- though the term ``overriding'' may be misleading in that it may sound like something is being cancelled out, which is not intended. 2. In the part that gets mapped, image-schema structure is preserved, e.g., interiors of containers get mapped onto interiors, exteriors onto exteriors, etc." There is no problem here, given that the target domain is only partially, perps minimally, structured on its own. Rick goes on: Well, if the target domain has its structure inherently, then what's all the excitement about metaphorical projections and image schemata? The misunderstanding is in the phrase ``its structure'' which assumes ALL of its structure. The excitement is that it only has some, perhaps very little inherent structure, and gets the rest metaphorically. There is more excitement as well -- different metaphors add different structures (often inconsistent structures!) to whatever inherent structure there is in the target domain. Thus, there can be two distinct and inconsistent metaphorical structurings of a domain. There are many well-known examples. Take the target domain of marriage. Marriage can be conventionally conceptualized both as a haven and as a struggle. Marriage can be conventioanally conceptualized both as the formation of a single person or as a means for personal growth. Such contradictory conventional conceptualizations via metaphor are common. Rick writes: That is, it seems to me that there are (at least) two possibilities, one which strikes me as very interesting, and one which strikes me as sort of interesting. They are, in order: a) Metaphorical mappings, from 'image-schematic' sources to 'abstract' targets are what underwrite and licence the very structure of (including legitimate inferences and reasoning about) the target domain. b) Metaphorical mappings are ways in which terms and language from one domain of experience can be used to help describe reasoning about some other target domain, but this is just a linguistic convenience in that the target domain is inherently structured and thus supports intrinsic relationships and inference independently of the mapping. My Reply: My actual position is (c): Some of both -- a lot of (a) and a little of (b). It is the partial inherent structuring of the target domain which cannot be overridden that can produce cases of (b). But the really interesting part is what is added, namely (a). The cases where there are contradictory metaphors shows that (b) cannot always be the case. There are other exciting things as well -- for example, blended spaces, where the metaphor holds in the new space. Take a Bugs Bunny cartoon where Elmer Fudd has smoke coming out of his ears. The Anger Is Heat metaphor structures that imaginative domain. Or take a culturally important case, one discussed at length by Emily Martin in her forthcoming book ``Flexible Bodies'': Many biologists and doctors really believe the military metaphor for the immune system. For them, the blended space in which the immune system ``fights off'' ``invading'' viruses is the only reality. On to the second misunderstanding. Rick writes:: I said that one of the claims of the invariance hypothesis was that structural relations from the source domain get mapped to the target domain, even in cases where the exact area of structural projection is not crucial to the mapping at hand. I am surprised that Lakoff would find this in any way controversial, as it seemed to me to be one of the primary theses of the image schemata/metaphorical mapping contingent. Perhaps I just used unfortunate and unclear language. All I meant is this: Take an instance of a metaphor, 1) Look how far we've come. Spoken by one member of a romantically involved couple (example from Lakoff (1990) "The Invariance Hypothesis: Is abstract reason based on image-schemas?" Cognitive Linguistics 1:1:39-47). This metaphor involves the projection of some structure from a source domain, journeys, to a target domain, romantic relationships. However, the source domain in this case includes many entities and much structure which are not explicitly mapped in this instance of metaphor. Nonetheless (and I understood this to be explicitly embraced by Lakoff) these aspects, which are not essential to the projection at hand, are still mapped to the source domain (perhaps implicitly). As Lakoff says: "Such a conceptual metaphor explains why new and imaginative extensions of the mapping can be understood instantly, given the ontological correspondences and other knowledge about journeys." The crucial part is: ``... these aspects, which are not essential to the projection at hand, are still mapped to the source domain''. Two things need to be distinguished: 1. The fixed conventional mapping, which is a fixed cross-domain correspondence. 2. Novel extensions of 1, which are not fixed. The novel extensions must be consistent with the fixed conventional mappings, but different novel extensions can be chosen on different occasions. The language ``are still mapped'' doesn't portray this accurately. Different novel extensions are mapped at different times on different occasions. Rick's statement that mappings occur `` even in cases where the exact area of structural projection is not crucial to the mapping at hand'' does not accurately describe this situation. This is undoubtedly a case of my not being clear enough. On to image-schemas. Rick writes: And as long as I have Lakoff's ear, and maybe some other informed ears as well, I have some questions about image schemas and the invariance hypothesis. First, I just get more and more confused as to exactly what is supposed to count as image-schematic structure. My initial understanding was something like Image-schematic structure is structure which is derived from low- level* sensory, motor, or sensorimotor experience. Specifically, it is distilled from instances of such bodily experience which have sufficient structural commonalities to support some degree of schematization. Okay as far as it goes. This was Johnson's account in The Body In the Mind. I think Regier has given us much more insight into image- schemas in his dissertation and in various short papers. (For copies contact him at regier@icsi.berkeley.edu.) Regier's basic idea is that certain kinds of neural structures -- topographic maps and orientation sensitive cells -- combine in assemblies that compute what we have been calling ``image- schemas''. He talks mainly about computational models of such assemblies in the visual system -- connected to the retina. But neural assemblies doing the same kinds of computations could presumably occur in other places in the brain. Thus, image-schemas on such an account need not be visual; they might be tactile or auditory, for example. One of the crucial things about Regier's neural assemblies of topographic maps and orientation-sensitive cells is that they have built-in inferential structure. Regier doesn't actually build the inference engine in the thesis, but the structure is there. Rick writes: ...I thought that one of the purposes of identifying image schematic structure with low-level bodily experience was to avoid problems with what Lakoff called (in WF&DT) objectivist metaphysics. The problem, as I understood it, was in assuming that the objective world has all this intricate structure independent of our knowledge or conceptualization of it, and the solution was supposed to be that the putative objective structure of the world is really no more than metaphorically projected structure from the less-metaphysically-extravagant domain of bodily experience.... ...I describe the source of image schematic structure as 'low-level' bodily experience, and it seems to me that, in order to carry out the interesting metaphysical project that Lakoff, as well as Johnson, advocate, 'low-level' must have some teeth... My Reply: I agree. Regier's enterprise is to do just that. Rick's summary: To summarize (and destroy any remaining patience the reader may have with my ramblings), what I take to be most interesting and most original about the Lakoff and Johnson project of explaining language use and understanding of 'objective' and 'abstract' domains in terms, ultimately, of metaphorical projections of image schemata seems to require (at least) two components: A) Image schemata must be derived from some domain of experience which is somehow less metaphysically suspect than the objective abstract domains which we are trying to account for (otherwise what's the point?), and B) The metaphorical projections which purport to account for this structure should not assume that the target domains have their structure inherently and independent of the mapping (otherwise we're just closet objectivists). My Reply: Basically, I agree. Both Johnson's and Regier's accounts of image- schemas are in the spirit of (A). My real account of target domains as having only some, often little, inherent structure is in the spirit of (B). I have a quibble about the ``closet objectivists'' remark. Even if serious use of metaphor to structure target domains did not exist, the rest of the cognitive semantic apparatus -- frame semantics, mental spaces, types of prototypes, radial categories, contested concepts, and metonymy would be more than enough to keep us from being ``closet objectivists.'' Rick writes: What confuses me is that the term 'image schematic structure' seems to be used more and more for sorts of structure which seem to have little to do with basic bodily experience...This bleaching of the term 'image schematic structure' troubles me... My Reply: It trouble me too. I try to be very careful about using the term. However, the term is sometimes used with less than full care -- often when the issues involved in some particular example are considered irrelevant. Here is how I see the problem as arising. When Turner and I came up with the idea in More Than Cool Reason, we spoke of ``generic-level structure''. We were careful not to use the term ``image-schematic structure'' because some of the structure we were referring to -- negation, ability, etc. -- had not yet been given an image-schematic analysis at all, much less a convincing one. Image- schematic structure was taken to be at the very least a special case of what we called ``generic-level structure.'' However, we had the intuition that, sooner or later, what we called generic-level structure would turn out to be image-schematic once the requisite analyses were done. The Invariance Principle was intended to cover two kinds of cases: (a) those where clear image-schematic analysis already existed, and (b) those where future analysis of what was preserved in the mappings would turn out to be image-schematic. We would like not to have to have an additional principle to cover cases of generic-level structure that isn't image-schematic. We believe that the equation ``Generic-level structure = Image-schematic structure '' will turn out to be true. But it still too early to tell. It occasionally happens that one metaphor analyst or another (even yours truly) speaks somewhat impetuously, claiming that some relation is ``image-schematic'' when the details of the analysis are not yet worked out but where the case `feels' kind of image-schematic and the analyst thinks it will eventually turn out to be so. We are human and this happens. It might very-well be the case that when ``daughter-of'' is properly analyzed, it will be image-schematic. One off-the-top-of-my-head analysis that comes to mind is that daughter is a relation and relations are links (that is, a link schema represents the relation); another is that daughters come from mothers, and that this ``come from'' is a metaphorical version of a path schema. These are not implausible analyses, but there is no serious linguistic analysis arguing for them at the moment. Rick's concern in such cases is appropriate. It is very important to keep track of exactly which image-schemas have been postulated, how plausible they are, what their inferential properties are, whether Regier-type models of them have been done, what the evidence is for analyses using them, etc. If the Invariance Principle does not account for ALL of the structure preserved in metaphorical mappings, then we should know that. At present, I believe it does, but I can't prove it for anything near all known cases. All that can be done at present is to show that there are thousands of cases that do work. Personally, I believe that Langacker's schemas and Fillmore's frames, when properly analyzed, will turn out to have image-schematic structure. But I can't prove it for all cases. I can't even come close. But, as time goes on, more and more of his cases turn out to be image- schematic -- e.g., causation, purpose, aspect, etc. Such cases lead me to think that my hunch is right. But it's still only a hunch. The way I think about the disparities between Ron's work and mine is that we are doing different parts of a big job and that Ron's analyses may well require even further analysis in terms of image-schemas and metaphor. But one cannpt reasonably make such a general statement -- you have to look at each case, do the analysis and give the evidence for it. There is another possibility that ought to be mentioned: Perhaps image-schemas are epiphenoma. Perhaps what is really involved are ensembles of neural structures of the type Regier describes, some of which have an imagistic character and some of which don't. Perhaps it will turn out to be structures of that kind that are ``mapped'' -- or rather employed when reasoning about an abstract domain. It should be mentioned that these are not PDP-type ensembles -- rather they are biologically motivated cases like topographic maps that do well- understood computations. We are definitely not advocating an aproach that says that the neurons, somehow or other, compute the concepts. Anyway, such a theory would fit the views that Johnson and I have about the bodily grounding of metaphor even more closely. Rick uses the term ``metaphysical project'' to describe what Johnson and I are doing, as if it were just another philosophical project. It isn't. Once we learn that reason is not transcendental -- not an objective property of the universe that we happen to mirror -- and once we learn that concepts are not mirrors of some external reality, then we are faced with explaining (1) what concepts are, (2) what makes them meaningful, (3) how they are grounded, and (4) how it is possible to engage in abstract reason. In short, we need a new kind of theory of meaning to make sense of the empirical data -- about basic-level concepts, spatial relations concepts, metaphorical concepts, color concepts, etc. In all these cases, the empirical data leads one to bodily grounding and metaphorical extension to abstract domains. It is not a philosophical project. It is an attempt to account for an extraordinarily rich body of data. At present our conjecture is, so far as I know, ``the only game in town'' -- the only theory that tries to give a theory of meaning and abstract reason in terms of bodily grounding while accounting for the mountains of linguistic and psychological data. This is not something that Langacker is trying to do -- he's more than busy enough doing what he does best. I should also mention that, although I think a Regier-type story that goes down to the neural level will ultimately be required, I think that the kind of phenomenological analysis done by Johnson in The Body in the Mind is an important part of the story about image-schemas. But it is not the place where the analysis stops. Indeed, I take such a phenomenological account as more data to be explained. Before I sign off, I want to thank Rick again for his thoughtful remarks. They are exactly the sort of queries that help clarify issues. Schematic conceptions of Latin as linked to Italian and as "leading to" Italian, and of Italian as "emerging from" Latin; and of mother as "linked" to daughter and as "leading to" daughter, and of daughter as "emerging from" mother, are, I take it, viewed as based in image-schemata by all those who use the term. None of this highly schematic structure is specific to or based in concepts of genetics, and of course, there are vast ranges of detail in the conceptual domains of progeneration and of historical relationships of languages that are not image-schematic. Jean Mandler, incidentally, has considered insightfully the ways in which image-schemata would be based not only in bodily experience but also in perceptual experience, and ways in which analysis of perceptual experience might lead to image-schemata as a basis of some kinds of conceptual categorization. From cogling-relay@UCSD.EDU Wed Jan 19 12:49:46 1994 Received: from ucsd.edu by cogsci.Berkeley.EDU (5.63/1.29) id AA15143; Wed, 19 Jan 94 12:49:39 -0800 Received: from localhost by ucsd.edu; id MAA26830 sendmail 8.6.4/UCSD-2.2-sun Wed, 19 Jan 1994 12:43:16 -0800 for cogling-list Received: from ubvmsb.cc.buffalo.edu by ucsd.edu; id MAA26777 sendmail 8.6.4/UCSD-2.2-sun via ESMTP Wed, 19 Jan 1994 12:42:55 -0800 for Received: from ubvms.cc.buffalo.edu by ubvms.cc.buffalo.edu (PMDF V4.2-14 #5889) id <01H7V7PQPNKW8Y50AC@ubvms.cc.buffalo.edu>; Wed, 19 Jan 1994 12:30:37 EST Date: Wed, 19 Jan 1994 12:30:37 -0500 (EST) From: BARRY SMITH Subject: Structure and Objectivism To: cogling@ucsd.edu Message-Id: <01H7V7PQSC1E8Y50AC@ubvms.cc.buffalo.edu> Organization: University at Buffalo X-Vms-To: IN%"cogling@ucsd.edu" Mime-Version: 1.0 Content-Type: TEXT/PLAIN; CHARSET=US-ASCII Content-Transfer-Encoding: 7BIT Status: R Re: Discussion between George Lakoff and Rick Grush Perhaps a mere philosopher can be allowed to comment on some of the above. Lakoff writes: 'Target domains of metaphorical mappings have some, perhaps very minimal, structure of their own -- BUT THEY ARE NOT COMPLETELY STRUCTURED ON THEIR OWN. Having some inherent structure is not the same as being completely structured. Metaphorical mappings typically add structure to what is inherently there.' My response: surely everything that is 'inherently there' must therefore also be completely structured, in its way. Even a lump of jello or a whirlpool or a conflagration is completely structured. What, therefore, does Lakoff mean when he talks of things being not completely structured? To take Lakoff's own example: suppose one person loves another; everything about this relationship is completely structured, down to the smallest movements of the lips. Of course one can say that structure gets added by metaphors; but a more careful account of this is needed than one according to which there are regions of inherent reality which are somehow only partially structured in such a way that metaphorical architecture can somehow fill the gaps. Metaphorically added structure and inherent structure are surely not in this sense comparable at all, and so do not need to be made compatible by postulating inherent structure-gaps. Note that Lakoff's very idea that 'there can be two distinct and inconsistent metaphorical structurings of a domain itself' proves that the notion of a domain's being only partially structured is not needed to allow metaphorical structuring to go to work. And surely a metaphorical structuring can easily be inconsistent with a domain's inherent structure (lots of cases come to mind of this). Or can Lakoff give an example of what he calls a 'partial inherent structuring of the target domain which cannot be overridden'? I doubt it (given the the bloody-mindedness of speakers of natural languages). Perhaps what Lakoff means when he refers to the target domain as being 'only partially, perhaps minimally, structured on its own' is that the target domain is not a part of inherent spatio- temporal reality at all, but is rather something general and abstract. But this would mean that there are two sorts of schema and two stages of abstracting going on: a first, metaphor-free stage which takes us from genuine reality to this 'only partially structured' stuff, followed by a second stage where metaphors go to work. This seems to be what Lakoff has in mind where he refers to metaphorical mappings from 'image-schematic' sources to 'abstract' targets. It would presumably be in keeping also with Lakoff's use of the phrase 'generic-level structure'. Two jobs need to be performed, then: one job of describing how the image- schemas relate to their abstract-generic targets, and a second job of describing how the abstract-generic targets relate to spatio-temporally reality itself. Is this reference to 'spatio-temporal reality' out-of-the-closet objectivism? Is there a spatio-temporal reality outside the realm of schematic and abstract-generic targets? Lakoff certainly seems to believe that there is, for example when he writes approvingly of 'Regier's basic idea' 'that certain kinds of neural structures -- topographic maps and orientation sensitive cells -- combine in assemblies that compute what we have been calling "image- schemas"'. Presumably this 'inherent [neural, physical] reality' is not confined to the spaces between people's ears but extends outwards also to fill the whole universe. Or perhaps Lakoff's general criticism of objectivism does not apply to physics and neurology? (Forgive me if this is a silly speculation, but I am new to this list.) Why, come to that, does the general criticism of objectivism not apply to work on metaphors and image-schemata? Why should just these domains (neural nets, the ways people use words) be privileged and all else be 'metaphysically suspect'? Lakoff writes: 'Image schemata must be derived from some domain of experience which is somehow less metaphysically suspect than the objective abstract domains which we are trying to account for.' What then are the examples of metaphysically suspect domains? Lakoff concludes: 'The metaphorical projections which purport to account for [the structure yielded by projections] should not assume that the target domains have their structure inherently and independent of the mapping (otherwise we're just closet objectivists).' But it was Lakoff himself who started off by talking of 'what is inherently there' and of its inherent (if purportedly partial) structure. What is going on here? I am confused. Barry Smith Department of Philosophy and Center for Cognitive Science, University at Buffalo From: George Lakoff Re: Reply to Barry Smith Dear Barry, Your note indicates some major misunderstanding somewhere. It might help to clarify things if I take a few steps back and start by talking about the enterprise I am engaged in. 1. As a cognitive scientist, I am interested in studying human reason and human conceptual systems. Human knowledge is structured in terms of concepts in that conceptual system. We are not studying human knowledge or the nature of the world in general. We are instead studying the system of concepts in terms of which we think and in terms of which knowledge is represented. 2. Some basic results from the mid-1970's lead us to believe that human conceptual systems are structured in terms of conceptual domains. Fillmore's work on semantic fields and frame semantic showed that each semantic field (e.g. a collection of words about the same subject matter) is structured by a `frame' or `schema'. His classic example was the commercial event frame, in terms of which words like ``buy, sell, cost, price, goods,...'' are defined. That frame has various participants, an initial and final state, and a sequence of events. It forms a gestalt. and it structures a conceptual domain of its own, distinct from, say, the conceptual domain of paleonotology. This of course permits the possiblity that some specimen, say a dinosaur bone, might fill the role of GOODS in some COMMERCIAL EVENT. But the dinosaur bone is not a part of the general COMMERCIAL EVENT frame. The COMMERCIAL EVENT FRAME is independent of particular instances of commercial events, and is used to structure all particular instances. The kind of evidence we have for the existence of such frames as this comes from general inferential relations among the words in the semantic field. Len Talmy has also done considerable work showing that our conceptual system is structured in terms of distinct conceptual domains. He has done a great deal of work showing that space is a separate conceptual domain, and has contributed enormously to our understanding of its structure. Note that this does not mean that in the real world space is separate from dinosaur bones, all of which occur in space. Nor does it mean that our knowledge about space does not contain knowledge about the things that occur in space. It does mean that our system of spatial concepts is independent of any particular piece of knowledge about any particular space or what occurs in it. And that system of spatial concepts forms a conceptual domain separate from the conceptual domains of commercial events or paleonotology. The point: Concepts from lots of conceptual domains are used in structuring any piece of complex real-world knowledge. But conceptual systems are structured so that conceptual domains are distinct, separate, and have internal structure of their own. That structure is relatively simple, compared to the complexity of our knowledge of the world. That is, my concept of what a commercial event in general is is much simpler than my knowledge about particular commercial events that happen to have taken place. Each conceptual domain has an inherent structure -- whatever must be there in ALL cases of the use of the schema that structures that domain. The inherent structure of each domain of concepts (e.g., general commercial events) is rather simple. In each commercial event, there is a buyer, seller, what is sold, what it is sold for, an intial state before the sale, the exchange constituting the sale, and the final state after the sale. I think that is all that is in the inherent structure of the general frame. Thare are of course lots of special case frames for special kinds of commercial events -- and they have more structure. And of course anyone's knowledge about a particular commercial event is likely to have lots more structure about the context of the event. And a particular commercial event IN THE WORLD, as opposed to someone's knowledge of it, would be structured as hell. 3. It has also been discovered that many (if not most) conceptual domains are structured metaphorically. The evidence comes from at least six sources: generalizations over polysemy, generalizations over inference structure, generalizations over novel extensions, generalizations over semantic change, psycholinguistic experiments, and evidence from gesture studies. For a discussion of such evidence and a bibliography, see my maper ``The contemporary Theory of Metaphor'' in the second edition of Andrew Ortony's METAPHOR AND THOUGHT, just out from Cambridge. For the experimental evidence see Ray Gibbs' great new book THE POETICS OF MIND, about to appear from Cambridge. What this means is that many domains are conceptualized not just in terms of their inherent structure, but also in terms of the structures of other conceptual domains. Each such metaphorical structuring is done by a mapping. The domain structured is the Target Domain. The domain in terms of which it is structured is the Source Domain. The mapping is called a ``metaphor.'' The study of conceptual metaphpr is the study of the properties of such mappings and of the particular mappings that are in the conceptual systems underlying various languages. Much of that study concerns the fixed, conventional mappings that have been found. These are fixed cross-domain correspondences, each of which provides a fixed, conventional way of conceptualizing some target domain. One of the things that has been discovered is that the inherent structure of a target domain is always preserved in such mappings. At this point, I think I can describe the misunderstanding in Barry's note. Barry writes: >Lakoff writes: 'Target domains of metaphorical mappings have some, perhaps very minimal, structure of their own -- BUT THEY ARE NOT COMPLETELY STRUCTURED ON THEIR OWN. Having some inherent structure is not the same as being completely structured. Metaphorical mappings typically add structure to what is inherently there.' >My [Barry's] response: surely everything that is 'inherently there' must therefore also be completely structured, in its way. Even a lump of jello or a whirlpool or a conflagration is completely structured. What, therefore, does Lakoff mean when he talks of things being not completely structured? To take Lakoff's own example: suppose one person loves another; everything about this relationship is completely structured, down to the smallest movements of the lips. >Of course one can say that structure gets added by metaphors; but a more careful account of this is needed than one according to which there are regions of inherent reality which are somehow only partially structured in such a way that metaphorical architecture can somehow fill the gaps. Metaphorically added structure and inherent structure are surely not in this sense comparable at all, and so do not need to be made compatible by postulating inherent structure-gaps. My Response to Barry: Barry is confounding three different things: (I) The world itself, knowledge about the world, and (III) the conceptual system used to structure knowledge about the world. The theory of conceptual metaphor is about (III). But Barry is talking about (I). It is clear why Barry is ``confused.'' Talking about the structure of a system of human concepts is very different from talking about the world. For this reason, most of Barry's `criticisms' make no sense at all. However, Barry's comments evoke important issues. Objectivist philosophy assumes that the world in itself comes with a structure -- a single, unique, correct structure from some God's eye view. Personally, being a mere human being, I am not in a position to know whether this is true or not. All I can do is try to comprehend the world using my human conceptual system -- including the parts that are metaphorical -- for example, my metaphors for time, event structure, causation, etc. Any `knowledge' that I have of the world has got to be represented using human concepts, which are grounded in the body and extended by various imaginative devices: frames, metaphors, prototypes, etc. There is no way that any human being can conceptualize anything without using a human conceptual system. Now human conceptual systems work pretty well for certain aspects of experience (dealing with basic-level experience, for example) and less well for other aspects. The sciences also can only use human conceptual systems. In the ``successful'' sciences, human conceptual systems have been consciously devised to allow us to comprehend certain very limited aspects of reality to a degree that we are pretty happy with -- a degree that allows us to do engineering of various kinds, for example. Do the human concepts used in ``successful'' sciences fit the structure of the world in itself if such a structure exists? There is no way that I or any other human being could tell, because all that I and the rest of us have available is human concepts. We have no direct access to whatever ``objective'' structure there might be from a God's Eye View. It may be that the ``successful'' sciences have done the best that one can do with only the resources of human conceptual systems -- or that they have done as well as we can tell. This is not to discredit science at all. Quite the opposite: doing that well with only human conceptual systems is a remarkable achievement. At this point I can reply to another of Barry's criticisms. Barry claims that I am an ``out-of-the-closet objectivist.'' Barry writes: >Is there a spatio-temporal reality outside the realm of schematic and abstract-generic targets? Lakoff certainly seems to believe that there is, for example when he writes approvingly of 'Regier's basic idea' 'that certain kinds of neural structures -- topographic maps and orientation sensitive cells -- combine in assemblies that compute what we have been calling "image- schemas"'. Presumably this 'inherent [neural, physical] reality' is not confined to the spaces between people's ears but extends outwards also to fill the whole universe. >Or perhaps Lakoff's general criticism of objectivism does not apply to physics and neurology? (Forgive me if this is a silly speculation, but I am new to this list.) Why, come to that, does the general criticism of objectivism not apply to work on metaphors and image-schemata? Why should just these domains (neural nets, the ways people use words) be privileged and all else be 'metaphysically suspect'? My Reply: First, I never used the term ``metaphysically suspect.'' The sentence that Barry claims I wrote is actually from Rick Grush's note that I was responding to. Though I agreed with the general gist of of what Rick was saying, I would never use such a term. Rick's the philosopher, not me. Indeed, I believe that what philosophers call metaphysics is largely metaphor. I won't try to defend that here, but Barry himself organized and attended a conference last summer where Mark Johnson and I spoke and Mark's paper made that case in great detail, looking primarily at Descartes' metaphysics and showing that Descartes's metaphysics was not merely metaphorical, but made use of common everyday conventional metaphors. I believe that science, in general, uses metaphor. I don't think there is anything wrong with that. Quite the opposite, I think that the metaphors that the ``successful'' sciences have come up with are great human achievements. Regier's modelling, for example, was done using the connectionist metaphor for neural systems, in which neurons are connectionist units, dendrites are connections, etc. This is an exteremly interesting and useful metaphor, which of course hides biological universes. But it is a metaphor that even a great many neuroscientists find useful. And I don't know of any comparably useful metaphor for neural systems. Why is this ``privileged'? It isn't. It's just the only game in town right now. Moreover, I took pains to point out that Johnson's phenomenological account was a necessary part of the whole explanatory story. In short, I am a scientist, and I am not an objectivist. Barry, being concerned with ``truth'' -- as he should be -- writes: >Perhaps what Lakoff means when he refers to the target domain as being 'only partially, perhaps minimally, structured on its own' is that the target domain is not a part of inherent spatio- temporal reality at all, but is rather something general and abstract. But this would mean that there are two sorts of schema and two stages of abstracting going on: a first, metaphor-free stage which takes us from genuine reality to this 'only partially structured' stuff, followed by a second stage where metaphors go to work....Two jobs need to be performed, then: one job of describing how the image-schemas relate to their abstract-generic targets, and a second job of describing how the abstract-generic targets relate to spatio-temporally reality itself. My reply: The question of how our conceptual system accords with reality is an important and interesting one. I have said all that I think I know about it in Women, Fire, and Dangerous Things, Chapter 17, starting on page 292. I would be happy to discuss that chapter. Finally, Barry writes: Can Lakoff give an example of what he calls a 'partial inherent structuring of the target domain which cannot be overridden'? I doubt it (given the the bloody-mindedness of speakers of natural languages). My reply: I gave one in my reply to Grush. The Conduit metpahor cannot override the inherent structure of the mental domain, part which includes: When one communicates an idea, one does not cease to have that idea. Example: I gave him that idea. This doesn't imply that I don't have it anymore. I have not found any case where the conduit metaphor or any other conventional conceptual metaphor for communication overrides this part of the inherent structure of the mental domain. I'm open to examples. Got any? From cogling-relay@UCSD.EDU Wed Jan 19 18:53:45 1994 Received: from ucsd.edu by cogsci.Berkeley.EDU (5.63/1.29) id AA18366; Wed, 19 Jan 94 18:53:39 -0800 Received: from localhost by ucsd.edu; id SAA25537 sendmail 8.6.4/UCSD-2.2-sun Wed, 19 Jan 1994 18:47:16 -0800 for cogling-list Received: from icsia.ICSI.Berkeley.EDU by ucsd.edu; id SAA25531 sendmail 8.6.4/UCSD-2.2-sun via ESMTP Wed, 19 Jan 1994 18:47:14 -0800 for Received: from icsib4.ICSI.Berkeley.EDU (regier@icsib4.ICSI.Berkeley.EDU [128.32.201.19]) by icsia.ICSI.Berkeley.EDU (8.6.5/HUB+V8$Revision: 1.19 $) with ESMTP id SAA13058 for ; Wed, 19 Jan 1994 18:47:11 -0800 From: regier@ICSI.Berkeley.EDU (Terry Regier) Received: from localhost (regier@localhost) by icsib4.ICSI.Berkeley.EDU (8.6.4/1.8) id SAA02099 for cogling@ucsd.edu; Wed, 19 Jan 1994 18:47:07 -0800 Message-Id: <199401200247.SAA02099@icsib4.ICSI.Berkeley.EDU> Subject: Image schemas and structure To: cogling@ucsd.edu Date: Wed, 19 Jan 1994 18:47:07 -0800 (PST) X-Mailer: ELM [version 2.4 PL21] Mime-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Content-Length: 2472 Status: R In the ongoing image-schema/metaphor discussion, George Lakoff made reference to the connectionist model which I built recently and described in my thesis. Here are my 2 bits on the topic: a quick intro to just what the model is, followed by my take on what its implications are for our purposes here. The model learns to linguistically classify perceived events, and can be trained on the spatial system of any of a range of natural languages. It has its limitations, a detailed listing of which I will spare you for the time being. For now, the model's relevance to discussions of image schemas and their nature lies in its structure. It exhibits a good deal of built-in adaptive "neural" structure, much of it motivated by neurobiological and in some cases psychophysical evidence. The important point is that in most cases it is possible to pinpoint some form of non-linguistic evidence for the structural device in question. Yet many of these devices wind up looking quite familiar from a cognitive linguistics standpoint. There is something which could be viewed as a container schema, something that could be viewed as a path schema, and so on. An example of non-linguistic motivation for a structural device which corresponds to a known image schema is apparent motion, which motivates the structure corresponding to the path schema. Apparent motion is a well-known although to my knowledge still incompletely understood psychophysical phenomenon in which a subject is presented with a visual stimulus at one location, then at another, and perceives not just the endpoints, but smooth motion from one location to the other. I.e. the subject is given the source and the endpoint, and perceptually infers the path. I take this as loose motivation for the tripartite (source, path, endpoint) structure of the structural device which can be interpreted as corresponding to the path schema. Other examples concern cells which are sensitive to orientations that do not correspond to actual luminance or color discontinuities in the image (can be taken as evidence for a structure corresponding to a simple above/below schema). Thus, the main two things I feel the model has to say about image schemas are: (a) Image-schemas can be thought of as adaptive neural structures embedded within a larger overall system. (b) When viewed this way, there are often non-linguistic sources of motivation for the structures/schemas. Terry From cogling-relay@UCSD.EDU Wed Jan 19 14:29:06 1994 Received: from ucsd.edu by cogsci.Berkeley.EDU (5.63/1.29) id AA16315; Wed, 19 Jan 94 14:29:00 -0800 Received: from localhost by ucsd.edu; id OAA12804 sendmail 8.6.4/UCSD-2.2-sun Wed, 19 Jan 1994 14:19:09 -0800 for cogling-list Received: from pitt.edu by ucsd.edu; id OAA12767 sendmail 8.6.4/UCSD-2.2-sun via SMTP Wed, 19 Jan 1994 14:18:44 -0800 for Received: by pitt.edu id AA06046 (5.65c/IDA-1.4.4.5 for cogling@ucsd.edu); Wed, 19 Jan 1994 17:18:41 -0500 Received: via switchmail; Wed, 19 Jan 1994 17:18:37 -0500 (EST) Received: from white.cis.pitt.edu via qmail ID ; Wed, 19 Jan 1994 17:18:07 -0500 (EST) Received: from white.cis.pitt.edu via qmail ID ; Wed, 19 Jan 1994 17:17:50 -0500 (EST) Received: from Messages.8.5.N.CUILIB.3.45.SNAP.NOT.LINKED.white.cis.pitt.edu.sun4m.412 via MS.5.6.white.cis.pitt.edu.sun4_41; Wed, 19 Jan 1994 17:17:50 -0500 (EST) Message-Id: Date: Wed, 19 Jan 1994 17:17:50 -0500 (EST) From: Jozsef A Toth Mime-Version: 1.0 Content-Type: text/plain; charset=US-ASCII To: cogling@ucsd.edu Subject: column detectors In-Reply-To: <9401162328.AA22157@cogsci.Berkeley.EDU> References: <9401162328.AA22157@cogsci.Berkeley.EDU> Status: R Excerpts from mail: 16-Jan-94 George Lakoff@cogsci.Ber (3310) > Terry Regier's thesis (for a copy write to > regier@icsi) gives some inkling of what image-schemas might be > from a neural perspective. His connectionist spatial model of > spatial relations computes image-schemas getting the topological > properties via models of topographic maps of the visual field > and the orientational properties via models of orientation-sensitive cells. 1. What is Terry Regier's full email address? 2. There is an interesting article in Science (Vol. 262, 29-Oct-93) entitled "Neuronal Mechanisms of Object Recognition" by Keiji Tanaka. It details primate probe studies and suggests single- and dual-cell evidence for columnar detectors that are "higher-order" than simple orientation selectivity cells. About 1300 (of what unit, I wasn't sure) such columns are suggested to exist in a region in the primate referred to as "TE"; which is directly connected with V1 (i.e., orientation selectivity). joe ---------------------------- Jozsef A. Toth University of Pittsburgh 600 Epsilon Drive Pittsburgh, PA 15238, U.S.A. (Voice) 412-624-6413 (FAX) 412-624-6436 INTERNET: jtoth+@pitt.edu From markt@umd5.umd.edu Wed Jan 19 06:25:55 1994 Received: from umd5.umd.edu by cogsci.Berkeley.EDU (5.63/1.29) id AA11054; Wed, 19 Jan 94 06:25:52 -0800 Received: from hamlet.umd.edu by umd5.umd.edu(8.6.4/1Dec93) with ESMTP id JAA24365; Wed, 19 Jan 1994 09:25:31 -0500 From: Mark Turner Received: from localhost by hamlet.umd.edu id JAA01049; Wed, 19 Jan 1994 09:25:30 -0500 Date: Wed, 19 Jan 1994 09:25:30 -0500 Message-Id: <199401191425.JAA01049@hamlet.umd.edu> To: lakoff@cogsci.Berkeley.EDU Subject: only game in town Status: R Your citation of Fodor's tag line to describe the cognitive linguistics project is rhetorically very witty and effective. Keep it up! While I was at San Diego, I introduced several people to Regier's thesis, and retailed some of the computations you presented when you spoke to the computer people at Maryland. Regier's work isn't sufficiently known, or understood, at San Diego. Very much looking forward to seeing you in the Bay Area next year. There is so much to catch up on. I am looking forward to spending a few weekends at least discussing with you the zillion examples of blending, and the astonishing intricacy of the theory of middle spaces. I should be able to send a draft of the Literary Mind before too long. This project has just taken me forever, although it is just a little book, meant to be popular and not to technical. Some of its chapters, especially the first four, date from late 1989! Time to get this out to the world of literary criticism and rhetoric, and to everybody else, too. From cogling-relay@UCSD.EDU Wed Jan 19 06:59:02 1994 Received: from ucsd.edu by cogsci.Berkeley.EDU (5.63/1.29) id AA11098; Wed, 19 Jan 94 06:58:55 -0800 Received: from localhost by ucsd.edu; id GAA03030 sendmail 8.6.4/UCSD-2.2-sun Wed, 19 Jan 1994 06:53:38 -0800 for cogling-list Received: from umd5.umd.edu by ucsd.edu; id GAA03027 sendmail 8.6.4/UCSD-2.2-sun via ESMTP Wed, 19 Jan 1994 06:53:36 -0800 for Received: from hamlet.umd.edu by umd5.umd.edu(8.6.4/1Dec93) with ESMTP id JAA27607; Wed, 19 Jan 1994 09:53:33 -0500 From: Mark Turner Received: from localhost by hamlet.umd.edu id JAA01402; Wed, 19 Jan 1994 09:53:32 -0500 Date: Wed, 19 Jan 1994 09:53:32 -0500 Message-Id: <199401191453.JAA01402@hamlet.umd.edu> To: cogling@ucsd.edu Subject: Anger, Heat, Blending, Lakoff, Elmer Fudd Status: R >There are other exciting things as well -- for example, blended spaces, >where the metaphor holds in the new space. Take a Bugs Bunny >cartoon where Elmer Fudd has smoke coming out of his ears. The >Anger Is Heat metaphor structures that imaginative domain. > >Or take a culturally important case, one discussed at length by Emily >Martin in her forthcoming book ``Flexible Bodies'': Many biologists >and doctors really believe the military metaphor for the immune >system. For them, the blended space in which the immune system >``fights off'' ``invading'' viruses is the only reality. At the October 1993 UCB-UCSD Cognitive Linguistics workshop, where Fauconnier and I presented the notion and the term "blended space," in a paper titled "Conceptual projection and middle spaces," we tried to make it clear that vivid literalizations of metaphor that do no central inferential work are misleading if thought of as the prototype of blending. Such examples can be misread as suggesting that blending is a secondary phenomenon, applied optionally to the products of conceptual metaphor, like paint to a house, for extra (spectacular) effect. They are perhaps the least interesting cases we know, although the most readily apprehended. Blending is a general instrument of cognition. It is often undetectable except on analysis. It often does central conceptual work. It often does central inferential work. It is not restricted to metaphor, but operates over the full range of instruments of conceptual projection, including non-metaphoric cases. It is a basic instrument of category extension, including culturally and scientifically important category extension (same-sex marriage, artificial life). Often what we know best will be the blended space (as in the case of the concept of the immune system above). Blending can be detected in cases of nonverbal action and action slips, so is not exclusively a linguistic phenomenon. And so on. George has not at all misunderstood, that we regard vividly literalized metaphor as blending, and he understands entirely (from our discussions at that workshop) that we view this as only one special case of blending, a case that does not manifest many of the most important powers of blending, but for the other members of this group, it might be useful for me explicitly to arrest and quash any suggestion that spectacular cartoon literalizations of metaphor are the prototype of blending. A fuller story is forthcoming (honest!) in the form of a technical report, in a few weeks. Cheers, Mark Turner From cogling-relay@UCSD.EDU Thu Jan 20 19:10:37 1994 Received: from ucsd.edu by cogsci.Berkeley.EDU (5.63/1.29) id AA27927; Thu, 20 Jan 94 19:10:33 -0800 Received: from localhost by ucsd.edu; id TAA14856 sendmail 8.6.4/UCSD-2.2-sun Thu, 20 Jan 1994 19:06:16 -0800 for cogling-list Received: from lanai.cs.ucla.edu by ucsd.edu; id TAA14852 sendmail 8.6.4/UCSD-2.2-sun via SMTP Thu, 20 Jan 1994 19:06:14 -0800 for Received: by lanai.cs.ucla.edu (Sendmail 5.61d+YP/3.23) id AA21436; Thu, 20 Jan 94 19:06:13 -0800 Date: Thu, 20 Jan 94 19:06:12 PST From: Dr Michael G Dyer Message-Id: <940121.030612z.21251.dyer@lanai.cs.ucla.edu> To: cogling@ucsd.edu Subject: "completely structured" and pointers to Regier-like research Status: R >My [Barry's] response: surely everything that is 'inherently there' must therefore also be completely structured, in its way. Even a lump of jello or a whirlpool or a conflagration is completely structured. An example of something that is 'inherently there' but NOT completely structured is a neural network whose weights are biased toward rapidly learning one class of mappings. Evolution, for example, might give an organism various biased networks, but experiences/interactions with the environment would be required to move the weights so that the knowledge would become properly 'structured' or manifest. By the way, I advised a PhD in the area of learning language by associating simulated image sequences with simplified phoneme sequences (that describe the image sequences). Here are pointers: Nenov, V. I. and Dyer, M. G. Perceptually Grounded Language Learning: Part 1 -- A Neural Network Architecture for Robust Sequence Association. Connection Science, Vol. 5, No. 2, 1993. Nenov, V. I. and Dyer, M. G. Perceptually Grounded Language Learning: Part 2 -- DETE: A Neural/Procedural Model. Connection Science, Vol. 5, No. 3, (to appear soon). Part 1 describes a more realistic neuron (termed a predictron) in which information shifts over time, from one dendritic compartment to the next, toward the soma (thus acting like a temporal delay line). Predictrons and other neural elements are organized to perform robust spatio-temporal associations. Here the organization of these predictrons is compared to the cerebellum and the rapid learning of sequences etc. is described. Part 2 describes an architecture (DETE) composed of modules made up of such neurons, in which separate, topographic maps are used and visual features (in regions of these maps) are bound to appropriate objects (in a scene) via phase-locking of neural firing. For example, the phrase "red ball moves diagonally up" is learned by showing examples of different red circular objects moving diagonally upward across the (simulated) visual screen. the system must learn individual words (e.g. "ball". "moves") first, then 2-word sequences, etc. before longer sequences can be learned. "diagonally" is represented as activation in diagonal regions of a map of neurons. there are maps for location, motion, color, shape, size and there are maps for motions (and zooming) of an eye and also motions of a simulated finger (for touching/pushing objects on the screen). The "schemas" formed are not anything like symbolic frames but are traces (stored in dendritic compartments) of invariant visual features. However, the DETE system cannot (as yet) learn anything abstract (like "owns") because it cannot see ownership (it can only see physical contact). Abstractions are not perceived directly, (so I'm willing to bet some set of "abstraction building blocks" will have to be innate within neural tissue of humans). -- Michael Dyer From cogling-relay@UCSD.EDU Thu Jan 20 06:31:36 1994 Received: from ucsd.edu by cogsci.Berkeley.EDU (5.63/1.29) id AA21566; Thu, 20 Jan 94 06:31:33 -0800 Received: from localhost by ucsd.edu; id GAA23019 sendmail 8.6.4/UCSD-2.2-sun Thu, 20 Jan 1994 06:27:35 -0800 for cogling-list Received: from HUGSE1.HARVARD.EDU by ucsd.edu; id GAA23013 sendmail 8.6.4/UCSD-2.2-sun via ESMTP Thu, 20 Jan 1994 06:27:33 -0800 for From: FISCHEKU@HUGSE1.HARVARD.EDU Received: from HUGSE1.HARVARD.EDU by HUGSE1.HARVARD.EDU (PMDF V4.2-12 #4716) id <01H7WF9LCD168WXUVQ@HUGSE1.HARVARD.EDU>; Thu, 20 Jan 1994 09:18:22 EDT Date: Thu, 20 Jan 1994 09:18:22 -0400 (EDT) Subject: ordered structures To: cogling@ucsd.edu Message-Id: <01H7WF9LCMOC8WXUVQ@HUGSE1.HARVARD.EDU> X-Vms-To: SMTP%"cogling@ucsd.edu" X-Vms-Cc: FISCHEKU Mime-Version: 1.0 Content-Type: TEXT/PLAIN; CHARSET=US-ASCII Content-Transfer-Encoding: 7BIT Status: R I would like to second most of Barry Smith's comments. The idea that the target domain in not structured very much does not seem to me to work. As one of the few developmentalists on this network, I would suggest that a lot of the answer is in ordered structures, with much of the ordering being evident in development. Thus a domain that is ordered in terms of, say perception-action structures (sensorimotor) could be restructured through higher-order representational structures. Children do this sort of restructuring all the time as they develop, and I think that adults do it too as they develop (learn, solve problems, extend their knowledge). Kurt Fischer From cogling-relay@UCSD.EDU Thu Jan 20 06:19:47 1994 Received: from ucsd.edu by cogsci.Berkeley.EDU (5.63/1.29) id AA21546; Thu, 20 Jan 94 06:19:41 -0800 Received: from localhost by ucsd.edu; id GAA22299 sendmail 8.6.4/UCSD-2.2-sun Thu, 20 Jan 1994 06:16:46 -0800 for cogling-list Received: from ubvmsb.cc.buffalo.edu by ucsd.edu; id GAA22292 sendmail 8.6.4/UCSD-2.2-sun via ESMTP Thu, 20 Jan 1994 06:16:42 -0800 for Received: from ubvms.cc.buffalo.edu by ubvms.cc.buffalo.edu (PMDF V4.2-14 #5889) id <01H7WF9N1XKO8Y4XCZ@ubvms.cc.buffalo.edu>; Thu, 20 Jan 1994 09:19:43 EST Date: Thu, 20 Jan 1994 09:19:43 -0500 (EST) From: BARRY SMITH Subject: Objectivism and Stuff To: lakoff@cogsci.Berkeley.EDU Cc: cogling@ucsd.edu Message-Id: <01H7WF9N1XKQ8Y4XCZ@ubvms.cc.buffalo.edu> Organization: University at Buffalo X-Vms-To: IN%"lakoff@cogsci.Berkeley.EDU" X-Vms-Cc: IN%"cogling@ucsd.edu" Mime-Version: 1.0 Content-Type: TEXT/PLAIN; CHARSET=US-ASCII Content-Transfer-Encoding: 7BIT Status: R I shall return to the big issues later. For the moment, why is the following not an answer to Lakoff's challenge to find an example of the conduit metaphor whose use in being passed on the ideas stay in the mind of the one who has them: Life was getting too hard. I needed to do something drastic before it was too late. Somehow I had to rid myself of all this philosophy and take up chicken-farming, which is what I was really cut out for. The young kid was really helpful. I gave him all my ideas, and I felt clean, real clean. Barry Smith From cogling-relay@UCSD.EDU Sun Jan 23 15:08:58 1994 Received: from ucsd.edu by cogsci.Berkeley.EDU (5.63/1.29) id AA15773; Sun, 23 Jan 94 15:08:53 -0800 Received: from localhost by ucsd.edu; id PAA23958 sendmail 8.6.4/UCSD-2.2-sun Sun, 23 Jan 1994 15:04:39 -0800 for cogling-list Received: from ubvmsb.cc.buffalo.edu by ucsd.edu; id PAA23955 sendmail 8.6.4/UCSD-2.2-sun via ESMTP Sun, 23 Jan 1994 15:04:36 -0800 for Received: from ubvms.cc.buffalo.edu by ubvms.cc.buffalo.edu (PMDF V4.2-14 #5889) id <01H814PJB8OW8Y5LEJ@ubvms.cc.buffalo.edu>; Sun, 23 Jan 1994 18:07:44 EST Date: Sun, 23 Jan 1994 18:07:44 -0500 (EST) From: BARRY SMITH Subject: The Big Issues To: cogling@ucsd.edu Message-Id: <01H814PJCB9U8Y5LEJ@ubvms.cc.buffalo.edu> Organization: University at Buffalo X-Vms-To: IN%"cogling@ucsd.edu" Mime-Version: 1.0 Content-Type: TEXT/PLAIN; CHARSET=US-ASCII Content-Transfer-Encoding: 7BIT Status: R 1. "Completely structured" and pointers to Regier-like research In response to my thesis that everything that exists must also be completely structured, Michael Dyer writes: >An example of something that is 'inherently there' but NOT completely structured is a neural network whose weights are biased toward rapidly learning one class of mappings. Could someone explain to me what is incompletely structured here? All the weights are just as they are, their biases are just as they are, the whole affair is deterministic. 2. Smith's Major Misunderstandings (comments on the debate between Lakoff and Grush) There is more agreement between Lakoff and myself than Lakoff presupposes, but I will continue to play devil's advocate for a bit longer, since I think also that there are fundamental flaws in Lakoff's project. Lakoff writes that he is "studying the system of concepts in terms of which we think and in terms of which knowledge is represented", pointing to basic results which tell us that human conceptual systems are structured in terms of conceptual domains. Already this word 'domain' should set warning-signals flashing, since 'domain' is a term of art that applies primarily to objects, to what is 'inherently there'. Here, though, it is being applied to subjects and their concepts, and with (deliberately?) confusing results. We see the confusion already when Lakoff writes: >The COMMERCIAL EVENT FRAME is independent of particular instances of commercial events, and is used to structure all particular instances. No it isn't. Particular instances are as they are. Frames structure our _knowledge_ of particular instances, or the ways we _talk about_ them. The same mistake appears again when Lakoff writes: >Len Talmy ... has done a great deal of work showing that space is a separate conceptual domain. No he hasn't. Space is space. There might be a separate family of spatial concepts. No family of concepts _is space_. One must, on the strength of these and related examples conclude that Lakoff is confounding two different things: (I) The world itself (e.g. commercial events) and (II) the conceptual system used to structure knowledge about the world (COMMERCIAL EVENTS). Lakoff writes: >Do the human concepts used in ``successful'' sciences fit the structure of the world in itself if such a structure exists? There is no way that I or any other human being could tell, because all that I and the rest of us have available is human concepts. We have no direct access to whatever ``objective'' structure there might be from a God's Eye View. Lakoff is making an error here that has been common since the beginning of idealist philosophy in the work of e.g. Kant and Berkeley.. (See, for a gloriously funny analysis of this error, the chapter "Idealism: A Victorian Horror Story" in David Stove's book_The Plato Cult and Other Philosophical Follies_, Blackwells, 1991). The error goes like this: A. The concepts used by scientists are human concepts. Therefore: B. we cannot know how reality is as it is structured in itself. B. does not, however, follow from A. Indeed, the very success of science in so many spheres (and not just 'engineering') suggests that we humans are in many domains pretty close to the God's Eye View. And if Lakoff himself did not think this about e.g. Regier's work, then he would not be so enthusiastic about it, after all. BS From cogling-relay@UCSD.EDU Mon Jan 24 07:24:19 1994 Received: from ucsd.edu by cogsci.Berkeley.EDU (5.63/1.29) id AA19570; Mon, 24 Jan 94 07:24:14 -0800 Received: from localhost by ucsd.edu; id HAA08357 sendmail 8.6.4/UCSD-2.2-sun Mon, 24 Jan 1994 07:19:51 -0800 for cogling-list Received: from ubvmsb.cc.buffalo.edu by ucsd.edu; id HAA08353 sendmail 8.6.4/UCSD-2.2-sun via ESMTP Mon, 24 Jan 1994 07:19:49 -0800 for Received: from ubvms.cc.buffalo.edu by ubvms.cc.buffalo.edu (PMDF V4.2-14 #5889) id <01H822OEB50Q8Y5NVA@ubvms.cc.buffalo.edu>; Mon, 24 Jan 1994 10:21:47 EST Date: Mon, 24 Jan 1994 10:21:47 -0500 (EST) From: BARRY SMITH Subject: Response to George To: cogling@ucsd.edu Message-Id: <01H822OEBENW8Y5NVA@ubvms.cc.buffalo.edu> Organization: University at Buffalo X-Vms-To: IN%"cogling@ucsd.edu" Mime-Version: 1.0 Content-Type: TEXT/PLAIN; CHARSET=US-ASCII Content-Transfer-Encoding: 7BIT Status: R More on the Great Issues: Not only do I own a copy of WFDT, but I read it, with great appreciation, almost as soon as it was published. I did not cite it in my last missive in order to keep the debate within reasonable limits, and because I agree with so much of what it contains, Moreover, the passages in this work which George mentioned do not, in fact, help to clear up the (my? his?) confusions between knowledge and object. These passages do however draw attention to a related problem in Lakoff's account of objectivism, namely his presentation of the only alternative to his 'experiential' account of truth as consisting in a >folk theory to the effect that: -There is one and only one correct way to understand reality. If truth is relative to understanding, and if understanding, according to our folk theory, is fixed, then (in that folk theory) then there is one and only one truth -- absolute truth. This is our normal folk theory of truth. All the folk I know, however, are perfectly clear in their minds both (1) that truth is an objective matter, so that no amount of understanding on our part will make it so if it ain't so, and (2) that there are many truths about even the simplest matter [the cat is on the mat, the mat is under the cat, the cat is exerting pressure on the mat, the four-legged living organism is stationary upon the sisal, etc. etc.]. There are, then, many ways to slice the cheese of reality, some of them far more interesting than this, all of them (if we're careful) a matter of what is objectively true. Lakoff thinks you can have (2) - with all the interesting business of metaphors which go along therewith - only if you reject (1). This is, again, a typical trait of idealism. Lakoff is right when he says that people are all the time finding themselves in a situation of the sort >where they can see that there are two or more equally plausible understandings ... What is not the case (contra Lakoff) is that they then standardly jump to the conclusion >that truth can be relative to those understandings. On the contrary, the existence of situations with different understandings would be a problem for the objective concept of truth only if there were understandings of the same situation which conflict in a strong (logical) sense. Yet (as I understood it) Lakoff's point about the incomplete structuredness of 'what is inherently there' was designed precisely to avoid such conflict. Lakoff is right to insist that it is a major problem for objectivist metaphysics (i.e. for the position of right-thinking ordinary folk since Aristotle, a position which takes truth to be an objective matter) to provide an account of the truth of sentences about humanly-created realities such as `wasted time'. In fact I believe his work, and that of Johnson, Talmy, etc. can be of great value in giving such an account. But the whole business becomes far too easy if one draws back from _this_ challenge and seeks instead an account of how such sentences can be merely 'experientially true'. At this point in his argument Lakoff writes that talk of 'wasting time' (and the like) >makes sense on an experientialist account of truth; it makes very little sense on an objectivist account of truth. Many of our most important truths are not physical truths, but truths that come about as a result of human beings acting in accord with a conceptual system that cannot in any sense be said to fit a reality completely outside of human experience. Does Lakoff really believe that all defenders of objective truth are physicalists? The Gestalt psychologists, for one, were passionate defenders of objective truth, yet produced wonderful theories about the human-created reality towards which many of our objectively true believes relate. Lakoff writes: >Since we act in accord with our conceptual systems, and since our actions are real, our conceptual systems have a major role in creating reality. Where human action is concerned, metaphysics (that is, our view of what exists and is real) is not independent of epistemology (in the broad sense of human understanding and knowledge). This is right, but again, it does not imply any challenge to objective truth: the fact that a given area of reality is human- created does not imply that mere human understandings of this reality are all we can have in the way of truth. On the contrary, if we follow Lakoff and >understand a statement as being \fItrue\fR in a given situation if our understanding of the statement fits our understanding of the situation closely enough for our purposes then all sorts of clearly false statements (e.g. that it is vital for the health of the Aryan race to exterminate Jews) will turn out to be true ('experientially') for certain folk. (This concentration camp example has of course been used against relativistic views of truth many times before.) Lakoff writes >truth cannot be characterized simply as correspondence to a physical reality, we must recognize truth as a human concept, subject to the laws of human thought. No serious defender of objective truth ever characterized truth in this fashion, to my knowledge. Lakoff's problem, then, is that his objectivist opponent is a straw man. Or to put the matter in another way: there are lots of interesting positions between Lakoffian experientialism and his straw-man physicalistic objectivism, positions which Lakoff chooses not to recognize. Your move, squire. Reply to Barry Smith: Let me start by thanking Barry for his patience and forbearance, as well as the for $1.25 royalty on Women, Fire -- which will be repaid with interest when I next get to buy him a caffe latte. I think the difference in world view behind this misunderstanding is about to become clear. The difference is in what counts as The Big Issue. For me, the Biiiigg Issssue is the nature of thought and language. That is an empirical issue. Relative to that, philosophical theories of truth become small, because they are constrained by the answer to the empirical question of what thought and language are like. That is what Johnson and I have been saying since Metaphors We Live By, and I'll go over the argument again given Barry's comments as a taking off place. But before I get to that, I owe Barry a response to a previous posting. Barry wrote: >For the moment, why is the following not an answer to Lakoff's challenge to find an example of the conduit metaphor whose use in being passed on the ideas stay in the mind of the one who has them: >Life was getting too hard. I needed to do something drastic before it was too late. Somehow I had to rid myself of all this philosophy and take up chicken-farming, which is what I was really cut out for. The young kid was really helpful. I gave him all my ideas, and I felt clean, real clean. This is, as any metaphor analyst will recognize, an example of the theory, not a counterexample to it. Let us start by looking at what is involved in it. First, there are two possible understandings. (1) ``Ideas'' means questions that the speaker is trying to answer or vague ideas that need to be fleshed out. Both of these involve a responsibility. The speaker has ``unburdened'' himself of the responsibility. That may or may not mean that he no longer has the ideas; in the most normal reading it only means that he no longer has the resonsibility. However, one could imagine a novel in which he no longer had the ideas. (2) One could imagine another novel in which (unlike the real world) it would be possible to tell someone everything one knows about a subject matter where one has rich knowledge, and then no know that subject matter any more. Now both readings involve nonnormal communication. In the protoypical case, when you tell someone something, what you tell them is relatively short and you still know it immediately afterwards. But there are real nonnormal cases of communication. For example, suppose you have an office next to mine and I need to tell you a phone number, and I keep repeating it to myself to remember it, and I go right into your office and tell it to you -- and then cannot remember it anymore. I think that is a real case -- so references to novels that contradict reality are unnecessary. Now let us return to the issue that this case was to address -- target domain ``overrides'' -- as I said, the term ``target domain accommodation'' would be better terminology. ``Give'', in the soruce domain characterizing object transfer in space, entails that the give no longer has the object after he gives it to someone else. In cases where this is consistent with ones knowledge of the situation at hand -- as in Barry's case or the telephone number case -- there is no ``override'' or ``accommodation'' to the target domain needed. The issue of overrides doesn't arise. Only in cases where the structure inhering in what you know about the situation would contradict a complete mapping does the issue arise. In such cases, the invariance hypothesis claims, knowledge about the subject matter at hand can override conflicting parts of the mapping. This occurs in the case of normal communication. In short, Barry's example was a simple-minded example of the theory in action. Another common example of ``accommodation'' or ``overrides'' is a sentence like ``I gave him a punch'' where the he does not have the punch afterwards. Knowledge about actions -- that they do not exist after they occur (though their effects night) -- takes precedence over the mapping of part of the source domain knowledge schema, namely, that if I give you an object, then you have it immediately afterwards. Turner's earlier posting gives lots of cites from his books where he discusses the function of the invariance principle and the workings of overrides. Now let us get back to the issue of what the Big Issue is. The question of what counts as the Big Issue is, as I understand, the question what priorities and commitments one has. As a linguist and cognitive scientist, my primary commitment is to study the nature of thought and language. This is an empirical enterprise. What Johnson and I claim is (1) that the answer to this empirical enterprise affects what possible theories of meaning and truth are consistent with the results of this enterprise. We claim that only an experientialist theory is consistent with the empirical results coming out of cognitive science and cognitive linguistics. We do not START with an experientialist position. Rather it is the only one we can maintain consistently given what has been discovered about language and thought. Conversely, we note that certain views about meaning and truth -- namely, objectivist theories -- are inconsistent with the empirical results about the nature of thought and language. On the other hand, many philosophers start by saying that they are concerned primarily with the Big Issue of meaning and truth and assume that they can get that straight first and then go on to study the relatively small empirical issue of what thought and language is like. In doing so, they commonly make on or another of the objectivist assumptions. What we note is that such a procedure will contradict empirical results about the nature of thought and language. What tupically happens is that they wind up ignoring most of language and thought, because it doesn't fit their preconceptions that come out of their objectivist assumptions. Now there is a reason why I spend so much time in Women, Fire distinguishing between Basic Realism and Objectivism (see the index -- I start on p. 158). Here's how I begin: Basic realism involves at least the following: -A commitment to the existence of a real world, both external to human beings, and including the reality of human experience. -A link of some sort between human conceptual systems and other aspects of reality. -A conception of truth that is not merely based on internal coherence. -A commitment to the existence of stable knowledge of the external world. -A rejection of the view that `anything goes' -- that any conceptual system is as good as any other. Objectivism, as I will describe it, is one version of basic realism. Experientialism is another. In other words, basic realism grants that there is world and it is the way it is. Now as a linguist and cognitive scientist, I am a basic realist about the nature of language and thought: they are real and there is a way they are and it is our job, using thought and language themselves, to characterize the way they are as well as possible. This does not commit me to objectivism. Objectivism adds a lot to basic realism -- all the stuff I discuss in chapter 11: an objectivist metaphysics and epistemology. The metaphysics includes a division into determinate entities with properties and relations holding among them at any time. It includes ``natural kinds'' -- classical categories defined by necessary and sufficient conditions out there in the world. Even Quine assumes all that as a precondition for his ``ontological relativity.'' (See his ``ontological Relativity and Other Essays, pp. 117-118.) Objectivism further assumes a theory of meaning that excludes such aspects of imaginative cognition as metaphors, frames, most prototypes, etc. Basic realism is neutral (says nothing at all) about whether the world in itslef has a structure or if so, what kind it has. Linguistics and cognitive science DISCOVER that the aspects of the world refered to as human language and thought do have a structure. They discover it by asking what generalizations, if any, there are over various kinds of phenonema and by experimental methods of various sorts. Starting with Basic Realism, we DISCOVER that objectivist epistemology and theories of meaning are false. Biology has discovered that objectivist metaphysics is false. (See WFDT, chapter 12.) And for a philosopher's argument against objectivist metaphysics, see Putnam's ``The Many Faces of Realism'' (pp. 32 ff.) Before going on to Barry's remarks, we should say something about the words ``true'' and ``truth.'' The question of what speakers of English mean by the words ``true'' and ``truth'' is an empirical question in linguistics -- in the branch called lexical semantics. But many philosophers ignore the empirical nature of the question and go about constructing, without empirical study, ``a theory of truth,'' in which the meaning of the word is taken for granted. Objectivist philsophers, in addition, make objectivist assumptions of various sort -- both objectivist assumptions about metaphysics and objectivist assumptions about epistemology and meaning. As a linguist and cognitive scientist, I have different commitments and priorities. I start by look at empirical results about language and thought, which yields all the results about metaphor, frames, cognitive grammar, image-schemas, etc. etc. Then, starting with the disposition of these Big Issues, I turn to the secondary question of what understanding of truth could be consistent with these results. I arrive at the conclusions presented in Women, Fire. Now the priorities I have -- study language and thought first -- are not arbitrary priorities. They are the only ones that make sense. The reason is that in order to make any claim at all about something being true, you must make that claim in language using a human conceptual system. All the data about truth -- data about the pairing of language and the understanding of language with data about how the world is understood -- presupposes an understanding of the human conceptual system and of language. Now it could be that, given the empirical results about thought and language, my conclusions are false -- empirically false. But to argue that, one has to start with the empirical results. Now the empirical ``results'' may be wrong. But to argue that, you have to go over the evidence and converging methodologies used to arrive at the results. If we agree on this, we can go on to discuss your reply to my chapter in Women, Fire. But I have a feeling that this is the point of disagreement. Am I right? George Addendum to Reply to Barry Smith: I think I can characterize the issue more sharply: Johnson and I claim that the empirical study of human conceptual systems, human reason and natural language is a prerequisite to any philosophical ``theory of truth''. There are two reasons: Reason 1: A `` theory of truth'' -- if is to be a theory at all must be about something, that is, it must account for data. What kind of data is relevant? Presumably, judgements by people that certain statements do or do not accord with certain situations. Thus, given a situation and a statement about it, would folks respond ``That's true'' or ``That's false'' or ``You could look at it that way.'' or ``That's misleading for the following reason: ...'' and so on for many other types of appropriate response. In such cases, each sentence in the data set for the theory must be understood -- and understood in terms of the conceptual system of the speaker. Similarly, each situation must be comprehended in terms of the perceptual and conceptual systems of the speaker. Both on the language side and the situation side, conceptualization relative to the speaker's conceptual system enters the picture where there is data from human beings about whether a statement is taken as true in a situation. And that his the only data that we have or can have about the matter. Reason 2: A ``theory of truth'' must be a theory. A theory of some subject matter mst be constructed by human beings using their human conceptual systems. What counts as a theory is an empirical issue. The data for the theory of what a theory is would include at least all the cases of scientific theories, e.g., evolutionary biology, contemporary physics, genetics, etc. That would require understanding theory construction in cognitive terms -- exactly what human conceptual resources go into a theory. And such a study presupposes a knowledge of what human conceptual systems are like. For example, Ron Giere has recently shown that classical physics uses image-schemas, basic-level categorization and radial categories. The moral: Understanding what a theory is requires empirical study of conceptual systems. Thus there are two reasons why the empirical study of conceptual systems and language is logically prior to a ``theory of truth'': First, because the data about truth requires an understanding of what conceptual systems are like, since conceptual systems nter into both the understanding of sentences and situations. Second, because an understanding of what theories are also requires an understanding of what human conceptual systems are like. Again, if Barry agrees to this, we can proceed to discuss his objections. But I have a feeling that this is the sticking point. The disparity, I believe, comes from the common philosophical assumption that human reason is transcendental and universal -- and that formal logic correctly characterizes transcendental and universal reason. The only ``ontological relativity'' -- the only human contribution -- is of the trivial Quinean sort. Here is the Quinean mythology: The world already comes with many classical categories (natural kinds) and with objects that have properties and stand in fixed relation to one another. Those properties and relations, and the classical categories defined by natural kinds, is what philsophers mean by the ``structure in the world.'' Those properties and relations in the world define potential classical categories. The only human contribution is to choose which of these potential categories to include in our conceptual systems and mark in our language. The potential categories are already implicit in the structure of the world -- which, by assumption, comes completely structured. We just ``carve up reality'' by picking out aspects of the structure inherent in the world. On such a view, the only empirical question is what categories particular languages ``carve out.'' This study is called ``naturalized epistemology'' -- the ``natural'' part is about natural languages and the ``epistemology'' part concerns the ``carved out'' out categories. Given such a view, one can construct a theory of truth in general without knowing what the particular classical categories carved out of the world by a language happen to be. No matter what they are, they will fit the structure of the world, since the properties and relations that characterize them are guaranteed by assumption to be in the world. Moreover, If reason is just logic, then that can be taken for granted, since both the world and any mere human conceptual system must both be structured by logic. On such a view, sentences can just be fit to the world, because the contributioons of conceptual systems is just irrevlevant by definition. And on this view, theories are sets of axioms in some logic, so no serious empirical study of conceptual systems is needed to discover what theories are like. What is wrong with this picture? Everything, as everyone on this net should know. The discoveries of cognitive linguistics and second generation cognitive science in general contradict the entire picture: Reason isn't logic. Conceptual categories for the most part aren't classical. Reason isn't transcendental, but comes out of the body. Conceptualization isn't just the carving up of a previously structured world: it involves the imposition of imaginative structure -- image-schematic, metaphorical, prototypical, and frame semantic -- and it involves not merely imposing mental spaces, but blended mental spaces. Once one knows this, then the empirical study of conceptual systems and language becomes a serious prerequiste to a theory of truth, for the reasons given above. From jfeldman@ICSI.Berkeley.EDU Wed Jan 26 17:49:58 1994 Received: from icsia.ICSI.Berkeley.EDU by cogsci.Berkeley.EDU (5.63/1.29) id AA18634; Wed, 26 Jan 94 17:49:53 -0800 Received: from icsib60.ICSI.Berkeley.EDU (jfeldman@icsib60.ICSI.Berkeley.EDU [128.32.201.114]) by icsia.ICSI.Berkeley.EDU (8.6.5/HUB+V8$Revision: 1.19 $) with ESMTP id RAA01580 for ; Wed, 26 Jan 1994 17:49:25 -0800 From: jfeldman@ICSI.Berkeley.EDU (Jerry Feldman) Received: from localhost (jfeldman@localhost) by icsib60.ICSI.Berkeley.EDU (8.6.4/1.8) id RAA02577 for lakoff@cogsci; Wed, 26 Jan 1994 17:49:21 -0800 Date: Wed, 26 Jan 1994 17:49:21 -0800 Message-Id: <199401270149.RAA02577@icsib60.ICSI.Berkeley.EDU> To: lakoff@cogsci.Berkeley.EDU Status: R George, Your megalomania seems to have burst its boundaries again. In your recent reply to Barry Smith you say, inter alia: > 1) Presumably, judgements by people that certain statements > do or do not accord with certain situations....In such cases, each sentence in > the data set for the theory must be understood -- and understood > in terms of the conceptual system of the speaker. > 2) That would require > understanding theory construction in cognitive terms -- exactly > what human conceptual resources go into a theory. And such a study