First Speaker: George Lakoff

Course Syllabus

Further Information

Invited Speakers

Article Summaries

Listserv Archive

The Neural Theory of Language Project


blending and conceptual integration


Questions and Summaries of George Lakoff's chapters

1) The Embodied Mind chapter three from Lakoff and Johnson: Philosophy in the Flesh

(discussed by Vanessa Gack)

Lakoff and Johnson (L&J) take issue with the Western philosophical theory of faculty psychology--that reason is seen as separate and independent o f what we do with our bodies. Citing evidence from cognitive science, L&J aargue that in some cases," rational inferences can be computed by the s ame neural architecture used in perception or bodily movement."

version I ("trivial"): conceptual structures are neural structures becaus e they are realized neurally

version II (strong): " An embodied concept is a neural structure that is actually part of, or makes use of, the sensory motor systems of our brai ns so that much of conceptual inference is sensory-motor inference" (p.7 ).

Some Questions:

1) What do we think of the assumptions which frame their inquiry?

a) categorization is a consequence of how we are embodied

b) categories, concepts and experience are inseparable

c) the architecture of your brain's neural networks determines what con cepts you have and hence the kind of reasoning you can do.

d) Whenever a neural ensemble provides the same output with different i nputs, there is neural categorization

e) a small % of categories are a consequence of conscious acts-- rather, through experience, categories are subject to unconscious reshaping and partial change

f) what we call concepts are neural structures that allow us to mentally reason about them

2) What do we think of the linguistic evidence that they cite in support of their argument?

a) color concepts

L&J say that our brains and bodies have evolved to create color out of f our factors: 1) wavelenghts of reflected light; 2) lighting conditions; 3 ) the three color cones in our retinas which absorb three wavelenghts of light; 4) the neural cirquity conncected to the cones. Thus, color is no t an internal representation of external reality -- but a function of the interaction between our brains' ability to make sense out of different l ighting conditions and surfacce reflectances. Colors are not substances or things. Therefore: we lose the corresponence theory of truth (sign3D external reality).

b) basic level categories ( eg chair in the middle of category hierarchy "furniture-CHAIR-rockingchair"). The basic level is the highest level at which 1) we have mental images th at stand for the entire category; 2) category members are regognized by p erception of overall shape; 3) we have motor programs for interacting wi th objects in the category; 4)most of one's knowedge is organized. L&J a rgue that the difference between basic-level categories and non-basic lev el categories is body-based; we interact at that level with the environme nt; consistent with part-whole structure, this level makes sense to us . (Note that they also apply basic-level categories to actions like swimmin g, emotions like anger...)

c) spatial relations concepts DEPEND on the body. They "characterize wha t spatial form is and define spatial inference" (p.23) (in front of, in b ack of...) " They have an "image schema, a "profile; and a "trajector-la ndmark structure."

c1 container schemas (bounded regions in space) highlight the interior of the schema, identify the boundary of interior as landmark and the obje ct overlapping the interior as trajector e.g. "sam is in the house," wher e house3Dlandmark relative to Sam the trajector (note: container schemas are cross-modal and can be applied to something we see, hear...)

c2 source-path-goal schemata have "a trajector that moves, a starting poi nt, a goal (destination), a route from source to goal, a trajectory of mo tion, the position of the trajector at a given time(t), the direction of trajector at (t), and final location of trajector. they depend on spatial logic:e.g. toward, away, through...

c3 bodily projections- we project our fronts and backs on to objects. W hile English is relatively impoverished (using terms like in front of, be hind...) other languages project bodily position as the primary means wit h which to characterize spatial relations. (e.g. in Mixtec, instead of s aying "he is on top of the hill" one says "he is located head hill."

3) What do we think of the neural modeling evidence that they cite in sup port of their argument?

Given spatial relations concepts, concepts of bodily movement (e.g. grasp , pull) and aspectual concepts that structure actions or events (e.g. sta rting, completed) Reason does not appear to be separate from perception and bodily movement. How might this work?

The Neural Theory of Language (NTL) group have been trying to make neural models of embodied cognition-- particularly language acquisition/use, th us linking neural computation with the forms of COMPUTATION required by l anguage AAND thought. They have five levels of methodology (35-37)

Level 1. Cognitive Science and Cognitive Linguistics

Level 2. Neurally Reducible Conventional Computational Models

Level 3. Structured Connectionist Models

Level 4. Computational Neuroscience

Level 5. Neuroscience

The link between Level 4 and Level 5 is given by computational neuroscience:

I'll only tersely summarize these experiments and let Lakoff do the expla ining.

a) NTL spatial relations learning task-Regier's Model R's model is hybrid, consisting of two parts:1) a connectionist model of neural structures; simulating the visual system whose job is characteriz e spatial relations & make distinctions; 2) a PDP connectionist model for learning via back-propogation. According to L&J, this model's significan ce is that "the fundamental CONCEPTUAL ROLES for making the right lingui stic distinctions among the vers are PLAYED by features of the MOTOR SYST EM.

b) NTL verbs of hand motion learning task-- Bailey Bailey used a computerized model of the body, an assumed collection of mo tor synergies(motor actions like pivoting the wrist, tightening a grip) a nd Petri nets to execute the motor schemas.Bailey videoed the computerize d body performing the hand movements, then had informants label the move ments with verbs.Then, Bailey made a learning mechanism that would learn the verbs from the computerized agent making the movements Results: on the sample of 18 English verbs, the recognition rate was 78% and the command-obeying rate 81%

c) NTL motor control and abstract aspectual reasoning task--Narayanan In this task, Narayanan constructed a neural (mappable by known methods o n to a neural model) theory of metaphor that used bodily movements to met aphorically discuss international economics (eg France falls into a reces sion). L&J argue that since the link between physical and economic domai ns could be modeled by neural connections--this work acts as an "existenc e proof."

4) What do we think of L&J's characterization of the relevant philosophic al issues? (p.8-9). And, to what extent does their evidence address their philosophical concerns?

a)categories exist independent human brains, bodies ; the relationship am ong categories is characterized by transcendent/ universal reason. Thus h uman reason is disembodied.

b)human concepts are the concepts of transcendent reason; categories are fixed and objective.

c)reason is culture-free and what makes us human has nothing to do with t he social or material world.

5) What are the implications for a theory of mind and how might this work impact the field of cognitive science?

a) human reason is a form of animal reason-- we might learn about embodied human reason through our study of embodied animal reason ?

b) human reason can be studied in neural network models ?

c) at last, evidence of embodied cognition-- faculty psychology is dead ?


While I felt comfortable with the strong version of L&J's argument, ["An embodied concept is a neural structure that is actually part of, or makes use of, the sensory motor systems of our brains so that much of concept ual inference is sensory-motor inference" (p.7).] I was not convinced b y the trivial version (conceptual structures are neural structures becaus e they are realized neurally). I do not understand the idea that-- becau se conceptual structures are realized neurally that they are neural struc tures. Neural structures perform a certain ffunction, but do they compri se that function? We say, for example, that respiration is realized by the lungs-- but we d o not say that lungs are respiration-- respiration is accomplished via th e lungs, alveoli, diaphragm, intercostal mmuscles etc. There is plastici ty in this system (when we can not breath through our nasal passages, we can breath through our mouths). Neural structures also appear quite plas tic, but I do not understand them to be interchangeable. As a consequence of my own limited understanding, I failed to ground the following assumptions:

c) the architecture of your brain's neural networks determines what conc epts you have and hence the kind of reasoning you can do.

f) what we call concepts are neural structures that allow us to mentally reason about them

This one, however, seems plausible enough:

d) Whenever a neural ensemble provides the same output with different i nputs, there is neural categorization

And I was satisfied with their characterization of mainstream Western phi losophy, but then again, I'm not a philosopher. Philosophers? Thus, I b ought these assumptions:

a) categorization is a consequence of how we are embodied

b) categories, concepts and experience are inseparable

e) a small % of categories are a consequence of conscious acts-- rather, through experience, categories are subject to unconscious reshaping and partial change

Re: evidence

I found the linguistic evidence quite interesting and convincing-- of emb odied reason, but not of concepts as neural structures. I found the compu tational evidence fascinating and plausible,, but not convincing.

I still believe that there is an enormous gap between the physical brain and the level of human concepts and language; we may need more than rese arch methodologies already in place within cognitive science to bridge th at gap. I, myself, am not convinced of how to link levels 1 and 2 or 1 a nd 5 I am looking forward to tomorrow's talk and hoping that in addressing my questions, he will convince me...

Philosophy in the Flesh

Lakoff & Johnson

Chapters 4 & 5

summary by Jeanne Milostan, Computer Science & Engineering


Chapter 4: Primary Metaphor and Subjective Experience

We saw in Chapter 3 that there is much evidence that, humans being evolved creatures, reasoning (rational thought) and categorization are inherently embodied. Chapter 4 continues the ideas, presenting evidence and arguments for the idea that human reasoning about subjective experiences is also embodied. Hence, we use our experience from sensory-motor domains to make subjective judgments and decisions within non-physical domains.

The use of sensory-motor knowledge to frame our views of abstract domains is inherent *in the way we think about them* in addition to the language we use to describe them. Lakoff & Johnson present a framework for why these conceptual metaphors are so widespread:

The Integrated Theory of Primary Metaphor

A) Conflation: early in our experience sensory-motor and subjective experiences tend to occur together and thus the concepts are formed in conjunction. As children gain more experience, the two domains can be distinguished, but the associations remain. I.e. More Is Up: Levels go up, not down, as we add more. (Recall the problem children have with conservation re: Piaget)

B) Primary Metaphor: These early associations form "primary metaphors" from which more complex metaphors can be built.

C) Neural Theory of Metaphor: The associations made by conflation are the result of concurrent activation in separate domains; this simultaneous activity results in permanent neural pathways being learned between the two areas. (Hebbian learning on a meta-scale?)

D) Conceptual Blending: Different domains are coactivated and connections between them can be formed, leading to both on-the-fly and culturally conventionalized activations. This is in addition to the associations formed during conflation, and accounts for the conventionalization of more complex metaphors.

The rest of the chapter fleshes out each of these points in more detail and with examples. Several points fall out of this framework:

Primary Metaphor (More Is Up) in inevitable. This is a result of our being human and interacting with the environment. These should also account for some of the universal concepts seen within widely disparate cultures.

Metaphors are Cross-Domain Mappings. This chapter concentrated on mappings from sensory-motor experience to making subjective judgments.

Can we think without metaphor? Yes, but it is difficult to do so without conscious effort:

I was blindsided by the flu. What I mean is the flu hit me when I wasn't looking. But no, that's still metaphorical. I caught the flu... wait a minute... I became ill with the flu when I wasn't expecting it.


Chapter 5: The Anatomy of Metaphor

Building on the bases laid down in previous chapters (reasoning is embodied, and primary metaphors are inevitable), chapter 5 explores how complex metaphors are constructed and why specific metaphors can be so pervasive within a culture.

Examining the metaphor A Purposeful Life is a Journey, Lakoff & Johnson go into great detail about how cultural beliefs (People are supposed to have purposes in life) combine with primary metaphors (Action Is Motion and Purposes are Destinations) and facts (A long trip with a series of destinations is a journey) to form pervasive metaphors: A Purposeful Life Is A Journey

Lakoff & Johnson then spend quite some time exploring the implications of this metaphor. Specifically, they explain how our common-sense reasoning about the "journey" domain allows us to then reason automatically about the "life journey" domain. They look at how these metaphors are not just figures of speech, but reasoning tools and bases by which we make everyday decisions (this in particular is covered extensively in L&J's other works). They look at how these complex metaphors, especially when conventionalized to the extent that people hardly even realize that they are metaphors, can then be used to construct even more complex metaphors (Love Is A Journey) which again are used as the basis for reasoning.

As summarizer, I find that the discussion presented in Chapter 5 seems to follow from the framework laid out in Chapter 4, supported again by the other works of the authors and their collaborators. The main weight of the argument seems to then lie on that framework, particularly part A (conflation) and C (Narayanan's neural theory of metaphor, from Chapter 3). The evidence for these two points are what I feel need to be more thoroughly explored.

| Syllabus | Speakers | Article Summaries | xxxxx | xxxx |

Contact: Gilles Fauconnier
Web Design: Omar Alhassoon