James R Hurford Language Evolution and Computation Research Unit, - - PowerPoint PPT Presentation

PROTO-HUMAN COGNITION IN NON-HUMAN ANIMALS

James R Hurford Language Evolution and Computation Research Unit, University of Edinburgh

Descartes

Animals as automata, and humans as automata+Reason

Darwin

Humans are evolutionarily related to apes and all other animals.

Morgan's Canon

“In no case is an animal activity to be interpreted in terms of higher psychological processes, if it can be fairly interpreted in terms of processes which stand lower in the scale of psychological evolution and development.”

Anthropomorphism versus continuity

• object permanence,
• episodic memory,
• metacognition,
• competence with abstract relations,
• transitive inference,
• subitizing,
• the where/what-dorsal/ventral separation,
• global and local attention.
• `frame-of-reference' systems,

Seeds of language: precursors of aspects of language in animal life

Displaced reference

Human language enables us to refer to things distant in time and space. No animal communication system allows the animal to refer to distant things, (except honeybee communication). But many animals can solve object permanence tasks.

Species Visible/invisible How long? Domestic chicks Neither Squirrel monkeys Visible only Cotton-top tamarins Both Parrots Both Dogs Both Up to 4 mins Chimpanzees Both Overnight Humans over 3 Both Very long times

Relative performance on object permanence

A solipsistic, private view of the origins

• f reference.

Private attention to objects (visible or hidden) is the basis for later-evolved public reference to

• bjects (e.g. via language).

Animals’ abilities in object permanence tasks are the evolutionary seed of displaced reference in language.

Episodic memory (mental time travel)

Humans describe specific events, often in the distant past (or future). Tulving (and Suddendorf) claims that episodic memory is unique to humans. But there are degrees of ‘episodic-like’ memory. ‘Episodic-like’ memory in animals is usually restricted to the domains animals are interested in (e.g. food).

Clayton et al.: Scrub jays remember where they hid food, of what type and how long before. Schwartz & Evans: apes have episodic-like memory. MacDonald: Gorillas could remember where food was hidden 24 hours later.

Schwartz: the gorilla King could remember the order in which he had been given food items, 5 minutes apart. (Backwards only!) Schwartz: the gorilla King could remember which of three possible events, which had no lasting effect, he had seen up to 15 minutes earlier. Menzel: The chimp Panzee could remember the hiding of food several days later.

Episodic memory – a methodological problem

Did Panzee recall the hiding of food (episodic memory)

• r where food lies hidden (semantic memory)?

How could we tell the difference? Humans are “inferentially promiscuous” (Susan Hurley) Possible experiment: (1) Prominently show a chimp a key being placed on a high shelf – this should be a salient event, but irrelevant (yet) to the chimp’s normal life. (2) Later, train the chimp to connect the key to the unlocking of a food cupboard; (3) one day, appear to have lost the key, and see if the chimp ‘tells’ you it is on the shelf.

Metacognition

Humans express propositional attitudes to states of

• affairs. E.g.

I (don’t) know/believe/think/hope that this is correct. “When an organism knows what it knows, its actions are different from an organism that is locked out of its library of knowledge” (Marc Hauser) Uncertainty monitoring: Smith and Washburn: Pigeons and chimpanzees trained to respond `Yes’, `No’ or `Don’t know’. The ‘Don’t know’ response typically takes longer than the more certain responses and is given at the borderline between categories. Keddy-Hector et al. Piglets backing out of a maze when they `realize’ they have made a wrong choice.

Animals form perceptuo-motor categories Only apes and humans can use these categories in further learning. Reversal learning experiments

Training regime Prosimians Great apes First session: Stimulus-1 > reward Association of Formation of Concept Stimulus-2 > nothing Stimulus-1 with reward. of Stimulus-1, and . association with reward. Second session: Stimulus-1 > nothing Association of Application of negation, Stimulus-2 > reward Stimulus-2 with reward. or oppositeness to . (Dicult unlearning Concept of Stimulus-1 . and relearning.) NOT( Stimulus-1 ) . associated with reward. . (Relearning facilitated by . use of acquired concept.)

Reversal Learning

Animals can represent abstract properties and relations Alex the parrot. What colour is this? Red RED(x) COLOUR(RED) What’s same about these? Shape. SQUARE(x), RED(x), PLASTIC(x) SQUARE(y), BLUE(y), WOOD(y) SQUARE(z), GREEN(z), GLASS(z) SHAPE(SQUARE)

Animals can do transitive inference Baboon social hierarchy (3160 possible pairs) (X dominates Y & Y dominates Z) X dominates Z Chickens submit to a chicken who has beaten a chicken who has beaten them. Chickens challenge a chicken who was beaten by a chicken they have beaten. Lab studies on ordering: A > B > C C > D > E A > G E > F > G

“The magical number 4” (Nelson Cowan)

• Short-term memory is limited to about 4 objects.
• Subitizing collections of objects, up to about 4.
• Human sentences are limited to about 3

participants (subject, direct object, indirect object).

The size and shape of minimal subscenes

Watch carefully: how many stars are there?

There is a psychological limit on perceivable discrete numerosity around 4-5. Short-term memory (Nelson Cowan, “The Magical Number 4”) Many animals can discriminate discrete numerosities up to about 3-4. (Stanislas Dehaene, The Number Sense.)

For any language, the maximum number of obligatory noun phrases in any sentence (with any verb) is 3. E.g. Mary awoke (1 NP) Mary ate breakfast (2 NPs) Mary put her keys on the table (3 NPs)

• A limit of four objects in visual working

memory (Luck and Vogel, 1997)

• Mean short-term memory capacity in

adults of 3-5 chunks (Cowan, 2001)

• Deictic subsystems are limited to about

three degrees of difference (e.g. Japanese ano, kono, sono).

• Minimal subscenes (Arbib) contain up to

about four separate participants. The basic clauses of languages are adapted to describing such minimal subscenes.

In a given perceptual scene, different languages tend to agree in the way in which they conflate perceptual information into concrete objects, which are then lexicalized as nouns. There is more variation in the way in which languages conflate relational components into the meanings of verbs and other predicates. To put it another way, verb conflations are less tightly constrained by the perceptual world than concrete noun conflations. Loosely speaking, noun meanings are given to us by the world; verb meanings are more free to vary across languages. (Gentner, 1981, p.169)

Primacy of objects

There is some human-universality in how we carve up actions and events into manageable

• packages. Schleidt and Kien (1997) analyzed

film of ‘the behavior of 444 people (women, men, and children) of five cultures (European, Yanomami Indians, Trobriand Islanders, Himbara, Kalahari Bushmen)’ (p.7). Based on careful definitions and cross-checking by independent analysts, they found that ‘human action units are organized within a narrow and well-definable time span of only a few seconds. Though varying from 0.3 seconds up to 12 seconds or more, most of the action units fall within the range of 1-4 seconds’.

Perceptual events last for about three

• seconds. We re-analyze scenes

roughly every three seconds.

1st cohort 2nd cohort later ISN MAN PUSH WOMAN FALL > MAN WOMAN PUSH FALL > MAN PUSH WOMAN . MAN PUSH FALL WOMAN (Sequence of two (Verbs brought adjacent) (Adjacent verbs Noun-Verb clauses) collapsed to one)

Nicaraguan Sign Language evolved from small sentence-packages to larger ones.

This is now near the natural limit for simple clauses.

Animals (don’t) have propositional representations of the world

Logically minimal propositions: E x [LION(x)] There is a lion

Many animals have simple predicate-argument structure.

• The most basic semantic distinction is between

predicates and arguments – PREDICATE (x).

• This asymmetric relation is the first sign of

semantic structure.

• The innermost brackets in any complex semantic

formula are those separating predicates from their arguments:

believe(john, ( x y (woman(x) & man(y) & love(x,y)))) E A

Many animals have simple predicate-argument structure. A parallel between the predicate-argument structure PREDICATE(x) and distinct neural mechanisms for (i) Attending to an arbitrary object - (x) - (via the dorsal stream), (ii) Categorizing an object - PREDICATE(x) - (via the ventral stream).

(x) (y) (z)

(z)

Dorsal stream locates object Ventral stream categorizes object

RED(z)

`` ... the most primitive contact that the visual system makes with the world (the contact that precedes the encoding of any sensory properties) is a contact with what have been termed visual

• bjects or proto-objects ... As a result of

the deployment of focal attention, it becomes possible to encode the various properties of the visual objects, including their location, color, shape and so on.'' (Pylyshyn (2000):206)

So, I am committed to one-place predicates as the basis for representations of scenes/objects/events in animals. How can at least some scenes/objects/events be represented solely in terms of one-place predicates? I appeal to two psychological themes:

• Global versus local attention
• Frame of Reference

Global and local attention

RED YELLOW PURPLE SKIRT DANCE GIRL GIRL

Quick global attention delivers something like with predicates only approximately bound

Global and local attention

Another example

KISS BOY GIRL SMILE

Quick global attention delivers

Frame of Reference

Judgements are made relative to their contexts. E.g. What is judged as WHITE in half-light is judged as GREY in full light. Jokisch & Troje (2003) Fast-striding animals are seen as relatively small, and slow-striding animals are judged to be relatively big. Sarris (1998) Chickens can be trained to make perceptual judgements such as ‘big for a red cube’ and ‘small for a green cube’.

An iconic notation, taking a cue from event semantics

KISS AGENT BOY PATIENT GIRL SMILE KISS(e) & AGENT(x) & BOY(x) & PATIENT(y) & GIRL(y) & SMILE(y)

Boxes correspond to individual variables, e, w, x, y, z.

An iconic notation, taking a cue from event semantics

AGENT GIRL SMILE AGENT(y) & GIRL(y) & SMILE(y)

Boxes correspond to individual variables, e, w, x, y, z.

(No ontological distinction between a one-participant event/state and an individual object.)

Meaning after public symbols have emerged

Overt, public symbols facilitate (or even enable) acquisition of more abstract concepts, e.g. relations between relations. Many animals can learn the abstract concepts

• f SAME/DIFFERENT, and generalize these

to new stimuli. This is a first order task – a relation between objects. A second order task, learning relations between relations, is facilitated by having a public symbol.

tri

~ tri

First: private concepts. (Much?) later: public symbols.

Animals can represent abstract properties and relations, if helped by a public symbol for a more basic relation. Relations between relations SAME DIFFERENT DIFFERENT DIFFERENT DIFFERENT SAME

Sherman and Austin first learned holistic sequences: Pour water, pour juice, give banana, give apple. At this stage they produced random collocations, e.g. banana juice give. Next they were trained that *pour apple, *pour banana, *give juice, *give water are ‘wrong’. Now, they correctly generalized to, e.g. Pour milk, pour Coke, give orange, give candy.

Grammatical distribution can influence category formation

What the animals had learned was not only a set of specific associations between lexigrams and objects

• r events. They had also learned a set of logical

relationships between the lexigrams, relationships of exclusion and inclusion. More importantly, these lexigram-lexigram relationships formed a complete system in which each allowable or forbidden co-

• ccurrence of lexigrams in the same string (and

therefore each allowable or forbidden substitution of

• ne lexigram for another) was defined. They had

discovered that the relationship that a lexigram has to an object is a function of the relationship it has to

• ther lexigrams, not just a function of the correlated

appearance of both lexigram and object. (Deacon,

1997, p.86)

Arbitrary (public) labels affect private perceptual discrimination

Four experiments investigated the influence of categorization training on perceptual discrimination. Ss were trained according to 1 of 4 different categorization regimes. Subsequent to category learning, Ss performed a Same- Different judgment task. Ss' sensitivities (d's) for discriminating between items that varied on category-(ir)relevant dimensions were measured. Evidence for acquired distinctiveness (increased perceptual sensitivity for items that are categorized differently) was obtained. One case of acquired equivalence (decreased perceptual sensitivity for items that are categorized together) was found.

Goldstone R. (1994) Influences of categorization on perceptual discrimination. J Exp Psychol Gen. 1994 Jun;123(2):178-200.

Robert L. Goldstone (1998) PERCEPTUAL LEARNING Annu. Rev. Psychol. 49:5856121(998)

Laboratory experiments by Goldstone (1994) have suggested that physical differences between categories become emphasized with

• training. After learning a categorization in which one dimension was

relevant and a second dimension was irrelevant, subjects were transferred to same/different judgments ("Are these two squares physically identical?"). Ability of Ss discriminate between stimuli in the same/different judgment task was greater when they varied along dimensions that were relevant during categorization training, and was particularly elevated at the boundary between the categories. Further research showed that category learning systematically distorts the perception of category members by shifting their perceived dimension values away from members of opposing categories (Goldstone 1995). In sum, there is evidence for three influences of categories on perception: (a) category-relevant dimensions are sensitized, (b) irrelevant variation is deemphasized, and (c) relevant dimensions are selectively sensitized at the category boundary.

Balaban and Waxman (1992) found that speaking a word-label as the baby is exposed to a picture enhances their capacity to make categorical distinctions. This was in contrast to a control condition with a mechanical tone sounding while the baby was looking at the picture, which made no difference to the baby’s apparent judgement. The babies in this experiment were nine months old.

In the test conditions, two distinct objects were shown moving from behind a screen and then back behind the

• screen. While an object was in view, the experimenter said,

for example, ‘Look, a duck’ or ‘Look, a ball’, or, for both

• bjects ‘Look, a toy’. Thus the baby subject was exposed to

labels for the objects, but in one condition the labels were different (duck/ball), while in the other condition, the labels were the same (toy) for both objects. In the two-label case, the babies looked for significantly longer if the removal of the screen revealed, surprisingly, only one object. For the one- label case, the results were similar to the baseline condition. What seems to be happening here is that the explicitly different labelling of two objects leads the baby to expect two objects to be behind the screen, whereas labelling them both the same, as toy, does not produce this expectation.

(Xu, 2002)

Katz (1963) showed children four different geometrical

• shapes. With one group of children, each separate shape

was identified by its own nonsense syllable; with the other group, only two syllables were used, each syllable to a specific pair of shapes. After this training, the children were tested on whether two presented shapes were the same or

• different. The children who had received only two arbitrary

syllables tended more often than the other children to judge two shapes identified with the same syllable as the same

• shape. In later studies, Goldstone (Goldstone, 1994;

Goldstone, 1998) showed, more subtly, that the dimensions along which objects had been categorized in training (e.g. shape, colour) also had an effect on subsequent similarity judgements.

Merci. Thank you.