Week 3: Meaning

Question 1

Four types of semantics in Pulvermuller’s paper (essential reading)

Answer 1

Referential
Combinatorial
Emotional-affective
Abstract

Question 2

Referential semantics

Answer 2

Establishes links between symbols and the objects/actions they refer to

Question 3

Combinatorial semantics

Answer 3

Enables the learning of symbolic meaning from context

Question 4

Emotional-affective semantics

Answer 4

Establishes links between signs and internal states of the body

Question 5

Abstract semantics

Answer 5

Generalises over a range of instances of semantic meaning

Question 6

The ‘standard view’ of the mechanism of meaning

Answer 6

• Amodal, symbolic semantic system

- Separate from action/perception systems

Question 7

Grounding problem

Answer 7

A person just knowing the (syntactic) relationship between symbols cannot be said to know the (semantic) meaning of these symbols

Needs a link between sign, concept and referential information

Brought to light by Searle’s Chinese Room argument

Question 8

Harnad’s hybrid model

Answer 8

Attempt to solve grounding problem:

1) Perceptions lead to iconic representations of objects
2) Shared features of many iconic representations of the same thing brings a grounded categorical representation
3) Further symbolic representations are built by combining grounded ones (zebra: horse with stripes)

Question 9

Two problems with Harnad’s hybrid model

Answer 9

• Not biologically realistic

- Ignores other types of meaning apart from object reference

Question 10

Types of neurocognitive models for meaning

Answer 10

Symbolic model

Distributed model

Hybrid model e.g. integrated cell assembly (CA) model

Question 11

Integrated cell assembly (CA) model

Answer 11

CA = distributed circuit that ‘carries’ symbols in a network

Strengths:

• Biologically plausible(?)
• Solves grounding problem(?)

Question 12

Sensorimotor grounding of meaning?

Answer 12

Concepts and meanings need to be grounded in motor action and/or sensory perception. At least for some symbol types, an ‘amodal semantic system’ is insufficient.

Neural reuse of APCs in accordance with Hebbian learning

Could explain how cortical circuits for concepts develop and why they have specific:

• localisation
• temporal dynamics
• function

Question 13

Evidence for sensorimotor role in meaning

Answer 13

1) Explains otherwise puzzling (behavioural, neuroimaging, lesion) evidence for:
- Information sharing between e.g. motor, visual, auditory cortices
- Semantic (conceptual) processing reflected by actions and perceptions e.g. ACE
- SPP plus N400 paradigm

2) Biologically plausible theory given brain structure:
- APCs are distributed, long-range and multimodal, but not randomly so
- Sensorimotor and mirror neurons

3) Consistent with biologically-constrained simulation data and their ‘emergent topographies’:
- Specialised ‘semantic hubs’
- Category-specific semantic areas

4) Allows for many ways of learning

Question 14

Action compatibility effect (ACE)

Answer 14

People respond faster to sentences if the meaning matches the direction of the response
e.g. towards/away from something

Question 15

Grounding transfer

Answer 15

aka ‘Symbolic theft’

Word-word correlation learning (as opposed to word-world):
Previously learnt word + novel word form → new semantic circuit

Question 16

Grounding kernel

Answer 16

Required for grounding transfer:

A base vocabulary learnt in context of the action/object

Question 17

Semantic prediction and resolution

Answer 17

When people hear phrases strongly predicting an upcoming ‘critical’ word:

• ‘Semantic prediction potential’ (SPP) = anticipatory brain activity precedes the critical word, partly in sensorimotor systems relevant to the word meaning
• N400 during and after the critical word

Question 18

Objections to sensorimotor semantic grounding

Answer 18

Allegedly, effects are:

• not reproducible
• epiphenomenal, not functional/causal
• contradicted by evidence for ‘double dissociations’
• incapable of explaining abstract concepts
• restricted to only association of information, not true semantics
• task-specific (‘flexible’), not essential
• small

Question 19

Failed objections to sensorimotor grounding

Answer 19

Allegedly, effects are:

• not reproducible → nope
• epiphenomenal, not functional/causal → behavioural, TMS, lesion studies show causal role; effects are automatic in ‘non-attend’ paradigms(?)
• contradicted by evidence for ‘double dissociations’ → semantic circuits actually explain these double dissociations
• incapable of explaining abstract concepts → there are sensorimotor models aimed at doing exactly this

Question 20

Objections to sensorimotor grounding which are actually not objections to sensorimotor grounding

Answer 20

Allegedly, effects are:

• restricted to only association of information, not true semantics → necessary even though not sufficient mechanism for explaining meaning (Hebbian learning, correlation and timing drives neuronal connection strength)
• task-specific (‘flexible’), not essential → this flexibility is actually an important feature to model (different context-dependent modes of activating a semantic circuit)
• small → this is expected and more valuable if we (wisely) assume a many:many relationship between cognitive process and brain structure rather than 1:1