Visual Word Recognition 2

Question 1

what did James Cattell do?

Answer 1

• Frequency of usage of words in a language influences tasks involving printed words (Cattell, 1886)
• measurements of brain processes used a device called a chronoscope precisely measure how long it takes to make a decision on words
• accuracy in milliseconds

Question 2

Monsell et al (1898) investigation into word frequency effects in lexical decision, semantic categorisation and word naming.

Answer 2

• Stimuli: High (student / desk), medium (widow / furnace) and low frequency words (tyrant / shawl).
• Main effect of frequency. No interaction between task and frequency

Question 3

mehl et al. (2007) - Are women more talkative than men?

Answer 3

Large study with 396 participants revealed that men and women both speak about 16,000 word tokens per day. Thus, speaking and listening is about 32k per day.

Question 4

context diversity Adelman et al al. (2006)

Answer 4

• Contextual diversity (CD) not word frequency (WF) determines word naming and lexical decision times.
  CD: number of different contexts a word has been seen (number of different documents).
• Analysis based on different corpora (e.g. K&F, BNC) and lexical decision and word naming studies.
• Regression analyses showed an improvement of variance explained for WF and CD. However, improvement for CD higher than WF. Thus, CD more predictive of reaction times than WF.

Question 5

what did Jones et al. (2012) suggest about semantic diversity?

Answer 5

• it’s more important than CD because CD ignores information redundancy.
• Word BANK can occur in similar documents (e.g. about mortgages) or two very different documents (mortgages, and rivers). Although word can have a large document count it does not mean that these are reflecting truly distinct contextual uses of the word.

Question 6

outline Jones et al. (2012) reaction time experiment

Answer 6

Reaction times and naming latencies obtained from the English Lexicon Project (Balota et al., 2007). Document count and SD count obtained from three corpora

Question 7

Chateau and Jared (2000), impact of the exposure to print word recognition

Answer 7

• Exposure to print was measured using the Author Recognition Task
• 64 participants, 4 tasks (Lexical Decision, Form priming, Naming)
• conclusion: print exposure not only affects vocabulary and general knowledge (Stanovich & Cunningham, 1992) but it also enhances word recognition processes

Question 8

Mol and Bus (2011), meta analysis investigating the impact of print exposure from infancy to adulthood

Answer 8

• Print exposure in readers aged 3-5 was associated with oral language skills. Reading routines for children in school provide substantial advantages for oral language growth.
• Reading for pleasure may also improve academic success in college and university students.

Question 9

Aron and Snyder (2010), frequency effects for multi-word phrases

Answer 9

• Material: 4-word sequences (4-grams)
• Frequency information obtained from a large telephone conversation corpus (counts converted to frequency per million).
• Task: Phrase decision task have to indicate if the phrase is a possible sequence in English
• Participants: 26 students
  Results
• High bin: High frequency (HF) phrases (1040 ms) faster than low frequency (LF) phrases (1100 ms).
• Low bin: Also phrase-frequency effect (HF: 1059 ms, LF: 1125 ms).

Question 10

Siyanova-Chanturia et al. (2011) Binomial expressions

Answer 10

• Binomial expressions are three-word phrases that are formed by two content words of the same lexical class and a conjunction e.g. bride and groom, king and queen
• Participants (native and non-native English speakers) read sentences containing binomials (e.g. bride and groom) and their reversed forms (groom and bride) embedded in sentences.
• Eye movements were recorded used an eye tracker.
• Conclusion: Readers are sensitive to the frequency of multiword sequences.

Question 11

Lexical similarity coltheart (1977).

Answer 11

• Orthographic Neighbours: Number of words that can be created by changing one letter of a target word.
  e.g. target: MINE, neighbours: PINE, LINE, MIND, MINT (N=29)

Question 12

Andrews (1992) experiment on lexical decision task

Answer 12

Neighbourhood density was manipulated (number of neighbours, small N vs. large N) and word frequency (high vs. low).
results: significant interaction between frequency and neighbourhood density.

Question 13

Orthographic neighbours

Answer 13

Neighbourhood size/density (number of neighbours)
- Andrews (1989, 1992): facilitation (low frequency words)
- Coltheart et al. (1977): no effects
- Carreiras et al. (1997): inhibition

Question 14

what is the multiple read out model? Grainger and Jacobs

Answer 14

IA model with decision criteria
Three noisy response criteria:
- M: single word node activity (μ)
- ∑: summed activity of all active words (σ)
- T: time threshold (t)
with these three criteria the model can account for facilitation effects of neighbourhood density

Question 15

where are internal stores of knowledge of words?

Answer 15

• The mental lexicon
• Semantic memory (word meaning) - conceptual store

Question 16

what is decompositional theory?

Answer 16

• Word meanings best described in terms of sets of bivalent
• semantic features or semantic markings (+<feature> or -<feature>)</feature></feature>

Question 17

types of features

Answer 17

• Defining features = distinguished between defining and characteristic features
• Characteristic features = features possessed by most of a class of object, e.g., ‘flies’ and ‘builds a nest’

Question 18

semantic features (McRae, de Sa, & Seidenberg (1997))

Answer 18

Intercorrelated features:
- tend to occur together.
- living things tend to be represented by many intercorrelated features.
- many members of a natural kind category will share intercorrelated features
Distinguishing (or distinctive) features:
- enable us to distinguish among things.
- exclusive to single items within a category.
- artefacts tend to be represented by many distinguishing features.

Question 19

what is the semantic network model (Collins & Quillian, 1969)

Answer 19

Characteristics:
- Concepts represented by nodes in a network.
- Nodes joined together by links representing relations between concepts, e.g. set membership (Fido is a dog), set inclusion (dogs are animals), part-whole (a seat is part of a chair), and property attribution (canaries are yellow).
- The meaning of a word is determined by the place of the node in the network as a whole.
- Set inclusion links (usually called IS-A link) mean that nets are hierarchically organised.
- Semantic Economy.

Question 20

what did Hollan (1975) show about semantic network?

Answer 20

Showed that simple semantic networks and feature theories were formally equivalent.

Question 21

what did Rips, Smith and Shoben (1975) ahow about semantic networks?

Answer 21

Pointed out that formal equivalence does not imply psychological equivalence, i.e. different processing assumptions may results in more or less satisfactory performance models.

Question 22

what is prototype theory?

Answer 22

• Concepts are centred around a representation of a prototypical member of the class. e.g. ‘robin’ versus ‘ostrich’ for the concept ‘bird’.
• The meaning of a word is given by its prototype together with some specification of how far instances can differ from the prototype and still be a category member, i.e. ‘bird’.
• Boundaries may often be ‘fuzzy’ rather than well-defined.

Question 23

Outline sentence verification task

Answer 23

Subjects asked to respond true or false to sentences of the following kind:
- Set Inclusion: ‘A robin is a bird’; ‘A whale is a fruit’ and/or
- Property Attribution: ‘A robin has feathers’; ‘A whale has seeds’.
rationale
- Words presented in sentential context as in normal comprehension.
- Use of simple sentence frames to control differences in syntactic processing.
- Reaction times used to infer how knowledge about word meaning is stored/used in the process of comprehension.

Question 24

what is the principle of cognitive economy?

Answer 24

• Information about concepts is stored at the highest appropriate level in the hierarchy.
• e.g. ‘has feathers’ stored with ‘bird’ rather than with each different exemplar such as ‘robin’ or ‘canary’
• Early studies confirmed prediction that sentence verification times would increase as distance in the hierarchy increased

Question 25

what are the assumptions of the hierarchical model?

Answer 25

• It takes time to move through each step.
• Where one step is dependent upon completion of another step, the times are additive (S->P or S->S).
• Retrieval proceeds from one node in all directions at once (parallel).
• Average time for any step is independent of which particular level(s) is involved.

Question 26

what is the prototype effect?

Answer 26

Within-category typicality or prototype effects can be explained in terms of associative links of varying strength.
e.g., between ‘ostrich’ and ‘bird’ and between ‘robin’ and ‘bird’.

Question 27

variability of negative judgement times

Answer 27

Can be explained in terms of IS-NOT-A links of varying associative strength.
Also explains why it takes longer to reject closely related items, e.g. ‘A canary is an ostrich’ versus ‘A canary is a salmon’.

Question 28

problems with spreading activation

Answer 28

• Proliferation of auxiliary assumptions about way information is accessed.
• Impossible to design an experiment to test underlying theory of representation.

Question 29

what are distributional semantics (harris, 1954)

Answer 29

words with similar meanings are used in similar contexts

Question 30

what are count models?

Answer 30

using large corpora to obtain relatedness values between words (similarity in vectors representing the words)
the semantic similarity measure between words involves comparing the vectors between words

Question 31

predict models: continuous bag of words (CBOW)

Answer 31

predict target words based on context words. method uses a neural network that learns from a large corpora

Question 32

predict models: word2vec (Mikolov et al., 2013)

Answer 32

similarity measure involves using the weights from input nodes to the hidden units (e.g., 300) of word pairs (e.g., cosine distance)

Question 33

problem with distributional semantics

Answer 33

meaning=relation with other words. How do we know the meaning of those words? Also a problem for prototype and semantic network theory

Question 34

general problem of semantic theories

Answer 34

grounding problem: what is the relation between abstract symbols and our knowledge, experiences, actions, and perceptions?
maybe metaphors are used to represent abstract meanings