Chapter 6: Recognizing Visual Words
• We are not just interested in discovering how we decide if a printed string
of letters is familiar or not, but also how all the information that relates to a
word becomes available
• Magic Moment: point in time where a person has recognized a word but
not yet had access to its meaning. (Balota, 1990). A word’s meaning can
only be accessed after it has been recognized.
• Johnson-Laird (1975): depth of lexical access may vary.
• Gerrig (1986): different “modes of lexical access” in different contexts (ie-
when we are reading something and getting very little sense from it)
• Spoken vs. written Language: speech signal is only available for a short
time, whereas a written word is often available for as long as the reader
• But facilitation of recognition by words related in meaning is found in
studies of both spoken and visual word recognition, and selecting the
appropriate meaning of an ambiguous word is a problem for both visual
and spoken recognition
• Literacy is an important feature of modern civilization
Studying Eye Movements in Reading
• Important in helping us understand both how we recognize words and
process larger units of printed language
• Limbus tracking: infra-red beam is bounced off the eyeball and tracks
the boundary between the iris and the white of the eye (limbus). Good at
tracking horizontal eye movements, but poor at tracking vertical ones.
• Purkinje system: accurate at tracking both movements. There are
several sources of reflection from the eye (cornea, back of lens): System
computes the movments of the exact center of the pupil from this
• Saccades: when we read, the eye travels in jumps of about 20 to 60 ms in
duraciton, with intervals of 200-250 ms when the eye is still.
• Fixations: when the eye is still
• Very little info is taken in while the eye is moving in a saccade. During
fixation, 15 characters to the right and 3-4 to the left in English speakers.
• Asymmetry reversed for those who read from right to left.
• Skilled readers may be able to take in more info
• Fovea: most sensitive part of the visual field, corresponds to the central
seven characters of average size text .Fovea is surrounded by the
parafovea (extending 5 degrees on either side of the fixation point), where
visual acuity is poorer. Beyond that is the periphery (even poorer visual
acuity). Extract most of the meaning of what we read from foveal region. • Rayner and Bertera (1979): moving mask: creates a moving blindspot.
Even if the fovea was masked, the reading was possible from the
parafoveal region, but at a greatly reduced rate (12 words a minute).
• If both the foveal and parafoveal region were masked, no reading was
possible: participants knew there were strings of letters outside the
masked portion and could sometimes pick up on the occasional
grammatical function word (and), or obtain info about start of words.
• Most influential model of eye movement control: E-Z Reader model :visual
processing, and oculomotor control jointly determine when and where
eyes move when we are reading.
• When we read, fixate on a point, and then visual attention progresses
across the line of text until a point is reached where the acuity limitations
of the visual system then make it difficult to extract more info and
recognize new words
• Then, attention shifts and an eye movement is programmed into the
oculomotor system to move to the point of difficulty.
• Next, a saccade takes place to the new location, and the process is
• Two stages of saccades:
- 1) early labile stage, when planned saccade can be cancelled if it is no
longer necessary (identified word in proposed target location)
- 2) after labile stage, saccade cannot be cancelled.
• Most controversial assumption to the E-Z Reader theory: attention is allocated to
one word after another, in strictly serial fashion, shifting only after each word is
identified ensures that words are processed in correct order.
• Word identification: 2 stages
- 1) familiarity check: do I know this word: can trigger saccade
- 2) full lexical access: retrieval of meaning, representation of word
integrated with emerging linguistic structure: triggers shift in attention to
• Saccades and attention are decoupled in this model; have different sources of
control (familiarity and identification)
• Linguistic processing can affect eye movements- if analysis is wrong, might
return to earlier location. Higher level processes intervene only when something
Reaction time and other measures
• Naming task: participants are visually presented with a word that they have to
name; time it takes to start pronouncing the word aloud (naming latency) is
• Naming latencies: approx 500 ms.
• Lexical decision task: participants must decide whether a string of letters is a
word or a nonword. Displayed on computer screen- participant must press one key if it is a word and another key if it is a nonword. Experimenter measures
reaction times and eror rates.
• Problem: speed-accuracy trade-offs- faster a participant responds, more errors
• Absolute time is not very useful; more concerned with differences between
• Tachistoscopic identification: participants are shown words for a very short
amount of time.
• Researchers used to use a tachistoscope; now computers are used.
• Experimenter records the thresholds at which participants can no longer
confidently identify items.
• Subliminal perception: behavior is affected although participants are unaware
that anything has been presented.
• Priming: presenting material before the word to which a response has to be
• Most common: presenting one word prior to the target word. First word: prime.
• Stimulus-onset-asynchrony (SOA): time between the onset (beginning) of a
presentation of one stimulus and the onset of another.
• Prime does not have to be a single word, and does not have to be linguistic
What makes word recognition easier (or harder)?
• Frequency effects and semantic priming are found in both spoken and visual
Interfering with identification
• Can slow down word identification by making it harder to recognize stimulus (ie:
degrading its physical appearance- stimulus degradation)
• Can be achieved by reducing the contrast between the word and the
background, or by rotating word to an unusual angle.
• Presenting another stimulus immediately after the target, interferes with
recognition process (backwards masking)
• If masking stimulus is unstructured (just a patch of randomly positioned black
lights, burst of light): we call it energy.
• If masking stimulus is structured (letters or random parts of letters): pattern
• Energy masks operate on visual feature detection level: cause visual feature
shortage and make feature identification difficult. Interfere at the letter level and
limit time available for processing.
• Perception without awareness: form of subliminal perception
• Words that have been masked so that participants aren’t consciously aware of
them sometimes produce an effect, even to the level of semantic processing.
• Holender (1986): emphasized ensuring that participants are equally dark-adapted
during the preliminary establishing of individual thresholds and the main testing phase of the experiment. Otherwise, cannot be sure that information is not
reaching conscious awareness in testing phase.
• As of yet, it is unclear whether we can identify and access meaning-related
information about words without conscious awareness, but evidence leans in the
• Can also present a word, but delay the presentation of one or two letters at the
beginning of the word by backward masking those letters.
• In English, after 60 ms, it doesn’t make much difference, but before that, delaying
a consonant disrupts visual word recognition much more than delaying a vowel.
• Early on, consonant identification is particularly important for recognizing a word.
• In English, consonants have a more regular mapping from visual appearance to
sound, whereas vowels do not.
Frequency, familiarity and age of acquisition
• Frequency of a word is v. important in word recognition.
• Commonly used words more easily recognized, responded to more quickly.
• Effect first demonstrated in tachistoscopic recognition
• Whaley (1978): Frequency- single most important factor in determining speed of
responding during the lexical decision task.
• Effect of frequency is not just a result of differences between frequent and very
infrequent words, but also between common and slightly less common words
• Kucera and Francis (1967): most popular norm of frequency, listing the
occurrence per million of a large number of words in many samples of printed
• There are frequency differences between versions of English, and between
written and spoken frequency.
• CELEX database: stored electronically, so searchable. Useful for making up lists
of materials with specific characteristics
• But all this is only an approximation to experiential familiarity (Gernsbacher,
• I.e: approximation may not work for low-frequency words (psychologists might be
very familiar with terms that have low frequency in general language)
• Common words tend to be short
• Frequency entangled with age-of-acquisition (AOA): age at which you first learn a
• Normally children learn more common words first, but there are exceptions (ie.
• Later the AOA, the more difficult it is for someone with brain damage to produce
• Size of correlation between early learned items and frequency varies, from .68
• Has been suggested that frequency effects are really AOA effects, but some
studies also suggested that AOA effects have not controlled adequately for
frequency (not taking cumulative frequency into account- how often words will be
encountered in a lifetime) • French study showed that AOA effects persist even when cumulative frequency
is controlled for.
• Probable that both AOA and frequency have effects on word processing.
• AOA particularly affects word reading, but cumulative frequency has an effect on
• Tasks involving redundancy and regularity in the input-out-put mappings (ie-
reading, where letters map onto sounds in a predictable way) are less prone to
AOA effects, and are sensitive to cumulative frequency. But tasks with less
redundancy and regularity do show AOA effects.
• Experiment: Ellis and Lambon Ralph: introduced items into the training regime at
different times. Items learned early possessed an advantage independently of
• As a network learns more items, it becomes less plastic, and late items are not
efficiently or as strongly represented, because they are more difficult to
differentiate from items that have already been learned. Early learned items have
a head start that enables them to develop stronger representations in the
network. Late learned items can only develop strong representations if they are
presented in high frequency.
• Gough (1972): during word recognition, letters are taken out of a short-term
visual buffer one by one, at a rate of 15 ms per letter. Therefore, long words
would seem to be harder to identify than short words. But length effect
independent of frequency is hard to find.
• Complication: 3 different ways of measuring word length: how many letters, how
many syllables and how long it takes to say the word.
• Previously, it was thought that there was clear evidence that longer words take
longer to pronounce. But Weeks (1973) found that word length (measured in
letters) had little effect on naming words when other properties were controlled
• Thus, number of letters in a word has little effect for short words, but has some
effect for words between 5 and 12 characters long.
• Naming time increased as a function of the number of syllables in a word
(Eriksen, Pollack and Montague, 1970)
• We take longer to name pictures of objects depicted by long words, compared
with pictures of objects depicted by short words.
• N-statistic: number of words that can be created by changing one letter of a
• N is a measure of neighborhood size (density)
• Makes words with a high N easy to recognize when other factors have been
controlled for, although clear benefits are only found for low frequency words.
• Rime part of neighbors seems to be particularly important in producing facilitation • Andrews (1997): neighborhood size has more effect than neighborhood
• Words are generally responded to faster than nonwords
• Less plausible nonwords are rejected faster than more plausible nonwords
(pseudowords: ie, they follow rules of word formation)
• Once you have identified a word, it is easier to identify it the next time (repetition
priming). Facilitates accuracy of perceptual identification and lexical decision
• Effect lasts up to several hours or longer
• Repetition priming effects are stronger for low-frequency words than for high-
frequency words (frequency attenuation)
• Repetition effects have 2 components: brief lexical access effect (not sensitive to
frequency) and long term episodic effect (sensitive to frequency)
• We generally obtain facilitation by repetition priming only within one domain
(visual or auditory), but semantic priming (by meaning or association) works
across domains = episodic view.
• If words share letters: orthographically related.
• Seeing a word that looks like another word = orthographic priming or form-based
• Difficult to demonstrate
• Only effective with primes masked as short SOAs, so that prime wasn’t
• Efficacy of form-based primes depends on exact makeup of task. Can have
inhibitory effect, as well, if visually similar words are in competition.
• Form based priming is much easier to obtain if the prime is masked (because
masked priming is a purer form of priming that has no contribution from
• Identification of a word can be facilitated by prior exposure to a word related in
meaning (semantic priming)
• Identification made easier if it is immediately preceded by a word related in
• Effect can be found across many tasks and not limited to visual word recognition
• First word (prime) can speed up recognition of the second word (target) =
facilitation • But prime can also slow down identification of target = inhibition
• With very short intervals, priming can occur if the prime follows the target
• If target was presented for 50 ms, followed 80 ms later by prime, there was no
facilitation. But if the target was presented for only 30 ms, and followed 35 ms
later by the prime, there was significant backward priming of the target.
• Suggests that words are processed in a parallel if the time between them is
• Semantic priming is a type of context effect : words are rarely read in isolation.
• Processing might be speeded up if words related to the word you are currently
reading are somehow made more easily available, as they are more likely to
come next than random words.
Other factors that affect word recognition
• Grammatical category to which a word belongs
• Imageability, meaningfulness, concreteness
• Rubin (1980): frequency, emotionality and pronunciability are the best predictors
of performance in experimental tasks
• Whaley (1978): frequency, meaningfulness and number of syllables had effect on
lexical decision times, although AOA can be important variable
• Balota et al compared phonological, lexical and semantic variables on speeded
visual word naming and lexical decision tasks.
• They found that the contribution of the variables was highly task dependent.
• Semantic variables are especially important in lexical decision
• Also, syntactic environment affects word recognition (Garret, 1984)
• Refer to page 178: (1) and (2): in both cases, the target word is semantically
unpredictable from the context, but syntactic context affected lexical decision
times so that people were slower to