Were up for the 2nd assignment, the experiment'

1

• Were up for the 2nd assignment, the experiment.
  
• CogLab will be installed in 4the computer lab
• You will form groups of 4 students and pick one
  expt (no overlap allowed, 1st come - 1st served,
  requests by email by Friday 6th).
• Experiment executed in class are not allowed.
• 4 machines will be reserved for you to do it
  next Wednesday (no formal class).
• The results, commented based on the theoretical
  ground and the literature will be discussed in a
  class presentation on Friday, May 4th (20 mins
  per group).
• The outcome will weight for 10 to 15 of the
  final -overall- assessment

2
Outline

• Psycholinguistic
• Properties of language
• The mental lexicon
• Mental lexicon and semantic networks
• Biological evidence of mental lexicon
• Language comprehension
• The visual input
• The acoustic input
• Word processing and recognition
• Integrating words in phrases
• Language production
• Afasia
• Neurophysiology of language

3
Psycholinguistics

• The study of language from the point of view of
  psychology and cognition is called
  psycholinguistics.
• This domain can be subdivided in three
  sub-domains
• Linguistics, that is the study of the structure
  of language and of its changes its the core of
  the whole domain
• Neurolinguistics, that is the study of the
  relationship between brain and language
• Sociolinguistics, that is the study of the
  relationship between social behavior and
  language.
• The linguistics includes
• phonetics and phonology the sounds of language
  and their organization in complex systems
• morphology the internal structure of words
• syntax the combination of words in phrases
• semantic the meaning of language
• pragmatic the comunicative use of language the
  comunicative strategies used by humans while
  speaking.

4
General properties of language

• Establishing the distinctive properties of
  language causes a dispute among scientists, but
  there are at least five features on which there
  is a general agreement. Language would be
• Comunicative, in that it allows comunication
• Arbitrarily symbolic, in that it creates an
  arbitrary relationship between symbol and
  referent)
• Regularly structured, in that it has a structure
  and sequences with similar configurations have
  similar meanings. This allows sharing of
  language
• Generative and productive, within the limits
  established by the structural rules an infinite
  number of new expressions can be created
• Dynamic, in that all languages evolve
  continuously.

5
The mental lexicon

• The mental representation of language constitutes
  the so called mental lexicon (ML).
• The ML would include semantic and syntattic
  (i.e., the rules to combine words to form
  sentences) information as well as the details
  about the form of the words (spelling, sound,
  etc.).
• Many psycholinguistic theories attribute a key
  role to the ML, but there is debate as to whether
  there is a unique ML or different MLs for
  production and comprehension. Another possible
  distinction is for orthographic (visual-based)
  and phonologic (sound-based) MLs.
• A normal adult has a vocabulary of about 50 k
  words and can produce up to three words a second.
  Therefore, the lexicon must be organized in a
  highly flexible way, better than a simple
  dictionary
• it cannot be sorted alphabetically
• it does not have a fixed content
• some words are more accessible than others

6
ML and semantic networks

• Collin Loftus (1975) proposed that the ML is
  organized in semantic networks in which
  individual words are the nodes that are connected
  with other nodes.

• In a network, semantically related words are
  closer than unrelated words.

7
ML and semantic networks

• The strength of the connections and the distance
  between nodes of a network are determined by the
  semantic or associative relationship beteen nodes
  
• Activation would diffuse between adjacent nodes
  before than between distant nodes.

Car
Hot
Transport
Bus
Fire
Highway
Sunset
Car
Firetruck
Truck
Red
Dawn
Clouds
Ambulance
Violet
Orange
Green
Flower
Yellow
Rose
8
ML and semantic networks

• Empyrical support for the network-like structure
  of the ML comes from priming paradigms involving
  lexical decision tasks
• Word pairs are presented to the subjects.
• The first word is the prime.
• The second word can be a real word (semantically
  related or unrelated with the prime), a non-word
  (like fehuiag) or a pseudo-word (like coppee).
• The task is to decide as fast as possible if the
  second member is a word or not.
• LETS TRY IT
• Subjects are faster with semantically related
  words.
• In similar tasks in which the second member is
  always a word and the task is to read aloud it as
  fast as possible, related words show an
  advantage.
• Alternative models of semantic representations
  are, for instance, the feature models. They
  consist in linking to a word (e.g., bird) a list
  of features (wings, feathers, beak) that need to
  be activated in order for the word itself to
  activate. The number of features needed to
  activate a node depend on the strength of the
  node itself.

9
Biological substrate of ML

• The Wernicke aphasia appears following damages to
  posterior areas of the left emisphere. Patient
  with this desease show errors in language
  production called semantic paraphasiae. They can,
  for example, use the word horse when thy mean
  to say cow.
• A similar phenomenon is the one of the deep
  dyslexia, where in the presence of the written
  word cow they read horse.
• Patients with progressive semantic dementia
  initially show deficits of the semantic system of
  concepta in the absence of other impairments.
  They have problems assigning objects to the
  proper categories, and name the category name
  when asked to name instances of it (e.g. they say
  animal when they see a cat).
• Surprisingly, some deficits involve only specific
  semantic categories. For example, Warrington and
  McCarthy (1987) found patients with problems in
  pointing correctly to food or living item but
  were normal when tested with unanimated objects.
  Other patients showed the mirror pattern.

10
Biological substrate of ML

• Warrington interpreted these results as as due to
  two different types of storing, one more semantic
  (objects), another more bound to physical
  features (biological features).
• Analyses of the lesions and PET studies on normal
  subjects have evidenced the anatomical substrate
  of this dissociation (Damasio)

PET
11
Language comprehension

• The first task of a listener or of a reader is
  that of a perceptual analysis of the physical
  signal of language.
• Listening and reading imply different channels.
  Different modalities imply different types of
  segmentation. Let us take a look at the waveform
  of the word captain
• The pause occurring after two letters is not used
  to segment the groups of phonems in two words, as
  it is in the other example.
• Moreover, very often the space in between two
  written words does not correspond to a pause in
  the spoken language, as shown in the right
  example.

12
Language comprehension

• Spoken language relies upon different
  segmentation cues. The main one is the prosody,
  that is the information provided by the metric
  and the rythm of speech, by the accents or by the
  intonation.
• For example, in interrogative sentences the voice
  raises towards the end of the sentence (see
  previous slide).
• The prosody cues are absent in reading,
  especially when the text is read for the first
  time.
• To understand written language, the first step is
  to analyze the visual features of the letters.

13
The visual input

• The classical Pandemonium model by Selfridge
  (1959) explains the comprehension of written
  language from the input to the letter recognition

28 features
14
Word processing and recognition

• The lexical processing is based upon three
  processes
• access
• selection
• integration
• The access occurs when the perceptual input
  succeeds in activating the representations of
  words in the ML. Written and spoken language are
  well dissociated even at this stage.
• The lexical selection is then used to find the
  best match between input and ML.
• The integration links the different syntactic and
  semantic levels in a unique representation of a
  word.

15
Word processing and recognition

• Three types of model explain word recognition
• modular models
• interactive models
• hybrid models
• The modular models believe that recognition is a
  bottom-up process and that comprehension occurs
  through completely independent modules.
• The interactive models assume that recognition
  implies integration of different levels and that
  the context drives the lexycal selection changing
  the weights of the activation of individual
  nodes.
• The hybrid models assum that the lexical access
  is independent of context, while selection is
  dependent on both sensory (bottom-up) and
  contextual (top-down) information.

16
Integration of words into phrases

• The comprehension of language as a whole requires
  a level higher than that activated by individual
  words.
• Consider this sentence We have policies and
  rules in our District Plan to protect significant
  trees and bush in our city from damage and
  needless removal . How can I read this sentence
  without thinking of your president?
• The context of the phrase drove me to one meaning
  rather than another. This process occurs in
  real-time and often we do not even think of the
  alternative meaning of these ambiguous words.
• indeed, probably none of you thought of a
  context driving a car with me as a passenger
• The comprehension of a word starting from its
  context implies the integration of syntattic and
  semantic levels. A successful comprehension of
  the linguistic input requires the ability to form
  and understand the overall grammatical structure.
  
• Indeed, searching for individual words within
  sentences semantically incoherent is faster when
  grammar is intact, indicating that the syntactic
  analysis occurs even in the absence of meaning.
• Taken from http//www.northshorecity.govt.nz/our
  _environment/trees/tree_protection.htm

17
Integration of words into phrases

• Patients with syntactic deficits suffers from
  agrammatic aphasia.
• The theory of Frazier (1987) on syntactic
  analysis postulates that each sentence has a
  preferred interpretation.
• In this model, the phrases are sorted
  compositions of words that can be represented as
  hyerarchical trees reflecting the structure of
  the phrase.

Phrase
Verbal phrase
Nominal phrase
Verb
Nominal phrase
Propositional phrase
The spy saw the agent with the spyglass
18
Integration of words into phrases

• The initial process of syntactic analysis would
  make a structural analysis with no influence from
  other processes.
• This operation is performed in the most
  parsimonious way in order to achieve a real-time
  comprehension.
• Two mechanisms would support parsimony the
  minimal attachment and the late closure. The
  former minimizes the number of semantic nodes,
  the latter is the tendency to assign the word
  processed to the phrase currently processed
  rather than to the next.
• In our example the interpretation on the left is
  preferred to that on the right.

Phrase
Verbal phrase
Nominal phrase
Propositional phrase
Nominal phrase
Verb
The spy saw the agent with the spyglass
19
Language production

• The starting points of language comprehension and
  production are opposite. Whereas comprehension
  starts from individual words to translate into
  concepts, production begins from concepts for
  which one has to find the right words.
• The main model for language production is the
  Levelt model (1989).
• The first step is preparing the message. To this
  aim, Levelt assumes two fundamental aspects of a
  conceptualizator macroplanning and microplanning
  
• Macroplanning decides the concepts to communicate
  (e.g. distance between the house and the park)
• Microplanning decides how to comunicate those
  concepts (the house is near the park o the park
  is near the house)
• The output of the conceptualizator module is the
  input for the formulator, that adjusts the
  message in the correct grammatic and phonologic
  form. The anomia is a deficit at this level of
  the model.
• In the last level of linguistic production, the
  articulator, the articulation is planned the
  syllables of individual words are mapped into a
  motor pattern to move the right muscles for
  speech. The dysarthria, which is often associated
  with the Broca aphasia, is a deficit at this
  level.

20
Language production
21
Aphasiae

• 40 of cerebral strokes causes aphasia, a
  general desease of language.
• Mr.Tan, a patient of Paul Broca, could produce a
  unique word, tan. The post mortem exam of his
  brain evidenced a damage to the posterior portion
  of the inferior left frontal gyrus, known since
  then as the Broca area.
• Following studies by Broca confirmed this area of
  the left emisphere as responsible for language
  production.
• In 1870 Wernicke described two patients who had
  strokes involvingmore posterior areas and showed
  comprehension deficits.These patients had fluent
  production, but sounds, words and phrases were
  meaningless.
• The autopsy showed damages in the posterior
  region of thesuperior temporal gyrus, known
  since then as the Wernicke area.
• Damages to the arcuate fasciculus, the links the
  two areas, define theconduction aphasiae. The
  most selective impairment in this caseis a
  deficit of repetition of words.

Broca Area