WORD ACQUISITION

We growing ……..
So far
◦ We have seen that infants may learn words by separating the speech stream using
perceptual information.

◦ This information is in the domain of phonology, i.e. sounds and their properties.

◦ Various cues like rhythm, phonotactics etc are used to attain this goal.

◦ This is also called phonological bootstrapping.

◦ And the resultant understanding is that of phonological word form

◦ Now, the next step is to pair word forms with meaning

◦ Now performing this mapping is not easy, to put it mildly.

◦ Children will need to navigate a number of issues before they can successfully
map form onto mapping.

◦ Two of the most important issues are: reference problem and extension problem.
The reference problem.
◦ Quine talked of radical translation problem (1960).
◦ This refers to a person translating a previously unknown, radically different
language.

◦ How would he go about it?

◦ He would start by what word ‘natives’ would use for a certain event like : a white
rabbit scurried by and the native saying ‘gavagai’.

◦ Now what does it really refer to? Rabbit? White? Scurrying? Animal? He seeing the
event?

◦ Possibilities are many: this is the reference problem

Daniel Everett: don’t sleep there are snakes

◦ Some examples:
◦ Daniel drops one stick on the ground. The Piraha man says: ‘xii xi bigi kaobii”
◦ Dan thinks it means: ‘you dropped one stick’/’stick falls to the ground’
◦ Later he finds out it means ‘stick it ground falls’

◦ Similarly, when he drops two sticks to the ground, the sentence received was “xii
hoihio xi bigi kaobii:
◦ He thought it means: two sticks fall to the ground.
◦ But later finds out it means: “a slightly larger quantity of sticks falls to the
ground”
◦ This language, Piraha, does not have words for hello, goodbye, thank you, I am
sorry, you are welcome etc

◦ When a Piraha arrives in a village, he might say “ I have arrived”. No one says
anything in reply.

◦ If one gives something to someone, no ‘thank you’ is used. Sometimes they say
something that literally means ‘transaction acknowledged’.

◦ The act might be reciprocated later in kind, but not verbally.

Something similar happens with the child too, only
more difficult

◦ A child learning his first words and their meanings has a similar problem. In fact
worse problem, as there is no pre-existing first language either.

◦ When s/he hears her mother say: “that’s a dog”, the meaning is actually not
exactly clear.

◦ There can be many possible meanings and the child has very limited
understanding of the language to find her way about.
Also

◦ Linguists know that there might be words for objects like ‘rabbits’ in a language
and try to figure it out.

◦ But children do not even know that.

◦ E.g. for children, the word ‘gavagai’ might even mean “we don’t have any word for
that animal in my language”
Worse

◦ Learning word-meaning mapping is difficult for concrete words.

◦ How does a child learn meanings of words that have no referent? Abstract words?

◦ Like: the, think, love

◦ These are problems, that ‘we’ know. But do the child really find the task difficult?
May be not.
◦ Despite the problems, children actually learn words impressively fast.

◦ Pinker (1994) : “lexical vacuum cleaners”.

◦ Bergleson & Swingley (2012) tested whether 6 to 9 month old infants understand
words denoting body parts and food items.
◦ They used eye tracking methodology
◦ basic premise: people tend to look at objects whose name they hear.
◦ So, the infants looked at a computer screen displaying objects while their mother
would say, for example, ‘look at the apple’
◦ The subjects looked at the intended object more times than could be attributed to
chance.
◦ This means: they could map the name on the referent.
◦ Fenson (1994) reported that average American 12 month olds understand 50
words.

◦ It is thus not surprising that children understand a large number of words by age
2 years and keep adding to the list too fast for record keeping.

◦ Production vocabulary is less than comprehensive, but that too improves rapidly.

◦ Children are also capable of ‘fast mapping’, learning a new word without directly
being told the referent.
◦ Chromium experiment (Carey & Bartlett 1978)
◦ Showing them two trays and saying ‘bring me the chromium tray, not the blue one, the
chromium one’
The Extension problem
◦ So, let’s say the child figures that ‘dog’ refers to her pet dog. Solves the reference
problem.

◦ Now, the word dog does not refer to one single instance of the animal.

◦ It also refers to all other breeds of dogs, from chihuahua to Labrador to corgi.

◦ How does she learn to ‘extend’ the meaning to the category?

◦ And also learn where to stop?

◦ For example, ‘dog’ cannot be used for cats and rabbits?
◦ Nor with things thematically associated with dogs, like leash, a bone etc
Extension problem is language specific

◦ Words do not map onto categories in the same way across languages

◦ This means that a child needs to learn to extend words in different ways based on
the language specific preferences.

◦ An important study in this regard showed how English and Korean differ in how
they describe the location of an object in relation to other objects (spatial
relations).

◦ In English, the distinction between ‘in’ and ‘on’ is important, they show the
difference between ‘containment’ and ‘support’
◦ In Korean, there is no word for ‘in’ and ‘on’. This language does not distinguish
between containment and support.

◦ Instead, this language distinguishes between ‘tight fit’ and ‘loose fit’.
◦ ‘Kkita’ in Korean is used to describe objects in tight fit condition.
◦ ‘Nohta’: ‘putting on a horizontal surface’, like putting a cup on table
◦ Nehta: ‘putting loosely around’. Putting books in bag, apple in bowl.
◦ Pwuchita: juxtapose surfaces. ‘magnet on fridge’
◦ ‘Ssuta’: put clothing on head (put hat on)

◦ Thus, spatial words do not always map on the same perceptual categories. Still
children typically learn to map these early.
An example from the study
THEORIES
What we need to start learning words
◦ Researchers have proposed a number of theories based on different principles to
explain how word learning really happens among infants/children.
◦ In terms of their point of departure, these theories can be of few types:

◦ Innate linguistic biases/properties

◦ Non-linguistic factors:
◦ Socio-pragmatic factors
◦ Attentional learning account

◦ Syntactic bootstrapping:
Linguistic Constraints theory part 1: innate
constraints

◦ Key point: since new words can potentially map onto any aspect of an
event, are we innately biased to consider some meanings before others?

◦ In other words, are there any constraints that would put some limits on
the possibilities?

◦ A number of possibilities have been proposed in this regard, called

assumptions.
Three important assumptions

◦ whole object assumption: Learners should assume any new word refers to the
entire objects rather than its parts. Thus, ‘this is a dog’, where dog refers to the
entire animal not its parts, like legs or ears.

◦ ‘Taxonomic assumption: while extending the mapping of a word on referent, the

extension can happen only to the member of the same taxonomic group.

◦ Mutual exclusivity assumption: learners should assume that objects have only one
name. thus if we already have a name for one object (dog), assume that other
words spoken in the presence of dog are not alternative names of dog but it refers
to something ‘about’ the dog. E.g. dog running.
Similarly, some authors have proposed ‘biases’

◦ The shape bias: this is said to bias children to label objects of similar
shapes with the same name.

◦ The function bias: this biases children to label novel objects that have the
same function with the same name.

◦ Pragmatic principle of conventionality and contrast:

◦ Conventionality: children assume that there is a conventional form that
should be used in a language community.
◦ Contrast: similar to mutual exclusivity, any difference in form means
difference in meaning.
Constraints theory. Part II
◦ Developmental lexical principles framework was proposed as a response to the
critiques of constraints theory I.

◦ this framework proposed six principles of word learning.

◦ These are organized in two tiers.

◦ Tier I: the principles that are innate and are needed to help infants ‘get off the
ground’ (make a beginning) and these are available to them from the very
beginning.
◦ These are:
◦ Principle of reference: words stand for objects, actions and attributes
◦ Principle pf extendibility: words do not necessarily refer to one single object, but a category
of similar objects
◦ Principle of object scope: words refer to objects primarily. If not it applies to actions and
events. Also, words refer to whole objects.
◦ Tier II principles: having only Tier I principles are not adequate for
children to learn. Also they may lead to lot of errors. So, they develop
more refined principles as they grow. These are more sophisticated.

◦ These are:
◦ Principle of conventionality: children need to learn the conventional
meaning of words, not their own creations etc.
◦ Principle of categorical scope: perceptual similarity is no longer
adequate. Children learn that words label taxonomic categories, first at
basic level and then at superordinate levels.
◦ Novel name-nameless category principle. Children at this stage assume
that a novel name refers to an object which does not yet have a name.
However, unlike mutual exclusivity principle, this principle does not
prevent children from having 2 names for one object( Fido & dog).
Non-linguistic factors
◦ Is it only linguistic constraints that help in learning words? Is it possible
that there are other factors that might be as important?

◦ A number of researchers have pointed out that children’s learning of

words is a result of their growing need to understand communicative
intent and joint attention.

◦ Others have suggested that powerful cognitive processing (attentional

learning) alone can explain word learning.
◦ The warring factions thus are:
◦ Socio-pragmatic factors
◦ Attentional learning
Socio-pragmatic account

◦ Bruner, Tomasello and others are the proponents of this account.

◦ According to this view, children do not need any linguistic constraints. Rather they
are guided by what others are trying to say. In other words, ‘social clues’ help
them learn the meaning of new words.

◦ Speakers, in any social situation, give adequate rich and diverse cues/hints to the
listeners about what they are referring to, by looking at the object or gesturing to
it etc.

◦ in this case, two abilities allow children to use these cues to learn words and their
meanings.
◦ Ability to establish joint attention
◦ Ability to understand speaker’s communicative intent
Support for this theory
◦ Toma studies: famous studies that show children do use social cues to learn novel words
◦ Children were given a toy while the experimenter held a different toy
◦ Half children were part of joint attention condition
◦ The other half were part of discrepant condition
◦ Joint attention condition: the experimenter waited till the child and she were looking at
the child’s toy and she labelled the child’s toy : it’s a ‘toma’.
◦ In the other condition: the experimenter waited till the child looked at her own toy and
then labelled her (experimenter’s) own toy as ‘toma’. In this case, the child was not
looking at the experimenter’s toy.
◦ In the test phase, children were shown both the toys and asked “where is toma?”
◦ Children 18 and 19 months old (not below that), from both conditions, identified the
toma correctly
◦ This proved that they were aware of the speaker’s attention, even in the absence of joint
attention.
Again,
◦ This theory is also good at explaining verb learning by children.
◦ Unlike nouns, learning verbs is particularly tricky, as verbs denote action that is
short-lived and intangible (hitting, e.g.).
◦ Also, most of the times when children hear a verb, the action is not ‘being
performed’, like ‘you are eating!’
◦ Tomasello and Kruger (1992) noted that mothers use verbs to either request the
child to complete an action, anticipate an impending action, or comment on an
action just finished, but rarely while it is happening.
◦ They say the answer lies in “intention reading”
◦ Children actively monitor the adult’s intentions.
◦ They show this through the ‘widget’ experiment
Widget experiment
◦ Experimenters created a novel apparatus where a toy spun around on a
turntable
◦ Children saw the event and heard the experimenter say a sentence that
would (1) refer to the toy: widget or (2)the action: widge it
◦ Social cues provided also were of two types:
◦ Action highlighted condition: children saw the experimenter ready the
turntable for action, place the toy on it, and then alternate her gaze
between the toy and the child
◦ Object highlighted condition: the experimenter picked up the toy and
alternated her gaze between the toy and the child.
◦ Test phase: can you show me widget/widge it?
◦ The results showed a mapping between condition and response
◦ They concluded that social cues are responsible for the result.
Thirdly,

◦ A final advantage of the socio-pragmatic theory is that it can explain the

cross linguistic difference in word categories.
◦ For example, the English IN Vs ON does not have an equivalent in Korean.
Rather they have a ‘tight fit’ Vs ‘loose fit’ distinction.
◦ because when the child pays attention to the language around her, she
does not need the universal biases.
◦ One interesting study in this regard found out that unlike European or
American parents, Kaluli people of Papua New Guinea do not engage
children in conversation, as they think children have no understanding of
the world around them.
◦ Still the children of the group learn language just fine.
◦ This is because they are able to tune it to the adult conversations around
them.
criticism
◦ Like all theories, socio-pragmatic learning theory has also been questioned on many
grounds. There are mainly two lines of arguments
◦ One: social learning cannot be the whole story, as word learning can take place without
social scaffolding
◦ The arguments provided by this group is that children are capable of learning words
before intention reading abilities are in place.
◦ Bates et al (1994) used MacArthur Communicative Development Inventory for thousands
of American English speaking children. This inventory assesses the language abilities of
infants and children through developmental stages.
◦ The study revealed that 8 month olds understood 30 words, some even 90 words.
◦ Babies who were taught baby-sign ( kind of a fun form of sign language) language, could
produce signs for food, drink and activities at 6 months.
◦ At this stage children have not fully developed joint attention and intention reading (which
develop between 9 and 12 months)
◦ Hence they must have used other means to learn words.
◦ Another set of criticisms in this line come from studies on autistic children.
Autistic children find it difficult to interpret social cues.
◦ Hence it must be difficult for them to learn words using social cues.
◦ In a lab based study this was found to be true.
◦ Parish-Morris and colleagues showed that children with autism consistently picked
up an interesting object as the referent of a novel word, inspite of the
experimenter clearly signaling a different (boring) object as the intended referent.
◦ However, in real world, children with autism are also found to learn words almost
at par with TD children, even with severe socio-pragmatic difficulties.
◦ Hence, the theory needs a relook.
◦ Two: we do not need socio-pragmatic skills to explain word learning
◦ Samuelson and Smith (1998) cited an influential study of Tomasello and group that
was used as a proof for socio-pragmatic bias and said that the same study can be
shown to prove that it is general purpose cognitive abilites like attention and
memory that are at the root of language learning.
◦ The experiment: (Tomasello et at 1996)
A. Training phase: two experimenter, the child and child’s parent played with novel
objects
◦ Then experimenter A and the parent leave the room.
◦ During this time, experimenter B introduces a fourth object to the child and places all 4
objects in a transparent box.
◦ Now experimenter A and parent return
◦ Experimenter A exclaims: ‘look I see a gazzer. A gazer’. (remember none of the objects had
been given a name, in the sense that children were not told which is called what. Nor
which is gazer)
B. Test phase: children are asked “give me the gazer”. All children preferred the fourth
object as the referent of the word.
◦ The authors claimed that it is by interpreting the experimenter’s intention when she said,
“I see a gazer” to express surprise at the fourth object that they picked up the fourth
object as a referent.
However,
◦ Samuelson & Smith said that it is a problem of interpretation.
◦ They argued that children were influenced by the novelty of events surrounding
the introduction of the fourth object (contextual novelty).
◦ The first three objects were introduced in the same context (with three adults
sitting on the floor). The fourth object was introduced in a different setting (with
one adult, sitting at the table)
◦ This, combined with preference for labelling novel objects with novel words
resulted in what Tomasello reported.
◦ Thus, children attached the new label to the new object, because of setting change
not because of experimenter’s intention.
◦ It had nothing to do children’s sophisticated ability to read the intention of the
experimenter.
◦ To prove their point, they re-run the experiment with minor changes.
◦ In the new version of the experiment:
◦ Experimenter A did not leave the room
◦ The fourth object was introduced in the presence of A, so it was not new to her.
◦ However, a new setting was added here too. The fourth object was introduced
sitting at a table, thathad bright coloured cloth.
◦ Children still preferred the fourth object for gazzer.
◦ Hence, they argued, it is a simple case of cognitive bias.

◦ But, the story does not end here. There are many other interpretations of the same
experiment given by later researchers.
Attentional learning account: Samuelson & Smith

◦ This account puts the child’s cognitive abilities at the centre of the
learning process
◦ This underlines the ‘associative learning’ ability of children
◦ This ability refers to when two cues co-occur in predictable ways.
◦ For example, the word ‘fork’ will always occur in the presence of objects
of a particular shape. Eventually the child learn to associate the shape
with the word.
◦ This type of learning does not depend on any specific word-learning
mechanism.
◦ Rather it is based on general process of ‘perceiving, remembering and
attending’, which are sufficient in themselves.
◦ Within this theory, three more general processes are considered
important, along with associative learning.
◦ These are:
◦ Attention mechanism: children attach labels that attract their attention.
these can be novel objects, salient objects or objects another person is
showing interest in.
◦ Memory: it is important because the learner needs to keep in mind all
the contexts where the label was sued in order to extend it properly.
◦ Cross situational learning: to remember which situations are associated
with a word and to pay attention as to what object or event is common
across all these situations.
◦ For example, let’s go back to the little girl playing with her pet dog, when her
mother exclaims: ‘that’s a dog!’
◦ The attentional learning theory would predict that the girl will associate the word
with the object because that object is the current ‘focus’ of her attention.
◦ Eventually she learns to extend the term to other members of the same species
through a kind of ‘shape bias’. However, as per this theory , this bias is not innate,
rather learnt through the child’s experience in different situations.
Evidence in favour of attentional learning theory

◦ A number of studies have provided support for this theory

◦ Smith & Yu (2008) is another particularly influential study in this regard.
◦ This study used children aged 12 & 14 months and showed that even children this
small can use cross-situational learning, that is central to the attentional learning
theory
◦ Exp:
◦ Training phase: children saw picture-pairs of objects and heard a name. for
example, they saw star shaped and square shaped object and heard the word
‘bosa’
◦ Then they see the star shaped object alongside a round shaped object and hear the
same word ‘bosa’.
◦ Third stage within training phase, they see square and round object and hear
‘gasser’.
◦ None of the training trials provided adequate information about exact name of
each object. However, if the children can map objects and names across trials,
then she can predict the correct names.
◦ Test phase:
◦ They were presented with the novel word and two objects, one target and another
distractor.
◦ Children looked for significantly longer time to the target word, showing
recognition.
◦ Thus, the authors claimed that children remembered object-word mapping across
trials to come up with the correct mapping.

◦ Many other studies also found similar results.

Syntactic bootstrapping

◦ According to this theory all the previous theories are good, but not sufficient. At
least not in case of verbs. They work better for nouns.
◦ For verbs, children need to pay attention to syntax.
◦ We have already seen why learning verbs pose a particularly difficult problem

◦ Gillette and colleagues showed this through a study where adults were shown a
number of videos of mother and child playing, with audio turned off.
◦ Their task was to identify the word the mother was saying at a particular point,
indicated with a ‘beep’
◦ Each word occurred 6 times, in different situations, this giving adequate clue
about them
◦ 45% nouns and 15% verbs were identified by the participants
◦ They concluded that simply observing words paired with events, across multiple
situations, does not yield good result.
◦ Since we DO learn verbs too, this means there is way out.
◦ Syntactic bootstrapping is that way.
◦ According to this theory, mapping of semantics with syntax can be used to
learn verbs.

◦ For example, verbs referring to events having only one participants tend
to occur in sentences with only one noun phrase (I sleep; I fell
etc).intransitive
◦ Similarly, verbs that need two participants, where one participant
causes something to happen to another, they occur with two noun
phrases (I read the book, I ate the cake etc).transitive
◦ Verbs that denote transfer of an object from a donor to recipient will
have three noun phrases (I gave the gift to Mohan).ditransitive
◦ For example, on seeing a dog chase a cat, if the mother says, ‘the dog is
glorping the cat’, the child will understand that the meaning of glorping
refers to the action of the dog and therefore more likely to mean chasing
and not fleeing, since it has two nouns.
◦ Here the meaning is arrived at by taking cues from syntax. Hence this is
called syntactic bootstrapping.

◦ However, this theory also has faced criticisms as even 3 year olds cannot
understand meaning of verbs by just hearing them in a sentence. They
still need a referent.