Semantic Development

Levels of meaning

      1. Lexical semantics - word meaning (lexical contrast, synonyms, etc.)

      2. Propositional semantics - sentence meaning

            Thematic relations (agent, theme, etc.)

            Scope relations, e.g., ‘All of the arrows didn’t hit the target’

                  (¬ ∀ arrows) hit (target) vs. (∀ arrows) ¬ hit (target)

Quine’s Problem (Word & Object 1960)

Willard Van Orman Quine proposed the following scenario, which he refers to as a case of ‘radical translation’:

A rabbit scurries by, the native says ‘Gavagai’, and the linguist notes down the sentence ‘Rabbit’ (or ‘Lo, a rabbit’) as tentative translation, subject to testing in further cases. The linguist will at first refrain from putting words into his informant’s mouth, if only for lack of words to put. When he can, though, the linguist has to supply native sentences for his informant’s approval, despite the risk of slanting the data by suggestion (29).

For, consider ‘gavagai’. Who knows but what the objects to which this term applies are not rabbits after all, but mere stages, or brief temporal segments, of rabbits? In either event the stimulus situations that prompt assent to ‘Gavagai’ would be the same as for ‘Rabbit’. Or perhaps the objects to which ‘gavagai’ applies are all and sundry undetached parts of rabbits; again the stimulus meaning would register no difference (51-52).

The intrepid linguist, and all children, must solve Quine’s problem of radical translation. Should children use the English word ‘Rabbit’, an Upper Chehalis word like s-L’aláš ‘deer; it is a deer’ or the Japanese word usagi ‘rabbit stuff’ to translate ‘Gavagai’? Quine argued that the visual scene, his ‘stimulus situation’, did not provide enough evidence to decide between these choices. He suggests, instead, that the evidence is supplied by the linguistic context in which the words are used in each language. English marks common nouns with articles, Upper Chehalis marks its words with aspect prefixes and Japanese marks its nouns with numeral classifiers. Quine’s semantic holism thesis is that the interpretation of each word in a language is dependent upon the interpretation of other words in the language.

General issues

      1. Semantic features vs. holism (Quine - Word and Object)

            Semantic features similar to elements ‘the ultimate constituents of meaning’

                  E.g., bird {wings, feathers, beak, ...)

            Holism - the meaning of each word depends on the meaning of other words

      2. Perceptual features (Clark 1973) vs. Functional features (Nelson 1974)

            Perceptual features: shape, smell, color, feel, sound; ball - [+ round]

            Functional features: uses, interactions; ball - s.t. to throw, bounce

      3. Definitional (Fodor 1975) vs. Clusters (Wittgenstein 1953)

            Definitional - necessary and invariant features; ball - [+ round]

            Clusters - family resemblances (Rosch 1973); ‘game’, ball - prototypical features

      4. Innate (Fodor 1975) vs. Learned (Anglin 1977)

            Innate - universal and permanent features; ball - always [+ round]

            Learned - language specific and adjustable; ball - this particular ball

Reference Theories of Word Meaning

Ask anyone about the meaning of a common noun such as chair or carburetor and chances are they will point to an instance of such an object if it is handy. This type of ostensive definition satisfies our intuitions about what words mean by pointing to something the word denotes, and has prominent advocates such as John Stuart Mill. Does reference provide a valid theory of meaning?

Referential theories of meaning define the meaning of words or phrases in terms of the things (or actions, qualities, relations, etc.) that the word or phrase denotes. If we ask someone for a cup and they hand us a sponge, we are apt to think they did not understand the word cup. More precisely, we can equate the meaning of a noun with the set of things that are that object.

      CUP = the set of things that are cups

      SPONGE = the set of things that are sponges

Just to keep things straight, I will put a word in italics when I mention it and will use capital letters to refer to the meaning of a word. So CUP stands for the meaning of the word cup. A cup is the thing referred to by the English word cup.

Problem 1: Semantic Intension

A referential theory of meaning accounts for our ability to point to the things that words denote, but it has critical shortcomings. The philosopher Gottlob Frege pointed out a critical flaw using the phrases morning star (Phosphoros) and evening star (Hesperos). These phrases have the same referent, but different meanings. Using them in a sentence makes this difference obvious.

      The morning star is the morning star.

      The morning star is the evening star.

The first of these sentences does not tell us anything new, while the second sentence does. A referential theory of meaning does not predict this difference. Why not?

Frege distinguished between a word’s semantic extension and its semantic intension. The semantic extension of a word is the set of things the word denotes, while its semantic intension is the concept or meaning of the word. The meaning of a word determines the things it refers to, but it cannot be equated with its denotations.

Problem 2: Words without Semantic Extension

Not all words or phrases have a semantic extension, e.g. Santa Clause, witch, phlogiston. We are investigating whether the word meaning has a semantic extension. How do such words invalidate reference as a theory of meaning?

Problem 3: Opaque Contexts

Sentence meaning displays a similar distinction between extensional and intensional meanings.

      George is the best student in linguistics.

      I dreamed that George is the best student in linguistics.

The first sentence asserts that George is one and the same shining paragon of linguistics. In other words, it asserts that George and the best student in linguistics have the same semantic extension. If George is the best student in linguistics then the first sentence is true. Assuming the first sentence is true, however, does not guarantee the truth of the second sentence. The truth of the second sentence depends on what I dreamed and not on the truth of George being the best student in linguistics. We can usually equate the semantic extension of a sentence with its truth value, but we see that the truth of the embedded clause has no effect on the truth value of the second sentence. Propositional attitude verbs (dream, believe, want) create opaque contexts where the truth value of the embedded clause is unrelated to the truth of the whole sentence. We need something more than the reference of the embedded clause to understand meaning in opaque contexts.

How would the four theoretical approaches that Aslin & Pisoni (1980) outline for phonological development apply to the acquisition of intensional meaning?

      1. Semantic Learning Theory (behaviorist)

      2. Attunement Theory (constructionist)

      3. Universal Theory (innatist)

      4. Maturational Theory (restructuring)

Predictions

These predictions beg the question of what we know about semantic contrasts in the world’s languages. Meaning is among the least studied feature in linguistics. Few cross-linguistic studies of semantics exist. Recent studies reveal profound differences in meaning exist across languages.

Melissa Bowerman (1989) pioneered the cross-linguistic study of topological encoding. She initially looked at the encoding of in and on in English, Dutch, Spanish and Berber.

Other researchers have noted that similar differences exist in verb meaning. The separation domain presents a primary example of cross-linguistic differences in verb meaning (Pye 1996).

Cross-linguistic differences in meaning have been documented in many domains. Such differences pose the central induction problem for language learners who are restricted to positive evidence. Little research exists on semantic variation among adult speakers.

References

Bloom, P. 2002. How Children Learn the Meanings of Words. Cambridge, MA: MIT Press.

Bowerman, M. 1989. Learning a semantic system: What role do cognitive predispositions play? In M. L. Rice and R. L. Schiefelbusch (eds.), The Teachability of Language, 133-169. Baltimore: Brooks.

Bowerman, M. and S. C. Levinson (eds.). 2001. Language Acquisition and Conceptual Development. Cambridge: Cambridge University Press.

Kinkade, M. D.1983. Salish evidence against the universality of ‘noun’ and ‘verb’. Lingua 60: 25-39.

Pye, C. 1996. K’iche’ Maya verbs of breaking and cutting. Kansas Working Papers in Linguistics 21.87-98.

Quine, W. V. O. 1960. Word and Object. Cambridge, MA: MIT Press.

Theories of semantic development

1. Semantic Feature Hypothesis (Eve Clark 1973)

      a. children begin with superordinate features and learn subordinate features

            [ANIMATE] –> [ANIMATE, CANINE]

b. predicts early overextensions based on perceptually salient features, e.g., movement, shape, size

antonyms like ‘same’ vs. ‘different’ and ‘more’ vs. ‘less’ are first acquired with similar meanings

            Children first acquire ‘unmarked’ features, e.g. ‘big’ vs. ‘little’

2. Functional Core Concept theory (Katherine Nelson 1974)

      a. children begin with a concept that is not purely perceptual, and attach a word to it

      b. follow four steps (Table 8.13, p. 400)

            1. identify the object, e.g., ball

            2. recognize context-bound functional features, e.g., throw, roll, bounce

            3. identify core ‘invariant’ functions, e.g., roll, bounce

            4. acquire word

      c. predicts words acquired early would be underextended

      d. Nelson does not state how the theory can be extended to adjectives or verbs

Both theories assume a definitional approach to word meaning

Both theories treat word meanings in isolation from other words

3. Prototype theory (Eleanor Rosch; Anglin 1977, Bowerman 1978a)

a. definitions of key terms (Anglin 1977:27)

      1. Term of reference ‘a word. . . which denotes or refers to real objects’

2. Extension: ‘all the objects which an individual is willing to denote’ with a term of reference

3. Intension: ‘the set of properties which an individual believes to be true of the instances of the category denoted’ by a term of reference

4. Concept: ‘all of the knowledge possessed by an individual about the category of objects denoted’ by a term of reference. ‘This knowledge includes both knowledge of extension and knowledge of intension.’

b. acquisition

      1. child forms a perceptual schema of an object analogous to a prototype

            the schema begins at an intermediate semantic level;

            it also includes core functional features

the child’s extensional knowledge is not well coordinated with it intensional knowledge

                  Children’s word overextensions are based on perception

Their word definitions (intensions) are primarily functional, e.g. ‘apple’ st eaten

2. school-aged children learn to coordinate their lexical knowledge into hierarchical systems

Acquisition Studies

      1. Anglin (1977) noun acquisition Study 1

            a. order of acquisition–superordinate, intermediate, subordinate

Brown (1958b): children start with intermediate level, e.g., ‘money’, not ‘thing’ or ‘dime’

b. problem: child could use an intermediate adult word as a superordinate or subordinate term

            c. Anglin’s stimuli:

            He used 4 sets of 3 pictures rated by adults for each word

                  3 ‘prototypical’ animals: King Charles spaniel, African elephant, cat

                  3 ‘peripheral’ animals: bullfrog, monarch butterfly, marsh hawk

            20 children between 2 and 5 years old participated

      The children were asked to name each picture in the set and then asked ‘What are all these?’

            d. Results (Table 8.14, p. 407)

                                          Child

‘There is neither a unidirectional specific-to-general progression in vocabulary development nor a unidirectional general-to-specific progression.’ (67)

            e. factors governing semantic development

                  1. relevance–is the word important to the child

                  2. function–groupings that are useful to the child

                  3. frequency–how often does the child hear the word

      2. Study 2 extension

            a. Anglin’s stimuli:

            b. participants

                  18 children with ages 2;6 to 4;0

                  18 children aged 4;6 to 6;0

                  18 adults

            c. each asked ‘Is this a (e.g.) fruit?’ when shown instances and non-instances

            d. Results

                  1. children showed both overextensions and underextensions (most frequent)

                  2. Denied woman was an animal and praying mantis and caterpillar were animals

            e. Explanation

                  1. individual variation

                  2. the specific concept

                        Flower overextended to elephant ear, coconut, and a philodendron

                        Plant was usually underextended; rejected trees as examplars of plants

                  3. the stimuli

                        Peripheral examplars were less likely to be correctly identified

      3. Study 3 intension

            a. 14 children between 2;8 and 6;7

b. each asked to discuss the meaning of 12 words: dog, food, flower, vehicle, animal, apple,

                  rose, car, collie, fruit, plant, and Volkswagon

                  What is a ____?

                  Tell me everything you can about ______.

                  What kinds of _______s are there?

                  What kind of thing is a _____?

                  Tell me a story about a _____.

            c. children were not able to give a set of defining properties for a word’s meaning

                  e.g., Peter (2;8) for ‘dog’, ‘it goes woof, barks’–instance oriented

Bowerman (1978a)

a. diary study of her two daughters, Eva and Christy

b. findings (Table 8.15, p. 410)

      Eva ‘moon’ used for ball of spinach, hangnails, the letter D

      Overextensions of non-nominals based on child’s subjective experiences

            There!       Experience of completing a project

c. both complexive and noncomplexive overextensions occur from the beginning

d. complexive use was somewhat more frequent for action words than for object words

e. the girls assigned different meanings to the same words, e.g.,

Christy ‘on-off’: ‘any act involving the separation or coming together of two objects’ (p. 271), e.g., getting socks on or off; getting off a spring-horse; taking pop-beads on and off; separating stacked cups; putting phone on hook; etc.

Eva ‘off’: ‘removal of clothes and other objects from the body’ (p. 271), e.g. for shoes, car safety harness, glasses, pacifier, bib, diaper, etc.

f. the complexive examples suggest an underlying organization around a prototype

Eva used the word ‘kick’ on seeing a picture of a kitten with a ball near its paw; watching cartoon turtles on TV doing the can-can; pushing her stomach against a mirror.

g. Bowerman concludes (p. 278) ‘These findings suggest that there is less discontinuity between child and adult methods of classification than has often been supposed.’ She does not explain how her daughters acquire adult extensions.

Solving Quine’s Problem (Word & Object 1960)

‘radical translation’–does ‘gavagai’ refer to the whole object, the collection of undetached parts, or a stage?

1. Hypothesis testing (induction) can lead to radically different interpretations

2. Solution–universal constraints that predispose children to the proper adult extensions

      a. Whole object bias (Markman 1994; also Golinkoff, Mervis & Hirsh-Pasek 1994 JCL)

A novel label is likely to refer to the whole object and not to its parts, substance, or other properties.

      b. Taxonomic bias

Labels refer to objects of the same kind rather than to objects that are thematically related.

      c. Mutual exclusivity bias (Also Clark 1987, The Principle of Contrast)

            Words are mutually exclusive.... Each object will have one and only one label.

3. Problems

      a. The proposed constraints cannot be extended to other lexical categories, e.g., verbs

      b. The constraints are vague

            i. the specific object or objects of the same kind

            ii. ‘the same kind’

      c. The constraints are false (Gathercole 1989 JCL, e.g., couch/sofa; -s/-z)

Recent Studies

Soja, Carey & Spelke (1991 Cognition 38.178-211 ‘Ontological categories guide young children’s inductions of word meaning)

a. Examined children’s understanding of the distinction between some sand and some dog. Is the distinction between objects and substances innate?

      b. Subjects: 24 two-year-olds (x = 2;1)

      d. Stimuli: novel objects and substances, e.g., apple corers, plumbing fixtures; coffee

      e. Results: percent object responses

      f. Conclusion:

            i. Children can distinguish between objects and substances by two years of age

                  They generalize by shape when the referent is an object

                  They generalize by texture when the referent is a substance

ii. Children do not rely upon linguistic cues to make the distinction between objects and substances

Imai & Gentner (1997 Cognition 62.169-200 ‘A cross-linguistic study of early word meaning: universal ontology and linguistic influence’)

a. Subjects: American and Japanese

      14 children at 2;1; 15 children at 2;8; 14 children at 4;2; 18 adults

b. Stimuli

c. Results: percent object responses

d. Conclusions:

      i. Both American and Japanese subjects distinguish between objects and substances

ii. American subjects had a higher proportion of shape responses in simple object and substance trials

iii. Japanese children showed no bias for simple object trials; American adults have an object bias; Japanese adults have a substance bias

      iv. Japanese subjects show a clear substance bias for substance trials

v. All children make a distinction between objects and substances, but language influences where the division is made

Landau, Leyton, Lynch & Moore (PRCLD 1992)

a. Subjects: 3;0

b. Stimuli

	This is a dax.	Is this a dax?

c. Conclusion: Children’s attention to object shape is subject to their theory about the object

Verb Meaning (Choi & Bowerman, 1991 Cognition ‘Learning to express motion events in English and Korean’)

      Talmy (1985) observed that languages may be divided into:

      Satellite-framed languages (e.g., English)            Verb-framed languages (e.g., Spanish)

      The bottle       floated       into the cave         La botella        entró    a la cueva  flotando

      Figure  Motion+Manner    Path                Figure  Motion+Path                Manner

      Korean verbs also conflate motion and path

a. Subjects: Choi visited 4 Korean children (14-34 months old) every 3-4 weeks; compared with Bowerman’s diary study of 2 children learning English

b. Results:

      i. Children displayed cross-linguistic differences by 17-20 months of age

      ii. Korean children extended ppayta to flat surfaces or tight clothing

      iii. English children extended in to a position between two objects

c. Conclusion: Contradicts theories that children come equipped with domain-specific semantic primitives for space and spatial motion (e.g., Landau & Jackendoff, Behavioral and Brain Sciences 16, 1993)

1. Hypothesis testing (induction) can lead to radically different interpretations

2. Solution–universal constraints that predispose children to the proper adult extensions

      a. Whole object bias (Markman 1994; also Golinkoff, Mervis & Hirsh-Pasek 1994 JCL)

A novel label is likely to refer to the whole object and not to its parts, substance, or other properties.

      b. Taxonomic bias

Labels refer to objects of the same kind rather than to objects that are thematically related.

      c. Mutual exclusivity bias (Also Clark 1987, The Principle of Contrast)

            Words are mutually exclusive.... Each object will have one and only one label.

3. Problems

      a. The proposed constraints cannot be extended to other lexical categories, e.g., verbs

      b. The constraints are vague

            i. the specific object or objects of the same kind

            ii. ‘the same kind’

      c. The constraints are false (Gathercole 1989 JCL, e.g., couch/sofa; -s/-z)

Brown (2001 ‘Learning to talk about motion UP and DOWN in Tzeltal’ in Bowerman and Levinson, Language Acquisition and Conceptual Development)

Tzeltal Mayan has an Absolutive system of spatial reckoning based on the predominant uphill/downhill lay of the land (along a South (uphill)-North (downhill) axis)

                              Verb                Position                       Relation

            UP            mo ‘ascend’   kaj ‘be above’            ajk’ol ‘uphill’

            DOWN     ko ‘descend’  pek’ ‘be low down’   alan ‘downhill’

      e.g., ‘The rain is descending’ (i.e., coming from the south)

              ‘It (a puzzle piece) goes in downhillwards’

English has a Relative system of spatial reckoning (e.g., front/back, left/right)

Tzeltal children begin to use the Absolute vocabulary in the one- and two-word stage to refer to vertical relations (with verbs of falling and climbing) as well as horizontal relations (movement between houses or of toy cars on the flat patio)

The core semantics for the children’s verbs mo/ko are restricted at first to local places (particular houses in the local compound). The children generalize these verbs to novel contexts such as moving objects into trees, onto beds, up onto the roof, and to and from particular houses

Conclusions

Tzeltal children learn the language-specific constraints on verb use in context by a process of induction across instances of use, instances which provide both vertical contexts and landslope contexts for the same words

General lexicalization properties of the language rather than innate cognitive universals provide the basis for language-specific hypotheses about possible word meanings

Further examples of semantic specificity of Tzeltal verbs

Pye, Loeb & Pao (1996 ‘The Acquisition of Breaking and Cutting’ CLRF 27.227-236.)

Selectional restrictions on verbs are one of the main sources of semantic differences between languages. The breaking domain provides a good example of this variation. English speakers apply the word break to a wide array of objects: sticks, stones, bread, furniture, and machines. At the same time, English insists on marking the difference between breaking and tearing things. We break one and three dimensional objects, while we tear two dimensional objects.

                  Other languages break up this domain in different ways.

English

            Subjects:   16 American children (2;2-5:5)/22 adults (shown in parentheses)

            Results:     Percentage of children (adults) responding with break

K’iche’

            Subjects:   6 K’iche’ children (4;0)/5 adults (shown in parentheses)

            Results:     Percentage of children (adults) responding with q’upi:j

Chinese

            Subjects:   8 Chinese children (3;0-5;5)/13 adults (shown in parentheses)

            Results:     Percentage of children (adults) responding with duàn