• Tidak ada hasil yang ditemukan

In this chapter, we define essential aspects of the problem children face in language acquisition. We characterize the severity of a “Projection Prob- lem” by an analysis of the types of evidence children might be expected to need in order to acquire a language. We see that not only is “negative evidence” generally not accessible to children, but also that “positive evidence” in the speech stream is fundamentally indeterminate with regard to the knowledge children must acquire.

Both are always only “indirect.” Given that the evidence available to children is fundamentally insufficient, we consider two possible approaches to “bootstrap- ping” from external (non-linguistic) evidence which they might alternatively try to solve the problem of language acquisition, and show that neither of these can solve the fundamental problem. We conclude that the evidence available to chil- dren can not in itself fully determine the linguistic knowledge they must acquire, and that no form of “bootstrapping” from solely extra-linguistic evidence can solve this problem. We therefore must lookwithin children, beyond the input, for an explanation of language acquisition.

3.1.1 What evidence do children need?

We saw in chapter 2 that children must acquire a generative system which allows infinite possibilities in language, and also rules out infinite impos- sibilities. For this, we would expect them to need both positive and negative evidence.

3.1.1.1 Positive evidence

Children need experience of whatdoesoccur in a language, i.e., “pos- itive evidence”; if they had never heard Hindi, Tulu or English, we would not expect them to acquire it. Speech streams constantly around infants, appearing to provide such evidence directly.

However, evidence presented to children (the “input”) can only be effective if they process it correctly (cf. chapter 6). Even if they do attend to and parse input correctly, this evidence is finite; it will never fully determine the infinite expres- sions which are possible in a language. Adults do not introduce every possible lexical item in English which children will acquire (on average 50,000–250,000 28

words; Aitchison 2003). Adults cannot define what any particular expression will mean. While they can pick out and indicate an individual exemplar of a “dog”

to a child, this positive evidence does not directly determine what the adult is referring to, e.g., some property of the situation containing the dog perhaps. It does not determine that the term will extend from “Snoopy” to a “chihuahua”

(cf. chapter 10). Similarly, the input does not sort out for children the distinction between “grey tabbies” and “great abbeys.”

Since the language learner must attain infinite generative capacity, at best, pos- itive examples, demonstrating possible language, can lead children to a potential chain of inference.

3.1.1.2 Negative evidence

If we were learning chess, we would seek both positive and negative evidence. We would be taught not only specific moves which were possible, but those impossible. Certain formal languages have been shown to be unlearnable without negative evidence, on the basis of induction from positive evidence alone (Gold 1967; Kapur 1994).1

Parents may attempt to provide direct negative evidence by saying, “No that’s not an apple” when a child refers to a pumpkin with the word “apple”; or “No, do not say ‘flied’” to one who just said “he flied over the rainbow.” However, children are not, for the most part, offered direct “negative evidence,” and when it is offered, they frequently rebuff it (chapter 6). When corrections are attempted, parents tend to confront the meaning of an utterance, not its grammatical form – syntactic or phonological – as in (1a) and (1b) (Brown and Hanlon 1970, 49).

Children too insist on relating the parent’s comments to meaning, not form, as in the case of the noun case marking error in (1c);

1. a.Sarah: “There’s the animal farmhouse”

Mother: “No, that’s a lighthouse”

b. Eve: “Mama isn’t boy, he a girl”

Mother: “That’s right”

c.A three-year-old boy: “Her is being mean”

Mother: “No, SHE is being mean”

Child: “Yea, that’s right”

Perhaps parents provide “implicit” negative evidence to children by repetitions of their ungrammatical utterances, or by requests for clarification,2although whether children consult these adult behaviors in building their grammar remains doubtful (cf. chapter 6).

1The result in Gold 1967 depended on assumptions about inductive language learning, e.g., assuming a “conservative” learner, which did not change hypotheses unless contradictory evidence was available (Kapur 1994).

2E.g., Bohannon and Stanowicz 1988, Demetras et al. 1986, Hirsh-Pasek et al. 1984, Penner 1987.

30 c h i l d l a n g ua g e 3.1.1.3 Overgeneralization

If children must acquire language on the basis of positive evidence, what would keep them from overgeneralizing on the basis of the evidence received, and how do they come to eliminate or retreat from incorrect forms in favor of correct alternatives? Children do, in some cases over the first years of life, over- generalize, e.g., overregularizing verbs in (2), although such overgeneralizations appear to be limited (see later chapters, 6, 11). (See Marcus et al. 1992, Clark 1982, Bowerman 1988.)

2. Overgeneralization

“His doggie bited him untied” (M, 5.10) –Telling how “tied-up” man in a TV show was freed(Bowerman 1988, 1982a, b).

“My teacher holded the baby rabbits and we patted them” (Cazden 1972).

3.1.1.4 Indirect negative evidence

A parent may simply “repair” a child’s error bynotusing the child form, thus providing “indirect negative evidence.” Most probably “bited” will not occur in a child’s environment. While a child hears himself referred to as “John”

or “Johnny” or “son,” he is not likely to also hear himself referred to as “Sam.”

However, children must be able to make use of such indirectnegativeinforma- tion. What does not occur must first be perceived, and if it isperceivedit doesn’t necessarily constitute evidence that an expressioncannotoccur. Children who had never heard “The Cat in the Hat” would not be entitled to conclude that this expression was impossible. Presented with a new animal termed a “wug,” they must be able to conclude that there can be several “wugs,” even if they have never heard the plural form of this term before; and they do so productively (Berko 1958, Potts et al. 1979).

A learner must determine when non-occurrence matters. If we are stopped at a red light and the car in front of us does not turn right, this event may or may not be significant. This non-turning event could indicate that a right turn on red is not allowed in this state. It is only so significant, however, if we are interpreting this event with regard to a prior hypothesis about the possibility of right turns on red. Without the existence of this hypothesis, the event of a car stopped at a red light would mean nothing in particular. What does not occur is infinite. Not only is the car not turning right, but it is not making a U-turn, not going straight ahead, not blowing its horn, etc.3Non-occurrence is computationally intractable without a prior hypothesis or expectation that certain occurrencesarepossible in a particular situation.

If children do not hear (3a) or (3b), but do hear (3c), how do they know whether (3a), (3b) or both are significant negative evidence?

3. a. Is the cat who is in the puddle in the hat?

b. *Is the cat who in the hat is in the puddle?

c. Is the cat who is in the hat in the puddle?

3Thanks to Guy Carden for the example.

Computing indirect negative evidence depends on pre-determined hypotheses regarding “possible” language (see in chapter 4 no. 16 [p. 58] on abduction). Yet this is what we are trying to explain: how do children come to know the significant hypotheses about their language?4(Cf. chapter 6.)

3.1.1.5 Primary linguistic data (PLD)

In the Initial State, infants for the first time hear sounds in the speech stream and begin to consult these physical phenomena in the process of language acquisition.5

Children born in the United States may hear an utterance like (4). Without knowledge of language, this corresponds simply to an acoustic stimulus, visually represented in (5); after phonetic analysis, it corresponds to (6). Children born in Sri Lanka may hear an utterance in Sinhala meaning (7), corresponding to an acoustic stimulus represented in (8), represented as (9) after phonetic analysis.

English

4. This is a story about Cinderella 5.

6. ðis iz ə stor`i əbawt sindərelə

Sinhala

7. This is a story about mother.

8.

9. me: katandəre: amma gænə this story mother about

We assume that all normal-hearing children, from birth or earlier, will be continually exposed to a wide range of ambient language (ranging from single word utterances to utterances more complex than [4] or [9]) not only between child and parent, but between adults and between other children, in all of the situations of children’s normal life. In this sense, they are thoroughly “awash” in potential evidence.

However, the sound input, e.g., that corresponding to (4) or (7), recorded in (5) and (8), consists simply of “[v]ariations in air pressure in the form of sound waves” which “move through the air somewhat like ripples on a pond” (Ladefoged 1993, 160). The sounds themselves are “fleeting and transient . . . Even during the brief existence of a sound . . . there is nothing that can be seen; there is no visible connecting link between a speaker and listener. There is air around, but it

4See Kapur 1994 for analysis of indirect negative evidence and linguistic theory; Bowerman 1988;

Marcus et al. 1992; Morgan and Travis 1989; Randall 1992.

5The term “Initial State” does not involve a temporal or age-based notion, but rather refers to the state of beingprior to experience. For the adult acquiring a new language we may assume that the adult is once again in an Initial State (Flynn and Lust, 2002). See chapter 4.

32 c h i l d l a n g ua g e

is not normally possible to see any changes in the condition of the air when it is conveying a sound” (Ladefoged, 1996, 1).6

Phonetic analyses in (6) and (9) represent the linguist’s transcription of this speech. The adult who knows either language – English or Sinhala – hears sound in (5) or (8), analyzes it, digitizes it (see chapter 2), remembers and reconstructs it according to adult “intuitions,” exposing its units.7 Unless we are bilingual, we can so transform the speech stream for only the language we know; the other remains just sounds. Children, initially, must find a way to map from (5) to (6), or (8) to (9), without knowing any language, without “adult intuitions,” and without being taught any form for the representation of these sounds.

We speak of this initial input to the language learner such as (4)–(5) or (7)–(8), and the full range of ambient language in all contexts, as “primary linguistic data”

(PLD). It is the grist for the mill of language acquisition.

10. Primary linguistic data(PLD)

The actual original finite language data to which children are exposed, and from which they must map to knowledge of a specific language; a combination of sound and extra-linguistic experience.

On the basis of such data, children must somehow eventually acquire a specific language. We call this the “Projection Problem.”

11. The Projection Problem

The problem of mapping (“projecting”) from the finite initial specific experiences of PLD to knowledge of a specific language. (cf. Baker 1979)

In general, children may be said to need to “crack the code” of the physical stimu- lus to which they are exposed, and project from the data to linguistic knowledge.

3.1.1.6 Cracking the code

To solve the Projection Problem, children must convert the acoustic stimulus from a continuous stimulus, e.g., (5) or (8), to a discontinuous or digital (unit-based) representation, e.g., (6) or (9). That is, children must discover the units that function in the language and their organization. We have seen that discrete units exist at several “levels of representation” in language knowledge (cf. figure 2.3).

6Spectrographs capture and display the acoustic energy of the input speech as a function oftime, frequencyandamplitudeof the sound waves (Crystal 1997a, 136–137, and Ladefoged 1993, 191–

214.)

7The linguist’s phonetic transcription is used here in (6) or (9) to overcome the inconsistencies of spelling, and to capture the “real” sounds of the utterances. They use some version of a “phonetic alphabet” (e.g., some version of an International Phonetic Alphabet, IPA). The reader not familiar with this alphabet may consult Fromkin and Rodman 1998, Ladefoged 1993, Pullum and Ladusaw 1986, or Cipollone et al. 1998. Study of first language acquisition requires the adoption of a phonetic alphabet because it requires capture of the precise sounds of a child’s utterance in an unambiguous way.

12.

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

[ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ] [ ]

As in examples (5) and (8), children must discover the sounds (which combine into words), the words (which combine into phrases), the phrases (e.g., subjects and predicates which combine to form clauses), and the clauses and clause com- binations (which form sentences) (chapter 2).

3.1.1.7 Summary

While both positive and negative evidence appear to be necessary for children to acquire a language, neither appears to be directly available to them.

3.2 The nature of the evidence: searching the speech