• Tidak ada hasil yang ditemukan

The role of the input 1. Cue-category mappings

Dalam dokumen Studies on Language Acquisition 25 (Halaman 74-86)

language processes Liz Temple

4. The role of the input 1. Cue-category mappings

these nouns refer to professions with male-female gender differentiation (e.g. der Lehrer – die Lehrerin [male resp. female teacher]) and are thus associated with the natural gender principle. In the case of -o, one has to take the perceptual salience of the final full vowel into account which is untypical for German and could very well lead to an early establishment of this gender regularity. The extraordinary poor results for chen, tion, and -ung are surprising, considering the fact that these suffixes categorically indicate the gender of the respective noun, and will lead to a closer look at the input frequencies of nouns with respective formal gender indicators the subjects had been exposed to.

4. The role of the input

– Decision frequency: Cues for a frequently processed linguistic category (such as gender indicators in German) have a high general level of acti-vation.

– Detectability: Salient cues should be learned before non-salient ones.

– Availability: The extend to which a certain cue is available in the input.

– Reliability: The degree to which a cue correctly indicates the classification on the cases it is present.

– Validity: The product of availability and reliability.

– Conflict validity: The extend to which a cue helps a candidate to ‘win out’

over others in case of conflicting cues.

According to McDonald’s (1989) cue-category-mapping concept, these six properties define the order in which the individual cues are being acquired, as well as their power to influence linguistic decision in cases where com-petent speakers are confronted with a new categorization task like gender assignment to previously unknown nouns such as loans or nonce words.

On the basis of her analysis of various empirical studies on the acquisition of grammatical gender in German, McDonald (1989: 392) concludes that overall cue validity determines the acquisitional order of the different indi-cators to gender. Whereas cue availability constitutes the most important property in the early stages of learning, later on cue reliability gains more and more significance. Only in the late stages of acquisition, the conflict validity of a certain gender indicator becomes the decisive factor for the fine tuning of the learner internal gender assignment system.

4.2. Input frequencies of gender indicators and the acquisition of gender In order to investigate in what way the described cue-category-mapping properties account for the acquisition of gender by adult GFL learners, the results of gender assignment Study 1 (3.2.) will now be compared to the subjects’ input of nouns with the different gender indicators. Fortunately, the highly controllable nature of the acquisition process of the students who participated in the present study15allows for quite an accurate estimation of the L2 total input16. The nouns in the textbook up to the chapter which had been covered by the 1st year students at the time of the gender assignment study were thus registered according to their actual type and token frequen-cies and subsumed under the same gender assignment types and formal regularities described above17.

Figure 3. Correct assignments for the three gender classes – Study 1

Figure 3 depicts the overall accuracy rates for the three gender classes over all items and all subjects. Gender assignment was most accurate for the masculine items (49.3 %), followed by the feminine nouns (43.7 %), and the neuters (36.7 %)18. T-tests show that these differences are statistically significant (masc vs. fem: t (90) = 3.22; p = 0.002; masc vs. neut: t (90) = 6.1; p < 0.000; and fem vs. neut: t (90) = 3.92; p < 0.001). The superiority of masculine items is somewhat inconsistent with the results of Mills (1986: 89) who found that children acquiring German as their first lan-guage perform best on feminie gender assignments. The poor results for the neuter items, however, are parallelled by an investigation with adult L1 speakers of German, where Jacobsen (1999: 507) measured significantly longer latencies for neuters compared to the other two genders in an object naming task following gender marked article primes.

The question thus arises whether the three gender classes also substan-tially differ in the subjects’ input. Table 3 gives an overview over the type and token frequencies of nouns belonging to the three gender classes in the input to the students up to the time of the study. The masculine items which were classified most accurately in the study also have the highest type and token frequencies followed by feminine nouns, while neuters with their rather poor gender assignment were least frequent in types and tokens. The table thus clearly demonstrates how the results for the three gender classes directly correspond to the overall type and token frequen-cies of masculine, feminine and neuter nous in the input.

% correct

49,3

43,7

36,7

Masc nouns Fem nouns Neut nouns 30

35 40 45 50

A comparison of assignment latencies, however, discloses a different pic-ture (Fig. 4)19. The average gender assignment to feminine nouns (2193 ms) took the subjects significantly less time than to masculine (2338 ms) and to neuter (2273 ms) nouns (fem vs. masc: t (90) = 5.79; p < 0.000, fem vs.

neu: t (90) = –3.45; p = 0.001, and masc vs. neu: t (90) = 2.92; p = 0.004).

Figure 4. Latencies for the three gender classes – Study 1

A closer inspection of the distribution of the input nouns over the various formal and semantic gender regularities assists the interpretation of these findings. Table 4 considers only those nouns in the input that correspond to one (or more) of the gender regularities described above. A comparison of

2338

2193

2273

Masc nouns Fem nouns Neut nouns 2100

2150 2200 2250 2300 2350

latencies in ms

Table 3. Distribution of the three genders in the input and assignment accuracies – Study 1

Masculine nouns Feminine nouns Neuter nouns

Total Ratio* Total Ratio* Total Ratio*

Types 210 1 183 0.87 161 0.77

Tokens 696 1 641 0.92 536 0.77

Correct gender 49.3 % 1 43.7 % 0.89 36.7 % 0.75 assignment

Note: * This figure relates to the ratio of masculine : feminine : neuter nouns.

the number of different gender indicators for each of the three gender classes shows that there are more gender regularities for feminine (7) than for neuter (5) and for masculine (4) nouns. At the same time, the number of types as well as tokens in the students’ input which form exceptions to these gender regularities is largest for masculine nouns. We are thus con-fronted with the highest rule-exception-ratio for masculine nouns, second highest for neuter and lowest for feminine nouns, a difference which is clearly reflected in the assignment latencies for the three gender classes.

This overview of overall assignment of nouns to the three gender classes seems to indicate that – at least in the early stages of GFL acquisition – the total number of nouns in the input belonging to the individual gender classes has a significant influence on assignment accuracy, whereas the latencies seem to depend more on the relation between the number of nouns that follow gender regularities compared to the number of excep-tions for the respective gender class.

Seman. reg.

Formal reg.

Monosyll.

Nomin. verbs

Analogies Excep. formal

Excep. monosyll.

Tokens Latencies Types

Correct assignment

Figure 5. Influence of input types and tokens on gender assignment accuracy and latencies – Study 1

Table 4. Nouns corresponding to gender regularities vs. exceptions in the input Regular gender assignments Exceptions Rule-exception-ratio

No. of reg. Types Tokens Types Tokens Types Tokens

Masc nouns 4 57 180 47 149 1.21 1.21

Fem nouns 7 132 395 15 65 8.8 6.08

Neut nouns 5 20 77 8 33 2.5 2.33

Returning to McDonald’s (1989) predictions about the development of cue-category mappings in the acquisition process, we will now investigate the influence of type and token input frequencies for nouns conveying individual gender cues on the accuracy and reaction times for the corresponding items in the study. Figure 5 superposes the values for the type and token frequencies of the nouns in the input as well as for assignment accuracy and latencies for the types of gender regularities described above20. The graph depicts that the token frequency of nouns in the input intersects counter-symmetri-cally with the respective assignment latencies. A high token frequency is thus reflected in shorter reaction times, while types of gender regularities with relatively few tokens yield longer latencies. The values for the number of noun types following a certain gender regularity in the input, however, run basically parallel to the accuracy of gender assignment, indicating that high type frequencies for a certain gender regularity are parallelled by good gender assignment results to the nouns carrying the respective gender cue. The only exception to this pattern are the semantically motivated gen-der assignments. Here, the conceptual salience of the association between a semantic feature of the referent and the grammatical gender of the corre-sponding noun obviously strengthens these gender cues in spite of a rela-tively low type frequency in the input.

Figures 6a and 6b illustrate that this general tendency of concurrence between token frequency and assignment latencies on the one hand and be-tween type frequency and accuracy on the other hand also holds for the various formal gender regularities21. Although there are some minor devia-tions from the described pattern (e.g. for the nouns ending in -keit, -tion, -us, or -ung)22, the data in general support McDonald’s (1989: 392–393) claim

-a -at

-chen -e

-el -en

-er

Ge--ik -in

-keit -ment

-o -tion

-um -ung

-us Ex-e Ex-el

Ex-en Ex-er Tokens Latencies

Figure 6a. Influence of input tokens on assignment latencies – Study 1

that cue availability (i.e. frequency) is the most decisive factor for the development of cue validity in the early stages of language acquisition for an instructed L2 learning setting as well.

The results of the described study on gender assignment by adult stu-dents of German in an instructed learning environment thus show that even beginning GFL learners have already extracted some of the most valid gen-der regularities from the input and are applying this knowledge to the gengen-der categorization of unknown items. In the early acquisitional stages, input frequency and availability of the individual gender indicators are obviously more important than their absolute reliability for establishing cue validity.

The poor results for nouns forming exceptions to the most valid gender regularities for these learners (the association of word final -e with femi-nine and of -er with masculine gender) indicate that at this learning stage, gender cue strength is not yet influenced by the concept of conflict validity, as these gender cues even dominate deviant individual nouns with high in-put frequencies (see Fig. 2). A comparison of type and token frequencies for input nouns conveying certain gender regularities with the accuracy rate and the assignment latencies in the study furthermore demonstrates the tendency of high token frequencies to correlate in particular with shorter latencies, while high type frequencies are more closely related to gender assignment accuracy. These findings correspond to the prediction from Bybee’s (1988, 1995) Network Model, according to which the type rather than the token frequency of a morphological pattern is the decisive factor for its productivity.

-a -at

-chen -e

-el -en

-er

Ge--ik -in

-keit -ment

-o -tion

-um -ung

-us Ex-e Ex-el

Ex-en Ex-er Types Accurracy

Figure 6b. Influence of input types on assignment accuracy – Study 1

4.3. Resolution of semantic vs. formal gender cue conflicts

Although the development of cue conflict validity marks the last stage in the acquisition of cue-category mappings in McDonald’s (1989) concept, GFL learners nevertheless do have to deal with conflicting gender indica-tors from the very beginning of their language learning process on. Here, conflicts between semantic and formal regularities (e.g. das Mädchen [girl]: + formal cue, – semantic cue; der Junge [boy]: + semantic cue, – formal cue) are of special relevance for beginning learners. Depending on the language to be learned, the resolution of semantic vs formal gender cue conflicts will be biased in favor of the type of gender regularity predomi-nant in that language. In a series of experiments, Karmiloff-Smith (1979) demonstrated an unequivocal dominance of phonological over semantic regularities for French. For German, however, the picture is not that clear.

For native speakers at least, Köpcke (1982: 111) places the semantic gender regularities at the top of his gender cue hierarchy, a claim that is supported by a nonce word experiment conducted by Lang (1976). In this study adult speakers of German first assigned gender to nonce words on the basis of formal gender indicators, but later changed some of these gender assign-ment when the nonce words were given meanings associated with a certain gender, such as flower names (fem), alcoholic beverages (masc) or names of dances (masc). For children learning German as their L1, however, Mills (1985, 1986) found that both semantic and formal gender regularities are acquired simultaneously and that no gender cue type dominates over the other, a claim supported for children learning German as their L2 in an error analysis study by Pfaff (1992).

4.3.1. Study 2

In order to investigate how adult learners of GFL tackle the problem of conflicting gender cues, a study was conducted in which the stimulus items represented male and female persons, while the validity of the formal gen-der indicators on the nouns was varied systematically.

Subjects

The subjects were the same student groups who participated in Study 1 (2.1).

Stimuli

16 nouns that name persons or professions (8 male, 8 female) were selected and supplemented by drawings representing these persons. The formal gen-der regularities on the nouns either reinforce semantic gengen-der indication (der Kellner [waiter]) or created a conflict between semantic and formal gender cues (der Kunde [customer]). In order to control the influence of memorized article + noun combinations, half of the stimuli consisted of items known to all 91 subjects and the other half were constructed nonce words. Gender indicators were distributed over the stimulus items as fol-lows: 4 fem nouns conveyed convergent gender cues (-e and -in, 2 items each), 4 fem nouns conveyed conflicting gender cues (-er and -us, 2 items each), 4 masc nouns conveyed convergent gender cues (-er and -or, 2 items each), and 4 masc items conveyed conflicting gender cues (-e and -is, 2 items each).

Procedure

The influence of the natural gender of the referent on the assignment of grammatical gender to the respective noun can best be ascertained by a direct comparison of gender assignment to (a) solely the nouns and (b) the nouns together with the visual depiction of the referent. In order to avoid presenting each item twice to the subjects, a counter-balanced design employing two test sheets was chosen. Test sheet 1 included the items 1–8 as written nouns only, and gave nouns 9–16 together with the picture of the referent. Test sheet 2 presented items 1–8 together with the drawings and items 9–16 as written nouns only. The two test sheets were randomly issued to the students in both learning groups, and the subjects’ task was to indicate the appropriate grammatical gender of the nouns by supplying the nominative definite article (der, die, das) on the test sheets.

4.3.2. Results and discussion

Figures 7 a + b show the results for the items with congruent semantic and formal gender cues. It is obvious that the already well established formal gender regularities (75 % correct answers for masc items and 64 % correct answers for fem items without picture) are reinforced when the formal gender cue is furthermore supported by the visual depiction of the refer-ent’s natural gender (86 % correct answers for masc nouns resp. 80.5 % for fem items with picture).

If, however, semantic and formal gender indicators compete with each other, the results turn out quite differently. As shown in Figure 8a, subjects are quite strongly oriented towards formal gender cues (59.5 % of answers classified masc nouns as feminines), when known (sic!) items were pre-sented without visual depiction of the referent. And even when a picture clearly indicates the natural gender of the referent as male, 38.5 % of answers still assigned fem gender to those items.

A similar picture was obtained for the feminine items with conflicting gender cues (Figure 8b). The somewhat less dominant influence of the for-mal indicator of word final -er for masc gender when the pictures clearly identified the referents as female persons (30 %) can be attributed to the fact that -e fem obviously has been established more strongly by the subjects than -er masc, following cue input availability defined by the cue-category-mapping concept (McDonald 1989; cf. chap. 4.2).

Based on these results, the resolution of semantic vs. formal gender cue conflicts by beginning GFL learners can thus be recapitulated as follows:

86

3 11

75

13 12

masc fem neu

0 20 40 60 80 100

Assignment in %

with picture without picture

Figure 7a. Convergent gender cues; masc items – Study 2

8

80,5

10 22

64

14,5

masc fem neu

0 20 40 60 80 100

Assignment in %

with picture without picture

Figure 7b. Convergent gender cues; fem items – Study 2

In case of congruent semantic and formal gender indicators, both types of cues reinforce each other. If there is a conflict between both types of gender cues, the decisive criterion for the resolution of the conflict is the strength of the formal gender regularity. Cues with high availability in the learners’

input tend to dominate the semantic gender indication at this stage of GFL acquisition, whereas semantic cues dominate formal regularities with low input availability23. These observations correspond to Mills (1986: 113) findings from L1 acquisition of German: “In conclusion, an interaction between semantic and formal rules must be postulated which depends not on the categorization of the rules as semantic or formal, but rather on the relative ‘clarity’ of the rules in question within the gender system.”

30

59,5

10 63

19 19

masc fem neu

0 10 20 30 40 50

60 with picture

without picture

Assignment in %

Figure 8b. Conflicting gender cues; fem items – Study 2 48

38,5

12,5 22

59,5

18,5

masc fem neu

0 10 20 30 40 50

60 with picture

without picture

Assignment in %

Figure 8a. Conflicting gender cues; masc items – Study 2

4.4. Formal gender cue conflicts

While the resolution of semantic vs. formal gender cue conflicts can be in-vestigated directly by comparison, it is much more difficult to assess how students deal with inconsistent formal gender indicators. Following the ac-quisitional order of cue-category mappings in McDonald’s (1989) approach, cue conflict validity should be the last property to be learned, so that it could be assumed that beginning learners of German have not yet come to terms with this problem at all. A closer inspection of the latencies for the items of Study 1, however, reveals that a certain degree of inconsistence of some formal gender regularities is well reflected in the time subjects need to assign gender to the items in question. Figure 9 shows that the nouns with word final -en (regular items as well as exceptions) yielded the longest latencies (2620 ms resp. 2382 ms)24. Note that the accuracy rates for these nouns do not deviate significantly from the average results over all items.

In accordance with the assumption that the proportionally large number of exceptions to existing gender regularities for masculine nouns might be the reason for the longest latencies for this gender class (Table 4), the problem of cue conflict validity for individual formal gender indicators could have caused longer latencies for the assignment of gender to the nouns follow-ing this regularity. In the case of -en, the association with masculine gender is relatively reliable, as 72.1 % of nouns ending in -en are masculine according to the list of basic vocabulary (Wegener 1999: 513). In the sub-jects’ input, 10 types with 37 tokens followed this gender regularity. At the same time, however, there are not only some ‘real’ exceptions, such as das

All items -a

-at -chen

-e -el

-en -er

Ge--ik

-in -keit

-ment -o

-tion -um

-ung

-us Ex-e Ex-el

Ex-en Ex-er 2000

2100 2200 2300 2400 2500 2600 2700

Latencies in ms

Figure 9. Latencies for formal gender regularities – Study 1

Waschbecken [sink], but at least two conflicting gender regularities with word final -en: all nominalized verbs derived from the infinitive form end in -en and have neuter gender, e.g. das Skifahren [skiing], and the diminutive form -chen (das Mädchen [girl]) also ends in -en and categorically assigns neuter gender. Furthermore, the plural form of most feminie nouns ending in -e take the plural suffix -n, so that students will be confronted with final -en on a large number of nouns classified as feminines as well.

One can thus conclude from the latency results that learners are already working on the problem of gender cue conflicts even in the earlier stages of GFL acquisition, as is reflected in significantly longer reaction times for items with inconsistent formal gender indicators. Due to the complexity of the task, however, this property of cue-category-mappings will be mastered only in the later stages of the acquisition process.

Dalam dokumen Studies on Language Acquisition 25 (Halaman 74-86)