Computation of Word Associations Based on 
the Co-Occurences of Words in Large Corpora I 
Manfred Wettler & Reinhard Rapp 
University of Paderborn, Cognitive Psychology 
Postfach 1621, D-4790 Paderborn, Germany 
Abstract 
A statistical model is presented which predicts the strengths of word-associations from the 
relative frequencies of the common occurrences of words in large bodies of text. These 
predictions are compared with the Minnesota association norms for 100 stimulus words. The 
average agreement between the predicted and the observed responses is only slightly weaker 
than the agreement between the responses of an arbitrary subject and the responses of the 
other subjects. It is shown that the approach leads to equally good results for both English 
and German. 
1 Introduction 
In the association experiment first used by Galton (1880) subjects are asked to respond to 
a stimulus word with the first word that comes to their mind. These associative responses 
have been explained in psychology by the principle of learning by contiguity: "Objects once 
experienced together tend to become associated in the imagination, so that when any one of 
them is thought of, the others are likely to be thought of also, in the same order of sequence 
or coexistence as before. This statement we may name the law of mental association by 
contiguity." (William James, 1890, p. 561). 
When the association experiment is conducted with many subjects, tables are obtained 
which list the frequencies of particular responses to the stimulus words. These tables are called 
association norms. Many studies in psychology give evidence that there is a relation between 
the perception, learning and forgetting of verbal material and the associations between words. 
If we assume that word-associations determine language production, then it should be 
possible to estimate the strength of an associative relation between two words on the basis 
of the relative frequencies that these words co-occur in texts. Church et al. (1989), Wettler 
& Rapp (1989) and Church & Hanks (1990) describe algorithms which do this. However, the 
validity of these algorithms has not been tested by systematic comparisons with associations 
of human subjects. This paper describes such a comparison and shows that corpus-based 
computations of word associations are similar to association norms collected from human 
subjects. 
According to the law of association by contiguity, the association strength between two 
words should be a function of the relative frequency of the two words being perceived together, 
i.e. the relative frequency of the two words occuring together. Further more, the association 
strength between words should determine word selection during language or speech produc- 
tion: Only those words can be uttered or written down which associativeiy come to mind. 
If this assumption holds, then it should be possible to predict word associations from the 
common occurences of words in texts. 
IThis research was supported by the Heinz-Nixdorf-lnstitute (project 536) 
84 
2 Model 
According to the law of association by contiguity the learning of associations can be described 
as follows: If two words i and j occur together, the association strength aid(t) between i and 
j is increased by a constant fraction of the difference between the maximum and the actual 
association strength. This leads for association strengths between 0 and 1 to the following 
formula: 
aid (t + 1) = aid(t) + (1 - aid (t)). 01 if (i&j) (1) 
If word i occurs in another context, i. e. not in proximity to word j, the association strength 
between i and j is decreased by a constant fraction: 
aid (t + 1) = ai,j(t). (1 - 02) if (i&-~j) (2) 
Under the assumption that the learning rate 01 and the inhibition rate 02 are of identical 
size, the expected value aid of the association strength aid(t) from i to j for t ~ o~ is equal 
to the conditional probability of j given i (compare Foppa, 1965): 
ai,j = p(jli) (3) 
From these assumptions it could be expected that a stimulus word i leads to those response 
j, for which the value of equation 3 is at a maximum. 
Rapp & Wettler (1991) compared this with other predictions, where additional assump- 
tions on learning and reproduction were taken into account. With equation 3, mainly words 
with high corpus frequencies, e. g. function words, were predicted as associative responses. 
The predictions were improved when the following formula was used with an exponent of 
a = 0.66, and the word with the highest rid was considered to be the associative response. 
p(jli) (4) = p(j)'-"7 
The introduction of the denominator indicates that in the association experiment less frequent 
words are used than during language production. This inhibition of frequent words can be 
explained by the experimental situation, which furthers responses that are specific to the 
stimulus word. The exponential function can be interpreted as the tendency that subjective 
estimations are often found to be exponential functions of the quantities to be estimated. 
3 Association norms used 
For the comparison between the predicted and the associations of human subjects we have 
used the association norms coUected by Russet\] ~ Jenkins (Jenkins, 1970). They have the 
advantage that translations of the stimulus words were also given to German subjects (Russell 
& Meseck, 1959, and RusseLl, 1970) so that our model could be tested for English as well as 
for German. 
The Russell & Jenkins association norms, also referred to as the Minnesota word associ- 
ation norms, were collected in 1952. The I00 stimulus words from the Kent-Rosanoff word 
association test (Kent ~ Rosanoff, 1910) were presented to 1008 students of two large intro- 
ductory psychology classes at the University of Minnesota. The subjects were instructed, to 
write after each word "the first word that it makes you think of'. Seven years later, Russell 
Meseck (1959) repeated the same experiment in Germany with a carefully translated list of 
the stimulus words. The subjects were 331 students and pupils from the area near W~rzburg. 
The quantitative results reported on later will be based on comparisons with these norms. 
85 
The American as well as the German association norms were collected more than 30 years 
ago. The texts Which were used to simulate these associations are more recent. One might 
expect therefore that this discrepancy will impair the agreement between the observed and 
the predicted responses. Better predictions might be attained if the observed associations had 
been produced by the same subjects as the texts from which the predictions are computed. 
However, such a procedure is hardly realizable, and our results will show that despite these 
discrepancies associations to common words can be predicted successfully. 
4 Text corpora 
In order to get reliable estimates of the co-occurences of words, large text corpora have to be 
used. Since associations of the "average subject" are to be simulated, the texts should not 
be specific to a certain domain, but reflect the wide distribution of different types of texts 
and speech as perceived in every day life. 
The following selection of some 33 million words of machine readable English texts used 
in this study is a modest attempt to achieve this goal: 
• Brown corpus of present day American English (1 million words) 
• LOB corpus of present day British English (1 million words) 
• Belletristic literature from Project Gutenberg (1 million words) 
• Articles from the New Scientist from Oxford Text Archive (1 million words) 
• Wall Street Journal from the ACL/DCI (selection of 6 million words) 
• Hansard Corpus. Proceedings of the Canadian Parliament (selection of 5 million words 
from the ACL/DCI-corpus) 
• Grolier's Electronic Encyclopedia (8 million words) 
• Psychological Abstracts from PsycLIT (selection of 3.5 million words) 
• Agricultural abstracts from the Agricola database (3.5 million words) 
• DOE scientific abstracts from the ACL/DCI (selection of 3 million words) 
To compute associations for German the following corpora comprising about 21 million words 
were used: 
• LIMAS corpus of present-day written German (1.1 million words) 
• Freiburger Korpus from the Institute for German Language (IDS), Mannheim (0.5 
million words of spoken German) 
• Ma~nheimer Korpus 1 from the IDS (2.2 million words of present-day written German 
from books and periodicals) 
• Handbuchkorpora 85, 86 and 87 from the IDS (9.3 million words of newspaper texts) 
• German abstracts from the psychological database PSYNDEX (8 million words) 
For technical reasons, not all words occuring in the corpora have been used in the simulation. 
The vocabulary used consists of all words which appear more than ten times in the English 
or German corpus. It also includes all 100 stimulus words and all responses in the English 
or German association norms. This leads to an English vocabulary of about 72000 and a 
German vocabulary of 65000 words. Hereby, a word is defined as a string of alpha characters 
separated by non-alpha characters. Punctuation marks and special characters axe treated as 
words. 
86 
5 Computation of the association strengths 
The text corpora were read in word by word. Whenever one of the 100 stimulus words 
occured, it was determined which other words occured within a distance of twelve words to 
the left or to the right of the stimulus word, and for every pair a counter was updated. The 
so defined frequencies of co-occurence tt(i&j), the frequencies of the single words tt(i) and 
the total number of words in the corpus Q were stored in tables. Using these tables, the 
probabilities in formula (4) can be replaced by relative frequencies: 
tt(i&j) It(j) a Qo H(i&j) 
n(i) / Qo = H(j)o (5) 
In this formula the first term on the right side does not depend on j and therefore has no 
effect on the prediction of the associative response. With H(j) in the denominator of the 
second term, estimation errors have a strong impact on the association strengths for rare 
words. Therefore, by modifying formula (5), words with low corpus frequencies had to be 
weakened. 
. ~ H(i&j)/H(j) ~ fiir H(j) > ~.Q 
,'i,j = Q) f ir H(j) < Z" Q (6) 
According to our model the word j with the highest associative strength ~/,./to the stimulus 
word / should be the associative response. The best results were observed when parameter 
a was chosen to be 0.66. Parameters ~5 and 3' turned out to be relatively uncritical, and 
therefore to simplify parameter optimization were both set to the same value of 0.00002. 
Ongoing research shows that formula (6) has a number of weaknesses, for example that 
it does not discriminate words with co-occurence-frequency zero, as discussed by Gale & 
Church (1990) in a comparable context. However, since the results reported on later are 
acceptable, it probably gets the major issues right. One is, that subjects usually respond 
with common, i.e. frequent words in the free association task. The other is, that estimations 
of co-occurence-frequencies for low-frequency-words are too poor to be useful. 
6 Results 
In table 1 a few sample association lists as predicted by our system are compared to the as- 
sociative responses as given by the subjects in the Russell ~ Jenkins experiment. A complete 
list of the predicted and observed responses is given in table 2. It shows for all 100 stimulus 
words used in the association experiment conducted by Russell & Jenkins, a) their corpus 
frequency, b) the primary response, i.e. the most frequent response given by the subjects, 
c) the number of subjects who gave the primary response, d) the predicted response and 
e) the number of subjects who gave the predicted response. 
The valuation of the predictions has to take into account that association norms are 
conglomerates of the answers of different subjects which differ considerably from each other. 
A satisfactory prediction would be proven if the difference between the predicted and the 
observed responses were about equal to the difference between an average subject and the 
rest of the subjects. The following interpretations look for such correspondences. 
For 17 out of the 100 stimulus words the predicted response is equal to the observed 
primary response. Tiffs compares to an average of 37 primary responses given by a subject 
in the Russell & Jenkins experiment. A slightly better result is obtained for the correspon- 
dence between the predicted and the observed associations when it is considered, how many 
87 
Stim- Predicted ri,j Observed No. 
ulus Responses Responses Subj. 
l~lue green 2.144 sky 175 
red 1.128 red 160 
yellow 1.000 green 125 
white 0.732 color 66 
flowers 0.614 yellow 56 
sky 0.600 black 49 
colors 0.538 white 44 
eyes 0.471 water 36 
bright 0.457 grey 28 
color 0.413 boy 20 
butter bread 0.886 bread 637 
milk 0.256 yellow 81 
eggs 0.197 soft 30 
Ib 0.179 fat 24 
sugar 0.157 food 22 
fat 0.147 knife 20 
peanut 0.145 eggs 16 
fats 0.138 cream 14 
flavor 0.130 milk 13 
wheat 0.128 cheese 9 
baby mother 0.618 boy 162 
foods 0.427 child 142 
breast 0.353 cry 113 
feeding 0.336 mother 71 
infant 0.249 girl 51 
birth 0.245 small 43 
born 0.242 infant 27 
milk 0.208 cute 21 
her 0.206 little 18 
nursing 0.202 blue 17 
cold hot 1.173 hot 348 
warm 1.164 snow 218 
weather 0.736 warm 168 
winter 0.603 winter 66 
climate 0.474 ice 29 
air 0.424 Minnesota 13 
war 0.342 wet 13 
wet 0.333 dark 10 
water 0.330 sick 9 
dry 0.315 heat 8 
Table I: Comparison between the ten strongest pre- 
dicted and the ten most frequent observed responses 
for four stimulus words, rij was computed accord- 
ing to formula 6. 
subjects had given the predicted 
response: Averaged over all stim- 
ulus words and all subjects, a 
predicted response was given by 
12.6% of the subjects. By com- 
parison, an associative response of 
an arbitrary subject was given by 
21.9% of the remaining subjects. 
When only those 27 stimulus 
words are considered, whose pri- 
mary response was given by at 
least 500 subjects, an arbitrary 
response was given by 45.5% of 
the subjects on average. By com- 
parison, the predicted response to 
one of these 27 stimulus words 
was given by 32.6% of the sub- 
jects. This means, that for stim- 
ulus words where the variation 
among subjects is small, the pre- 
dictions improve. 
On the other hand, 35 of the 
predicted responses were given by 
no subject at all, whereas an av- 
erage subject gives only 5.9 out of 
100 responses that are given by no 
other subject. In about half of the 
cases we attribute this poor per- 
formance to the lack of represen- 
tativity of the corpus. For exam- 
ple, the predictions combustion to 
the stimulus bed or brokerage to 
house can be explained by specific 
verbal usage in the DOE scientific 
abstracts respectively in the Wall 
Street Journal. 
In most other cases instead of 
paradigmatic associations (words 
that are used in similar contexts) 
syntagmatic associations (words 
that are often used together) are 
predicted. Examples are the pre- 
diction of term to the stimulus 
long, where most subjects an- 
swered with short, or the predic- 
tion of folk to music, where most 
subjects responded with song. 
88 
stim 
afraid 
anger 
baby 
bath 
beautiful 
bed 
Bible 
bitter 
black 
blossom 
blue 
boy 
bread 
butter 
butterfly 
cabbage 
carpet 
chair 
cheese 
child 
citizen 
city 
cold 
comfort 
command 
cottage 
dark 
deep 
doctor 
dream 
eagle 
earth 
eating 
foot 
fruit girl 
green 
hammer 
hand 
hard 
head 
health 
heavy 
high 
house 
hungry joy 
justice 
king 
'lamp 
freq 
692 
615 
1157 
244 
812 
1295 
593 
541 
4250 
50 
1676 
1174 
863 
426 
68 
116 
138 
577 
566 
8897 
525 
8125 
2003 
386 
799 
137 
1695 
2418 
766 
629 
92 
1429 
2823 
1169 
1841 
1096 
1686 
173 
5146 
3502 
5350 
11433 
3497 
25220 
3059 
268 
246 
1314 
1983 
330 
par f (par) 
fear 
mad 
boy 
clean 
ugly 
sleep 
God 
sweet 
white 
flower 
sky 
girl 
butter 
bread 
moth 
head 
rug 
table 
crackers 
baby(ies) U.S.(A.) 
town 
hot 
chair 
order 
house 
light 
shallow 
nurse 
sleep 
bird 
round 
food 
shoe(s) 
apple 
boy 
grass 
nail(s) 
foot(ee) 
soft 
hair 
sickness 
'light 
low 
home 
food 
happy 
peace 
queen 
light 
261 
351 
162 
314 
209 
584 
236 
652 
751 
672 
175 
768 
610 
637 
144 
165 
460 
493 
108 
159 
114 
353 
348 
117 
196 
298 
829 
318 
238 
453 
550 
130 
390 
232 
378 
704 
262 
537 
255 
674 
129 
250 
583 
675 
247 
362 
209 
250 
751 
633 
pred 
am 
expression 
mother 
hot 
love 
combustion 
Society 
sweet 
white 
flower 
green 
girl 
wheat 
bread 
fish 
potatoes 
red 
clock 
milk 
care 
senior 
pop 
hot 
ease 
army 
cheese 
brown 
sea 
patient 
sleep 
bird 
rare 
habits 
square 
vegetable 
boy 
blue 
string 
On 
hit 
tail 
mental 
ion 
low 
brokerage 
eat 
fear 
criminal 
emperor 
light 
f (pred) , 
0 
0 
71 
10 
0 
0 
0 
652 
751 
672 
125 
768 
4 
637 
0 
0 
27 
0 
47 
10 
0 
0 
348 
76 
102 
111 
1 
77 
11 
453 
550 
0 
0 
0 
114 
704 
122 
0 
0 
1 
17 
0 
0 
675 
0 
174 
5 
1 
1 
633 
Table 2, part 1. Observed and predicted associative responses to stimulus words 1 to 50. 
The abbreviations in the headline mean: stim = stimulus word; freq = corpus frequency of 
stimulus word; par = primary associative response; f (pax) = number of subjects who gave 
the primary associative response; pred -- predicted associative response; f (pred) = number 
of subjects who gave the predicted assocdative response. 
89 
stim 
light 
lion 
long 
loud 
man 
memory 
moon 
mountain 
music 
mutton 
needle 
ocean 
priest 
quiet 
red 
religion 
river 
rough 
salt 
scissors 
sheep 
short 
sickness 
sleep 
slow 
smooth 
soft 
soldier 
sour 
spider 
square 
stem 
stomach 
stove 
street 
sweet 
swift 
table 
thief 
thirsty 
tobacco 
trouble 
whiskey 
whistle 
white 
window 
wish 
woman 
working 
yellow 
freq 
7538 
182 
16437 
230 
7472 
3230 
295 
1066 
3635 
39 
208 
1066 
311 
673 
3029 
1224 
1624 
457 
2158 
25 
854 
7388 
207 
1843 
1858 
690 
1681 
321 
154 
97 
1430 
796 
501 
98 
859 
7OO 
184 
2396 
63 
32 
1056 
1108 
63 
77 
4807 
816 
2061 
2995 
5366 
1188 
par 
dark 
tiger 
short 
soft woman(e) 
mind 
stars 
hill(s) song(s) 
lamb 
thread 
water 
church 
loud 
white 
church 
water 
smooth 
pepper 
cut 
wool 
tall 
health 
bed 
fast 
rough 
hard 
army 
sweet 
web 
round 
flower 
food 
hot 
avenue 
sour 
fast 
chair 
steal 
water 
smoke 
bad 
drink(s) 
stop 
black 
door 
want man(e) 
hard 
blue 
f (par) 
647 
261 
758 
541 
767 
119 
205 
266 
183 
365 
464 
314 
328 
348 
221 
285 
246 
439 
430 
671 
201 
397 
376 
238 
752 
328 
445 
187 
568 
454 
372 
402 
211 
235 
190 
434 
369 
84O 
286 
348 
515 
89 
284 
131 
617 
191 
124 
646 
132 
156 
pred 
dark 
sea 
term 
noise 
woman 
deficits 
sun 
ranges 
folk 
beef 
sharing 
flOOr 
Catholic 
sleep 
yellow 
Christianity 
flows 
smooth 
sugar 
pa~r 
cattle 
term 
motion 
hrs 
wave 
muscle 
drink 
army 
sweet 
tail 
root 
brain 
cancer 
kitchen 
corner 
potatoes 
rivers 
honour 
catch 
drink 
textiles 
rein 
beer 
train 
black 
glass 
I 
yr 
class 
green 
f(pred) 
647 
2 
0 
210 
767 
0 
168 
0 
0 
32 
0 
6 
189 
53 
19 
5 
0 
439 
83 
1 
15 
0 
0 
0 
0 
1 
10 
187 
568 
0 
22 
2 
1 
16 
20 
0 
0 
0 
2 
296 
0 
0 
52 
89 
617 
171 
2 
0 
3 
89 
MEAN: 2064.78 377.52 I 127.34 
Table 2, part 2. Observed and predicted associative responses to stimulus words 51 to 100. 
90 
Using the corpora listed in section 4, the same simulation as described above was conducted 
for German. For the computation of the associative strengths, again formula 6 was used. For 
optimal results, only a small adjustment had to be made to parameter alpha (from 0.66 to 
0.68). However, a significant change was necessary for parameters/~ and 7, which again for 
ease of parameter optimization were assumed to be identical. ~ and 7 had to be reduced by 
a factor of approximately four from a value of 0.00002 to a value of 0.000005. Apart from 
these parameters, nothing was changed in the algorithm. 
Table 3 compares the quantitative results as given above for both languages. The 
figures can be interpreted as follows: With an average of 21.9% of the other subjects giving 
the same response as an arbitrary subject, the variation among subjects is much smaller in 
English than it is in German (8.7%). This is reflected in the simulation results, where both 
figures (12.6% and 6.9%) have a similar ratio, however at a lower level. 
This observation is confirmed when only stimuli with low variation of the associative re- 
sponses are considered. In both languages, the decrease in variation is in about the same 
order of magnitude for experiment and simulation. Overall, the simulation results are some- 
what better for German than they are for English. This may be surprising, since with a 
total of 33 million words the English corpus is larger than the German with 21 million words. 
However, if one has a closer look at the texts, it becomes clear, that the German corpus, 
by incorporating popular newspapers and spoken language, is clearly more representative to 
everyday language. 
Description 
percentage of subjects who give the predicted associative 
response 
percentage of other subjects who give the response of an 
arbitrary subject 
percentage of subjects who give the predicted associative 
response for stimuli with little response variation" 
percentage of other subjects who give the response of an 
arbitrary subject for stimuli with little response variation* 
percentage of cases where the predicted response is 
identical to the observed primary response 
percentage of cases where the response of an arbitrary 
subject is identical to the observed primary response 
percentage of cases where the predicted response is given 
by no subject °* 
percentage of cases where the response of an arbitrary 
subject is given by no other subject** 
English German 
12.6% 6'.9% 
21.9% 8.7% 
32.6% 15.6% 
i 
45.5% 18a% 
17.0% ' 19.0% 
37.5~ 22.5% 
35.0% 57.0% 
'5.9% 19.8% 
Table 3: Comparison of results between simulation and experiment for English and German. 
Notes: ") little response variation is defined slightly different for English and German: in the English 
study, only thoee 27 stimulus words are considered, whose primary response is given by at least 500 
out of 1008 subjects. In the German study, only those 25 stimulus words are taken into account, 
wh~e primary response is given by at least 100 out of 331 subjects. **) for comparison of English and 
German experimental figures, it should be kept in mind, that the American experiment was conducted 
with 1008, but the German experiment with only 331 subjects. 
93. 
7 Discussion and conclusion 
In the simulation results a bias towards syntagmatic associations was found. Since the as- 
sociations were computed from co-occurences of words in texts, this preference of syntag- 
matic associations is not surprising. It is remarkable, instead, that many associations usually 
considered to be paradigmatic are predicted correctly. Examples include man -- woman, 
black ~ white and bitter ~ sweet. We believe, however, that the tendency to prefer syntag- 
matic associations can be reduced by not counting co-occurences found within collocations. 
Equivalently, the association strength between word pairs always occuring together in a strict 
formation (separated by a constant number of other words) could be reduced. 
When going from English to German, the parameters /~ and '7 in equation 6 needed 
to be readjusted in such a way, that less frequent words obtained a better chance to be 
associated. This reflects the fact, that there is more variation in the associative responses 
of German than of American subjects, and that American subjects tend to respond with 
words of higher corpus frequency. We believe that by considering additional languages this 
parameter adjustment could be predicted from word-frequency-distribution. 
In conclusion, the results show, that free word associations for English and German can 
be successfully predicted by an almost identical algorithm which is based on the co-occurence- 
frequencies of words in texts. Some peculiarities in the associative behavior of the subjects 
were confirmed in the simulation. Together, this is a good indication that the learning of 
word associations is governed by the law of association by contiguity. 
Although our simulation results are not perfect, specialized versions of our program have 
already proved useful in a number of applications: 
• Information Retrieval: Generation of search terms for document retrieval in biblio- 
graphic databases (Wettler & Rapp, 1989, Ferber, Wettler & Rapp, 1993). 
• Marketing: Association norms are useful to predict what effects word usage in adver- 
tisements has on people (Wettler & Rapp, 1993). Muitilingual assocation norms help 
to find a global marketing strategy in international markets (Kroeber-Riel, 1992). 
• Machine Translation: In an experimental prototype we have shown that associations de- 
rived from context are useful to find the correct translations for semantically ambiguous 
words. 
The successful prediction of different types of verbal behavior on the basis of co-occurrences 
of words in texts is a direct application of the classical contiguity-theory, or, in more modern 
neurophysiological terms, of Hebb's learning rule. Cognitive psychology has severely criti- 
cized contiguity-theory with the arguments that association theory did not produce useful 
results (Jenkins, 1974), and that associations are not the result of associative learning but of 
underlying semantic processes (Clark, 1970). Both arguments need a critical revision. Re- 
cent work with large corpora as well as a large number of connectionist studies have yielded 
very useful results in different psychological domains, and the high predictive power of the 
associationist approach makes that the intuitive appeal of cognitivist explanations is fading 
rapidly. 

References 
Clark, H.H. (1970). Word associations and linguistic theory. 
horizons in linguistics. Harmondsworth: Penguin, 271-286. 
In: J. Lyons (ed.), New 
Church, K.W., Gale, W., Hanks, P. & Hindle, D. (1989). Parsing, word associations and 
typical predicate-argument relations. In: Proceedings of the International Workshop on 
Parsing Technologies, Carnegie Mellon University, PA, 389-398. 
Church, K.W., Hanks, P. (1990). Word association norms, mutual information, and lexi- 
cography. Computational Linguistics, Volume 16, Number 1, March 1990, 22-29. 
Ferber, R., Wettler, M., Rapp, R. (1993). An associative model of word selection in the 
generation of search queries. In preparation. 
Foppa, K. (1965). Lernen, Ged~chtnis, Verhalten. KSln: Kiepenheuer & Witsch. 
Gale, W. A., Church, K. W. (1990). Poor estimates of context are worse than none. DARPA 
Speech and Natural Language Workshop, Hidden Valley, PA, June 1990, 283-287. 
Galton, F. (1880). Psychometric ezperiments. Brain 2, 149-162. 
James, W. (1890). The principles of psychology. New York: Dover Publications. 
Jenkins, J.J. (1970). The 1952 Minnesota word association norms. In: Postman, L., Keppel, 
G. (eds.): Norms of word association. New York: Academic Press, 1-38. 
Jenkins, J.J. (1974). Remember that old theory of learning? Well, forget it! American 
Psychologist, 29, 785-795. 
Kent, G.H. & Rosanoff, A.J. (1910). A study of association in insanity. American Journal 
of Insanity, 67 (1910), 37-96, 317-390. 
Kroeber-Riel, W. (1992). Globalisie~ung der Euro-Werbung. Marketing ZFP, Heft 4, IV 
Quartal, 261-267. 
Rapp, R. & Wettler, M. (1991). Prediction of free word associations based on Hebbian learn- 
ing. Proceedings of the International Joint Conference on Neural Networks, Singapore, 
Vol.1, 25-29. 
Russell, W.A. (197'0). The complete German language norms for responses to 100 words 
from the Kent-Rosanoff word association test. In: Lo Postman & G. Keppel (eds.), 
Norms of word association. New York: Academic Press, 53-94. 
Russell, W.A. & Meseck, O.R. (1959). Der Elnttui3 der Assoziation auf das Erinnern yon 
Worten in der deutschen, franzSsischen und englischen Sprache. Zeitschrifl fir ezperi- 
mentelle und angewandte Psychologie, 6, 191-211. 
Wettler, M. & Rapp, R. (1989). A connectionist system to simulate lexical decisions in 
information retrieval. In: Pfeifer, R., Schreter, Z., Fogelman, F. Steels, L. (eds.), 
Connectionism in perspective. Amsterdam: Elsevier, 463-469. 
Wettler, M. & Rapp, R. (1993). Associative analysis of advertisements. Submitted to 
Marketing and Research Today. 
