Borrowability and the notion of

basic vocabulary

Uri Tadmor, Martin Haspelmath and Bradley Taylor

Max Planck Institute for Evolutionary Anthropology

This paper reports on a collaborative quantitative study of loanwords in 41

languages, aimed at identifying meanings and groups of meanings that are
borrowing-resistant. We find that nouns are more borrowable than adjectives
or verbs, that content words are more borrowable than function words, and that
different semantic fields also show different proportions of loanwords. Several
issues arise when one tries to establish a list of the most borrowing-resistant
meanings: Our data include degrees of likelihood of borrowing, not all meanings
have counterparts in all languages, many words are compounds or derivatives
and hence almost by definition non-loanwords. We also have data on the age of
words. There are thus multiple factors that play a role, and we propose a way of
combining the factors to yield a new 100-item list of basic vocabulary, called the
Leipzig-Jakarta list.

Keywords: loanword, borrowing, basic vocabulary, stability, language contact

1. Assessing degrees of lexical borrowability

Predicting borrowing behavior is important in historical and comparative linguistics

for a variety of reasons. Most importantly, borrowing is often a confounding factor in
assessing genealogical relatedness of languages. If we were able to determine in gen-
eral which words are more or less likely to be borrowed, based on their meanings, we
would be in a better position to distinguish diffusional similarities from similarities
that are due to common ancestry.
Until now, not much has been known about borrowability in general. For some
grammatical domains, tentative borrowability scales have been set up (Matras 1998,
2007, Field 2002), and recent years have seen a lot of new research on grammatical
borrowing (Aikhenvald & Dixon 2007, Matras & Sakel 2007). However, no systematic
cross-linguistic research on lexical borrowing has been carried out before our project
(the Loanword Typology project, see §3). A tradition of research examines the stability

Diachronica 27:2 (2010), 226–246.  doi 10.1075/dia.27.2.04tad

issn – / e-issn – © John Benjamins Publishing Company
 Borrowability and the notion of basic vocabulary 227

of lexical meanings (e.g., Dolgopolsky 1986, Lohr 1998, Holman et al. 2008), but it
does not specifically address borrowing; in fact, Holman et al. emphasize that stability
in general and resistance to borrowing pattern differently.
Linguists have long been in agreement that “basic vocabulary” or “core vocabu-
lary” is more resistant to borrowing than less basic vocabulary, but what exactly is
meant by this is often left unclear. However, we can approach this question empirically.
If the kinds of words that are borrowed or are resistant to borrowing across languages
are not a random selection, but systematically tend to come from certain meaning
domains but not others, then we can come up with a list of hard-to-borrow vocabulary
by examining a representative set of languages with known loanwords. This is what we
have done in the Loanword Typology (LWT) project. One of our results is a list of basic
vocabulary based on data from 41 languages from all continents, the Leipzig-Jakarta
list of basic vocabulary (see §9).1 However, as will be explained later, it turned out
that low borrowability by itself was not a sufficient criterion and needed to be supple-
mented by a few other criteria.

2. The notion of basic vocabulary and the Swadesh 100 list

The term “basic vocabulary” can mean different things in different domains of linguis-
tics. For example, in second language acquisition “basic vocabulary” is that part of the
lexicon that “would have as wide a communicative range as possible using a minimum
number of words of general meaning” (McCarthy 1999: 233). In corpus linguistics, it
may be equated with the most frequent words. In historical and comparative linguis-
tics, basic vocabulary has typically been associated with stability, universality, simplic-
ity, and resistance to borrowing.
As Hymes (2006 [1971]: 254) pointed out, “[t]he notion of basic vocabulary is at
least as old as comparative linguistics”. Swadesh’s great contribution was in formulat-
ing a standard list of basic vocabulary, utilizing it for various lexicostatistical studies,
and perhaps most importantly — inspiring scholars around the world to use it for
thousands of languages.
In his writings, Swadesh was quite explicit in describing how he created and re-
fined his 100-item list. He describes the purpose of the list as follows (Swadesh 2006
[1971]: 19):
In counting and statistics, it is convenient to operate with representative samples,
that is, a portion of the entire mass of facts so selected as to reflect the essential
facts. For our lexical measure of linguistic divergence we need some kind of se-
lected word list, a list of words for which equivalents are found in each language
or language variant …

1.  The list is named after the locations where it was produced.
228 Uri Tadmor, Martin Haspelmath and Bradley Taylor

Although the major purpose of the list was to help determine relationships among
languages already known to be related, Swadesh also pointed out that “[t]he technique
of the diagnostic vocabulary, which was developed as an instrument of glottochrono-
logical measurement, has come to have other uses, above all some related to problems
of remote relationship” (Swadesh 2006 [1971]: 279). In other words, the list was seen
as a means of establishing previously unknown genealogical connections, not only
determining the degree of relatedness within known groups.
Swadesh defined basic vocabulary as “concepts and experiences common to all
human groups” (Swadesh 1950: 157) and “the fundamental everyday vocabulary of any
language — as against the specialized or ‘cultural’ vocabulary” (Swadesh 1952: 452).
Later he spelled out more explicitly the elements that constituted his basic vocabulary
(Swadesh 2006 [1971]: 275): “… universal and simple things, qualities, and activities,
which depend to the least degree possible on the particular environment and cultural
state of the group”, including “pronouns, some quantitative concepts, parts and simple
activities of the body, movements, and some general qualities of size, color, and so on”.
Explicitly excluded were “words of a cultural nature, words that in many languages are
sound-imitative (onomatopoeic) … [and] terms with very specific meanings”.
How did Swadesh create his 100-item list? Based on his intuitions, around 1948
he composed a list of meanings that suited his criteria, which he first tested on English
(Swadesh 1950: 161). As he later explained, “[t]he first research making use of the di-
agnostic list led to changes, the elimination of some elements and the substitution of
others, and finally the selection of the hundred words” (Swadesh 2006 [1971]: 275).2
Quantitative studies were apparently not part of his methodology.

3. The Loanword Typology project

In the Loanword Typology (LWT) project, which we coordinated between 2004 and
2009, each language was the responsibility of an author who is a specialist of the lan-
guage and its history. Each of the authors (or author teams) provided counterparts

2.  The number of items on the list fluctuated. The original list tested on English contained 225
items. The list in Swadesh 1950 contained 165 items, many of which, however, were culture-
specific (e.g., ‘canoe’, ‘moccasin’). In Swadesh 1950, he only used 121 of these, plus 44 extra
forms relevant to Salishan languages which were not used on the general list because they were
culture-specific (e.g. ‘canoe’, ‘moccasin’). The final list of 100 items can be found in Swadesh 2006
[1971], although it was formulated years before the publication of the first edition of that book.
Many linguists are not aware that the popularly used 200-item list was one of the intermediate
lists, not what Swadesh considered to be his final product. The 207-item list, which is also com-
monly used, was in fact never used by Swadesh himself. Rather, it contains all the meanings on
the 200-item list plus seven items that appear on the final 100-item list but not on the 200-item
list. Below, we will consider only the final 100-item list.
 Borrowability and the notion of basic vocabulary 229

(=translational equivalents) for lexical meanings on a fixed list of 1460 meanings (the
LWT meaning list). When the counterpart word was judged to be a loanword, supple-
mentary information was provided on what is known about the historical circum-
stances under which it was borrowed. The resulting combined database (which we
call the World Loanword Database, Haspelmath & Tadmor 2009b) is a lexical database
comprising 41 individual language subdatabases (available online at http://wold.liv-
ingsources.org).3 We hope that these languages are reasonably representative of the
world’s languages; while they do not come close to the ideal of a fully random sample,
we do not think that the biases of the sample lead to major skewings in the results. (In
contrast to grammatical typology, where genealogical and geographical bias is a well-
known confounding factor, language contact typology seems to be much less affected
by such factors.)
Each subdatabase contains about 1,000–2,000 words or counterparts of the mean-
ings on the Loanword Typology meaning list. The number of words in each sub-
database varies, because sometimes a language had no counterpart for a particular
meaning, while in other cases it had several counterparts. Considerable work by the
language experts went into these subdatabases, and each of them constitutes a separate
online publication. In addition, each author team wrote a prose chapter describing the
borrowing situation in their language. These chapters and a general description of the
project and its results are found in Haspelmath & Tadmor (2009a).
The LWT meaning list consists of 1,460 lexical meanings, most of which were
adopted from the meaning list of the Intercontinental Dictionary Series, which in turn
was based on Buck (1949). We opted for a meaning list that is significantly longer than
Swadesh’s list, because we wanted to get a more comprehensive picture of the lexicon

3.  The languages are:

Africa: Swahili, Iraqw, Gawwada, Hausa, Kanuri, Tarifiyt Berber, Seychelles Creole

Europe: English, Old High German, Lower Sorbian, Dutch, Romanian, Selice Romani, Kildin
Saami, Bezhta, Archi

Asia: Manange, Sakha, Mandarin Chinese, Thai, Vietnamese, White Hmong, Japanese, Ket,
Oroqen, Ceq Wong, Indonesian

Pacific: Malagasy, Takia, Hawaiian, Gurindji

Americas: Yaqui, Zinacantán Tzotzil, Q’eqchi’, Otomi, Saramaccan, Imbabura Quechua, Kari’na,
Hup, Wichí, Mapudungun

In selecting languages for inclusion in the project, an effort was made to represent the world’s
genealogical, geographical, typological, and sociolinguistic diversity. However, the overriding
factors were practical. Languages could only be included if a specialist in the language volun-
teered to invest the considerable amount of time and effort needed to complete the database
and to write a book chapter based on the findings. This has no doubt led to a bias in favour of
languages with many loanwords, and in favour of well-studied languages.
230 Uri Tadmor, Martin Haspelmath and Bradley Taylor

of each project language. Unlike Swadesh, who determined a priori what constituted
basic vocabulary based on his intuitions, and then proceeded to refine his list by trial
and error, we wanted the composition of any list derived from the LWT project to be
empirically based. We also added about 160 meanings — many of them having to do
with the modern world (‘hospital’, ‘newspaper’, ‘radio’) — because we were interested
in the extent to which loanwords or native neologisms are used for such new concepts.
Many of the meanings on the list were not culture-free, so often a language did not
have a counterpart of a given meaning. The geographical distribution of the languages
is shown in Map 1.

Map 1.  The geographical distribution of the 41 languages in the Loanword Typology

project

For each counterpart word, the database gives the (orthographic and/or transcribed)
form of the word, information about the analyzability of this word, a morpheme-by-
morpheme gloss (for analyzable words), information about loanword status, informa-
tion about the age of the word, and optional further information of various kinds.
For each loanword, the database gives the donor language and the source word (with
its meaning), as well as some information about the borrowing circumstances. The
loanword status is not a simple binary distinction, but a point on a scale between “No
evidence for borrowing” and “Clearly borrowed”. For calculating the loanword rates of
the project languages (as well as for statistics relating to semantic word classes and se-
mantic field as discussed in §4), we regarded as loanwords all words that were marked
as “clearly borrowed” or as “probably borrowed”.
One result of our project is a ranking of the languages with respect to the pro-
portion of (clear or probable) loanwords in their vocabulary. The leading borrower
is Selice Romani, with 63% loanwords, followed by Tarifiyt Berber (52%), Gurindji
(46%), Romanian (42%), and English (41%). The language with the lowest percent-
 Borrowability and the notion of basic vocabulary 231

age of loanwords is Mandarin Chinese (1%), followed by Old High German (6%) and
Manange (8%). The average in our sample is 24% loanwords, but it should be noted
that our sample is biased toward languages with many loanwords, because linguists
working on such languages were more interested in contributing to our project. More
details on the differences in the borrowing patterns of the various project languages
can be found in Tadmor (2009) and in the online World Loanword Database.

4. Differences among semantic word classes and semantic fields

4.1 Nouns vs. verbs (and adjectives)

It has long been known that languages are more likely to borrow nouns than verbs.
This is not only due to the fact that languages have more nouns than verbs. In the
consolidated database from our 41 languages, the verb-to-noun ratio is 1:2.5, but the
corresponding ratio among the loanwords is 1:5.5 (Table 1). While almost a third of all
nouns are loanwords, less than a sixth of the verbs are loanwords. Possible reasons for
this are discussed in great detail in Wohlgemuth (2009). Interestingly, adjectives (and
adverbs) are almost as hard to borrow as verbs — this is a much less well-known fact
which has hardly received any attention so far.

Table 1.  Semantic nouns, verbs, and adjectives

Semantic word class All words Loanwords Loanwords as % of total
Nouns 34,355 10,712 31.2%
Verbs 13,808   1,932 14.0%
Adjectives and adverbs   5,284    803 15.2%
All content words 53,446 13,446 25.2%

4.2 Content words vs. function words

Words with grammatical meanings (“function words”) are even harder to borrow than
verbs. As Table 2 shows, only about 12% of all function words are borrowed.

Table 2.  Content words vs. function words

Category All words Loanwords Loanwords as % of total
Content words 53,446 13,446 25.2%
Function words   4,071    492 12.1%
All words 57,517 13,938 24.2%
232 Uri Tadmor, Martin Haspelmath and Bradley Taylor

4.3 Differences among semantic fields

As already seen, not all word meanings are equally often borrowed. Cultural items,
such as words relating to religion, clothing, the house, and law, tend to be borrowed
often, as shown in Table 3, which ranks semantic fields by percentage of loanwords in
the combined database. The semantic fields that are used in this table are the fields of
Buck (1949), which are also retained in the Intercontinental Dictionary Series. At the
bottom of Table 3, we find semantic fields with relatively culture-free meanings, such
as words relating to sense perception, spatial relations, and body parts. Thus, our find-
ings broadly confirm the old view that words with culture-free meanings are less likely
to be borrowed.4 However, since we have information about each individual meaning,
we can be much more specific.

Table 3.  Semantic fields, ranked by loanword percentage

Semantic field No. of meanings Loanwords as % of total
Religion and belief 26 41.2%
Clothing and grooming 59 38.6%
The house 47 37.2%
Law 26 34.3%
Social and political relations 36 31.0%
Agriculture and vegetation 75 30.0%
Food and drink 81 29.3%
Warfare and hunting 40 27.9%
Possession 46 27.1%
Animals 116 25.5%
Cognition 51 24.2%
Basic actions and technology 78 23.8%
Time 57 23.2%
Speech and language 41 22.3%
Quantity 39 20.5%
Emotions and values 48 19.9%
The physical world 75 19.8%
Motion 82 17.3%

4.  A reviewer makes the interesting valid point that different semantic fields in Table 3 have
different proportions of nouns vs. everything else, and that this could account for some of the
differences. One of the reasons why the fields of Sense perception and Spatial relations show few
loanwords is that they contain a much greater proportion of adjectives and verbs than Clothing
and grooming or The house.
 Borrowability and the notion of basic vocabulary 233

Table 3.  (continued)
Semantic field No. of meanings Loanwords as % of total
Kinship 85 15.0%
The body 159 14.2%
Spatial relations 75 14.0%
Sense perception 49 11.0%
All words 24.2%

5. The most borrowing-resistant meanings

5.1 Meanings with the fewest (probable or clear) loanword counterparts

One of the main goals of the LWT project was to determine which word meanings are
least likely to be borrowed. This task, however, turned out to be far from straightfor-
ward. The first problem was that our database provided two different ways of deter-
mining the least borrowed meanings. The first was to count the percentage of “clearly
borrowed” and “probably borrowed” words among all counterparts for each meaning.
A ranking obtained using this method is provided in Table 4. The second method was
to assign gradually decreasing numerical values to each of the five loanword statuses5
and then to calculate the average borrowability scores for each meaning. This was the
basis of the ranking in Table 5.6
As can readily be seen, the rankings obtained using the two methods are rather
different from each other. In Table 4, we see that only 17 meanings on the LWT list
have no counterparts that are (clearly or probably) loanwords, among them 13 deictic
meanings, three verbal meanings, and one adjectival meaning. There is not a single
nominal meaning among them. That meanings of this kind should be at the top of the
list of low-borrowability items is not surprising, in view of the fact that function words
and verbs are generally much more borrowing-resistant than nouns, as we saw earlier.
This also means that a list of the 100 most borrowing-resistant meanings would be
very different from the Swadesh list and other high-stability lists, which contain few
deictic and grammatical meanings. In Table 5, we see that there are only five mean-
ings (he/she/it, we [inclusive], we [exclusive], up, this) that only have counterparts with
no evidence for borrowing, all of them deictics. Since the ranking in Table 5 utilizes
a graded score rather than an arbitrary cut-off point as the ranking in Table 4, it was
used as the basis for the final list.

5.  The assigned values were: “no evidence for borrowing”, 1.00; “very little evidence for borrow-
ing”, .75; “perhaps borrowed”, .50; “probably borrowed’, .25; and “clearly borrowed”, 0.

6.  Because of space limitations, only the first 25 items are listed in each table.
234 Uri Tadmor, Martin Haspelmath and Bradley Taylor

Table 4.  LWT meanings ranked by percentage of clearly and probably borrowed counter-
parts (top 25)
LWT label % of clearly and probably
borrowed counterparts
1 he/she/it 0.00%
1 we 0.00%
1 we (inclusive) 0.00%
1 we (exclusive) 0.00%
1 itch 0.00%
1 spin 0.00%
1 rise 0.00%
1 up 0.00%
1 day after tomorrow 0.00%
1 bitter 0.00%
1 how? 0.00%
1 where? 0.00%
1 which? 0.00%
1 why? 0.00%
1 this 0.00%
1 it 0.00%
17 say 1.11%
17 younger sister 1.11%
19 run 1.21%
20 married woman 1.34%
21 raw 1.60%
22 throw 1.79%
23 thatch 1.83%
24 there 1.85%
25 that 1.88%

6. Representation

The second problem to be tackled was that the meanings in our list differed markedly
in representation, i.e. the number of languages for which the database has counter-
parts. While all languages have counterparts for the deictics where, why, which, there,
here, how, and for basic verbs like rise, lie down, stand, some highly specific meanings
like mother-in-law of a man, netbag, tumpline, and larch are only represented in a few
of the project languages. If a particular meaning has counterparts in all 41 project
 Borrowability and the notion of basic vocabulary 235

Table 5.  LWT meanings ranked by borrowability score (top 25)

LWT label Borrowed score
1 he/she/it 1.00
1 we (inclusive) 1.00
1 we (exclusive) 1.00
1 up 1.00
1 this 1.00
6 where? 0.997
7 why? 0.995
8 which? 0.994
9 we 0.991
10 married woman 0.990
11 younger sister 0.989
11 rise 0.989
13 day after tomorrow 0.987
13 spin 0.987
15 stinking 0.982
15 bring 0.982
17 day before yesterday 0.981
17 there 0.981
17 lie down 0.981
17 stand 0.981
17 here 0.981
22 how? 0.980
23 run 0.976
24 behind 0.975
24 bitter 0.975

languages, none of which are loanwords, this is excellent evidence that this meaning
has very low borrowability. But if a meaning has a counterpart in only one project
language, and that word is not a loanword, that hardly constitutes any evidence. It
would be improper to make any generalizations based on evidence from just one lan-
guage (or from just a few languages). Yet if we ranked meanings purely by percentage
of loanword counterparts, the meaning with unborrowed counterparts in all project
languages and the meaning with just one counterpart that happens to be unborrowed
would receive the same score and would have an identical ranking. Obviously, such an
approach would be deeply flawed.
236 Uri Tadmor, Martin Haspelmath and Bradley Taylor

In order to address this problem, we computed the representation rate of each

meaning as the percentage of the project languages that had a counterpart for it. This
score eventually constituted one of the four scores which together made up our com-
posite score (see §9).

7. Analyzability

Some of the low-borrowability meanings in Tables 4 and 5, such as ‘younger sister’ and
‘day after tomorrow’, have many counterpart words that are analyzable. This means
that these words were most probably created in the language, and not borrowed from
another language. For such words the non-borrowed status is not surprising. Words
with such meanings are rarely borrowed, but not because of some inherent resistance
to borrowing — they simply tend not to be loanwords because they are often made
up from the resources of the language. In order to take into account the effect of this
factor, we computed an average analyzability (or simplicity) score for each meaning.7
Analyzable words included complex and compound words as well as phrasal ex-
pressions, and semi-analyzable words were those whose complexity is transparent only
to linguists, or words containing so-called ‘cranberry morphs’. Some of the meanings
in Table 4 and Table 5 such as ‘to stand’, ‘bitter’, and ‘we’ show a strong tendency to
have unanalyzable counterparts (analyzability score of over .85), while meanings such
as ‘day after tomorrow’ and ‘married woman’ are much more often analyzable (analyz-
ability score of under .65).

8. Age

The age of words has implications for their usefulness as evidence for borrowability.
The longer a word has existed in a language without being replaced by a loanword,
the better evidence it constitutes for the low borrowability of its meaning. Therefore
information about word ages was collected in a systematic way. For each word, the
contributors were asked to give an age, i.e. an approximate year or time period when
it was first attested or the oldest time period for which it can be reconstructed. This
is relatively easy only for languages with a long attested history such as English and
Dutch (where for many loanwords we know the decade in which they were first used,
at least in writing). Even for these languages, the age of very old non-loanwords (those
going back to Proto-Germanic or even Proto-Indo-European) can be estimated only
very roughly. However, such reconstructions are available for many languages even

7.  The following values were used: unanalyzable, 1.00; semi-analyzable, 0.75; analyzable, 0.50.
The minimum score is 0.50 rather than 0 to avoid overweighting this variable.
 Borrowability and the notion of basic vocabulary 237

Table 6.  LWT meanings ranked by age score (top 25)

LWT label Age score
1 fire 0.939
2 water 0.926
3 tongue 0.908
4 nose 0.906
5 wing 0.904
5 mouth 0.904
5 bone 0.904
8 arm/hand 0.903
9 wind 0.900
10 horn 0.898
10 the rain 0.898
10 to take 0.898
13 leg/foot 0.897
13 two 0.897
15 three 0.894
16 one 0.893
16 he/she/it 0.893
16 you (singular) 0.893
19 ash 0.891
20 blood 0.890
21 flesh/meat 0.889
22 ear 0.888
23 go 0.887
24 name 0.886
25 fish 0.885

where there is no long written history (e.g., for Mayan languages, or for Austronesian
languages). We have age information on 88% of our words.
Table 6 gives a ranking of lexical meanings by the average age of the counterparts
in the combined database according to their age scores.8 These scores are deliberately
similar to the borrowability scores (cf. §3), in that the oldest words (which are least
likely to be loanwords) are assigned a value of 1, and those which are attested only very

8.  For this ranking, we assigned the following age values: words first attested or reconstructed
earlier than 1000, 1.00; earlier than 1500, 0.90; earlier than 1800, 0.80; earlier than 1900, 0.70;
earlier than 1950, 0.60; earlier than 2007, 0.50. The age scale is nonlinear to compensate for the
uncertainty in estimates of older ages.
238 Uri Tadmor, Martin Haspelmath and Bradley Taylor

recently and might therefore be (undetected) loanwords are assigned a value of only
0.5. Of course, even the oldest words might be loanwords in the sense that they were
borrowed at an even earlier, prehistoric stage, but the very fact that they are old shows
that they have not been replaced by loanwords for a very long time.
The 25 meanings in Table 10 contain 12 body-part meanings (or 13 if ‘meat’ is in-
cluded), five meanings relating to nature, two basic verbs, three low numerals, and two
personal pronouns. With the possible exception of the numerals, these are culture-free
meanings in the sense that they represent concepts known to every human society.

9. The Leipzig-Jakarta list of basic vocabulary

If we take into account all four factors discussed above (borrowed score, representa-
tion score, analyzability score, and age score) and multiply them by each other, we
arrive at a composite score. The list of the 100 top-ranking items based on this score
are provided in Table 7.9 This is no longer a simple borrowability ranking, because the
borrowability score was just one of four scores used to derive the composite score. In
fact, it is a basic vocabulary list that takes into consideration the features normally
associated with basic vocabulary in historical and comparative linguistics: resistance
to borrowing (the borrowed score), universality (the representation score), simplicity
(the analyzability score), and stability (the age score).
The meanings on the Leipzig-Jakarta list can be broken down into the following
categories (items also on the Swadesh list are shown in boldface):
natural phenomena water, fire, night, wind, rain, smoke, stone/rock, salt, sand,
soil, ash, shade/shadow, star
human body parts nose, mouth, tongue, eye, tooth, hair, ear, arm/hand, neck,
breast, navel, liver, back, leg/foot, thigh, knee, skin/hide,
flesh/meat, bone, blood
animal and plant parts wing, horn, tail, egg, root, leaf, wood
humans and animals child (descendant), fish, bird, dog, ant, fly, head louse
cultural items house, name, rope
properties old, new, big, small, long, wide, far, thick, good, red, black,
heavy, sweet, bitter, hard

9.  The limitation to 100 items is somewhat arbitrary, and is in part in homage to Morris
Swadesh. However, for items significantly lower on the list, their ranking on the combined list is
less significant as it is often due to a single factor. (Incidentally, it was necessary to make several
minor adjustments to the computer-generated ranking, and several meaning labels were edited.
This is described in detail in Tadmor 2009.)
 Borrowability and the notion of basic vocabulary 239

actions go, come, run, fall, carry, take, eat, drink, cry/weep, tie, laugh,
suck, hide, stand, bite, hit/beat, do/make, burn (intr.), blow,
know, see, hear, give, say, crush/grind
deictic/grammatical 1sg pronoun, 2sg pronoun, 3sg pronoun, who?, what?, this,
one, not, yesterday, in

Table 7.  The Leipzig-Jakarta List of Basic Vocabulary

Rank Word meaning Borrowed Age score Analyzabil- Representa- Composite
score ity score tion score score
1 fire 0.965 0.939 0.995 1.000 0.901
2 nose 0.973 0.906 0.980 1.000 0.864
3 to go 0.963 0.887 0.974 1.000 0.832
4 water 0.909 0.926 0.987 1.000 0.831
5 mouth 0.920 0.904 0.982 1.000 0.817
6 tongue 0.934 0.908 0.954 1.000 0.808
7 blood 0.904 0.890 1.000 1.000 0.805
7 bone 0.918 0.904 0.971 1.000 0.805
9 2sg pronoun 0.958 0.893 0.933 1.000 0.798
9 root 0.944 0.869 0.973 1.000 0.798
11 to come 0.968 0.876 0.940 1.000 0.796
12 breast 0.947 0.856 0.967 1.000 0.783
13 rain 0.916 0.898 0.950 1.000 0.782
14 1sg pronoun 0.970 0.875 0.936 0.976 0.776
15 name 0.915 0.886 0.955 1.000 0.774
15 louse 0.950 0.861 0.946 1.000 0.774
17 wing 0.884 0.904 0.968 1.000 0.773
18 flesh/meat 0.877 0.892 0.986 1.000 0.771
19 arm/hand 0.881 0.903 0.966 1.000 0.768
20 fly 0.948 0.858 0.942 1.000 0.766
20 night 0.931 0.880 0.934 1.000 0.766
22 ear 0.896 0.888 0.961 1.000 0.764
23 neck 0.895 0.881 0.964 1.000 0.760
23 far 0.944 0.850 0.948 1.000 0.760
25 to do/make 0.947 0.877 0.914 1.000 0.759
26 house 0.893 0.876 0.969 1.000 0.758
27 stone/rock 0.895 0.882 0.958 1.000 0.756
28 bitter 0.975 0.872 0.889 1.000 0.755
28 to say 0.972 0.837 0.928 1.000 0.755
28 tooth 0.882 0.877 0.975 1.000 0.755
240 Uri Tadmor, Martin Haspelmath and Bradley Taylor

Table 7.  (continued)
Rank Word meaning Borrowed Age score Analyzabil- Representa- Composite
score ity score tion score score
31 hair 0.944 0.871 0.917 1.000 0.754
32 big 0.889 0.864 0.980 1.000 0.753
32 one 0.870 0.893 0.969 1.000 0.753
34 who? 0.968 0.838 0.924 1.000 0.749
34 3sg pronoun 1.000 0.893 0.955 0.878 0.749
36 to hit/beat 0.955 0.827 0.947 1.000 0.748
37 leg/foot 0.856 0.897 0.972 1.000 0.747
38 horn 0.840 0.898 0.987 1.000 0.745
38 this 1.000 0.851 0.897 0.976 0.745
38 fish 0.855 0.885 0.984 1.000 0.745
41 yesterday 0.958 0.843 0.922 1.000 0.744
42 to drink 0.904 0.877 0.934 1.000 0.741
42 black 0.951 0.866 0.899 1.000 0.741
42 navel 0.878 0.860 0.982 1.000 0.741
45 to stand 0.981 0.847 0.889 1.000 0.738
46 to bite 0.964 0.861 0.887 1.000 0.736
46 back 0.918 0.868 0.924 1.000 0.736
48 wind 0.828 0.900 0.987 1.000 0.736
49 smoke 0.916 0.863 0.929 1.000 0.734
50 what? 0.971 0.804 0.939 1.000 0.732
51 child (kin term) 0.929 0.866 0.930 0.976 0.730
52 egg 0.910 0.846 0.945 1.000 0.728
53 to give 0.913 0.878 0.907 1.000 0.727
53 new 0.920 0.860 0.920 1.000 0.727
53 to burn (intr.) 0.951 0.860 0.889 1.000 0.727
56 not 0.965 0.880 0.974 0.878 0.726
56 good 0.893 0.860 0.945 1.000 0.726
58 to know 0.933 0.856 0.908 1.000 0.725
59 knee 0.911 0.862 0.922 1.000 0.724
59 sand 0.901 0.866 0.928 1.000 0.724
61 to laugh 0.942 0.844 0.910 1.000 0.723
61 to hear 0.953 0.848 0.895 1.000 0.723
63 soil 0.900 0.883 0.954 0.951 0.722
64 leaf 0.897 0.823 0.977 1.000 0.721
64 red 0.926 0.864 0.900 1.000 0.721
66 liver 0.869 0.857 0.967 1.000 0.720
 Borrowability and the notion of basic vocabulary 241

Table 7.  (continued)
Rank Word meaning Borrowed Age score Analyzabil- Representa- Composite
score ity score tion score score
67 to hide 0.928 0.847 0.913 1.000 0.718
67 skin/hide 0.889 0.875 0.924 1.000 0.718
67 to suck 0.940 0.860 0.888 1.000 0.718
70 to carry 0.919 0.838 0.953 0.976 0.717
71 ant 0.865 0.850 0.975 1.000 0.716
71 heavy 0.911 0.874 0.901 1.000 0.716
71 to take 0.900 0.898 0.887 1.000 0.716
74 old 0.896 0.867 0.920 1.000 0.715
75 to eat 0.920 0.840 0.925 1.000 0.714
76 thigh 0.906 0.856 0.918 1.000 0.712
76 thick 0.950 0.827 0.906 1.000 0.712
78 long 0.956 0.824 0.898 1.000 0.707
79 to blow 0.962 0.857 0.878 0.976 0.706
80 wood 0.860 0.871 0.940 1.000 0.705
81 to run 0.976 0.833 0.867 1.000 0.704
81 to fall 0.946 0.825 0.903 1.000 0.704
83 eye 0.904 0.847 0.918 1.000 0.703
84 ash 0.853 0.891 0.921 1.000 0.699
84 tail 0.883 0.813 0.973 1.000 0.699
84 dog 0.838 0.869 0.960 1.000 0.699
87 to cry/weep 0.871 0.871 0.921 1.000 0.698
88 to tie 0.879 0.836 0.948 1.000 0.697
89 to see 0.918 0.842 0.900 1.000 0.695
89 sweet 0.914 0.857 0.887 1.000 0.695
91 rope 0.848 0.824 0.993 1.000 0.694
91 shade/shadow 0.887 0.840 0.931 1.000 0.694
91 bird 0.842 0.857 0.962 1.000 0.694
91 salt 0.848 0.838 0.976 1.000 0.694
91 small 0.909 0.790 0.966 1.000 0.694
96 wide 0.955 0.819 0.885 1.000 0.692
97 star 0.830 0.859 0.970 1.000 0.691
97 in 0.948 0.856 0.943 0.902 0.691
99 hard 0.918 0.833 0.903 1.000 0.690
100 to crush/grind 0.919 0.845 0.886 1.000 0.688
242 Uri Tadmor, Martin Haspelmath and Bradley Taylor

10. The Leipzig-Jakarta list vs. the Swadesh 100 list and three other
stability lists

There is a fair degree of correlation between the Swadesh 100 list and the Leipzig-
Jakarta list: 62 items on the lists overlap (those in boldface in the list). This means that
a total of 38 items on the Leipzig-Jakarta list do not appear on the Swadesh list and vice
versa. Swadesh’s intuitions thus turn out to have been good, although a 38% difference
is substantial and can lead to rather different lexicostatistical results. Moreover, our
findings indicate that quite a few items on the Swadesh list are not basic (see the rank-
ings in Table 8). At any rate, the major advantage of the Leipzig-Jakarta list is that it has
a strong empirical foundation and is thus a more reliable tool for scientific purposes.
Table 9 compares the 100 meanings of the Leipzig-Jakarta list (in the middle col-
umn) with the meanings on three stability lists: Dolgopolsky’s (1986) list of 23 stable
meanings, Lohr’s (1998: 54) list of 61 meanings, and the ASJP’s list of 40 meanings
(Holman et al. 2008: 336–339, 351–352). Dolgopolsky’s and Lohr’s lists were estab-
lished by looking at the kinds of meanings expressed by words that are reconstructed
for protolanguages of various families. The ASJP list consists of the 40 most stable
meanings of the Swadesh 100 list, where stable meanings are identified due to their
greater tendency to yield cognates within groups of closely related languages.
What is striking in particular is that many adjectival and verbal meanings appear
on the Leipzig-Jakarta list, but are not part of the stability lists. It thus seems that some
meanings may be subject to change (e.g., semantic change, or replacement by novel
formations), but not so much subject to borrowing.

Table 8.  Items on the Swadesh list but not on the Leipzig-Jakarta list
Item Our ranking Item Our ranking Item Our ranking
sit 106 sleep 155 bark 301
fingernail 107 white 157 walk 321
man 115 kill 159 swim 322
belly 118 many 166 seed 327
two 119 that 174 all 338
lie 121 sun 178 tree 345
cloud 123 woman 183 we 347
fly 134 dry 192 moon 358
head 137 grease 219 round 376
hot 143 heart 220 green 412
cold 146 yellow 232 person 531
feather 147 path 271
full 153 die 291
 Borrowability and the notion of basic vocabulary 243

Table 9.  Items on the Dolgopolsky list (boldface), the Lohr 61 list (italics), and the ASJP
40 list (underlined)
semantic domain meanings on Leipzig-Jakarta list other meanings
natural phenomena water, fire, night, wind, rain, smoke, stone/ mountain, sun, day, dark-
rock, salt, sand, soil, ash, shade/shadow, ness, light, moon, sky
star
human nose, mouth, tongue, eye, tooth, hair, ear, fingernail, heart, head,
body parts arm/hand, neck, breast, navel, liver, back, shoulder, stomach
leg/foot, thigh, knee, skin/hide, flesh/meat,
bone, blood
animal and plant wing, horn, tail, egg, root, leaf, wood tree
parts
humans and animals child (descendant), fish, bird, dog, ant, fly, person, snake
(head) louse
cultural items house, name, rope path
properties old, new, big, small, long, wide, far, thick, full, other, thin
good, red, black, heavy, sweet, bitter, hard
actions go, come, run, fall, carry, take, eat, drink, die, breathe, grind, cut,
cry/weep, tie, laugh, suck, hide, stand, bite, defecate, grow, lie (down),
hit/beat, do/make, burn (intr.), blow, know, press, sleep, smell (intr.),
see, hear, give, say, crush/grind split, spread, turn, wrap
deictic/ 1sg pronoun, 2sg pronoun, 3sg pronoun, two, three, four, we, over
grammatical who?, what?, this, one, not, yesterday, in

11. Conclusions

The large-scale comparative study of loanwords that we carried out with more than 40
colleagues has yielded many results, only some of which we have reported on here. We
found that nouns are borrowed much more often than verbs and adjectives, and that
content words are borrowed much more often than function words. We also found
that different semantic fields of words show different degrees of borrowability, as was
suspected before.
We then focused on establishing a list of the least borrowable word meanings,
to be compared with Swadesh’s 100-item list that was intended to be a list of basic,
non-cultural items, not necessarily a list of borrowing-resistant meanings. However,
deriving a single ranking from our data was not straightforward, because a number of
factors other than the raw borrowability rate were found to be important: the represen-
tation rate (the degree to which meanings have counterparts in our data), analyzability
(the degree to which words have complex counterparts), and age. These correspond to
the notions of universality, simplicity, and stability, respectively, which have long been
244 Uri Tadmor, Martin Haspelmath and Bradley Taylor

associated with the notion of basic vocabulary in historical and comparative linguis-
tics. Thus, we ended up with a list that takes into account all of these factors, which we
call the Leipzig-Jakarta list of 100 basic word meanings. This list is similar to Swadesh’s
100-item list and various other lists of stable word meanings, but resistance to borrow-
ing and stability are not exactly the same concepts.
The language sample and the resulting database are not free from bias, and the way
the results were calculated was necessarily arbitrary to a certain extent. Nevertheless,
we consider the Leipzig-Jakarta list to have an advantage over Swadesh’s 100-item list,
in that it was empirically derived, rather than being based on intuition.

References

Aikhenvald, Alexandra Y. & R.M.W. Dixon, eds. 2007. Grammars in Contact: A cross-linguistic
typology. Oxford: Oxford University Press.
Buck, Carl Darling. 1949. A Dictionary of Selected Synonyms in the Principal Indo-European
Languages. Chicago: University of Chicago Press.
Dolgopolsky, Aharon B. 1986. “A probabilistic hypothesis concerning the oldest relationships
among the language families in northern Eurasia”. Typology, Relationship and Time: A col-
lection of papers on language change and relationship by Soviet linguists ed. by Vitalij V.
Shevoroshkin & Thomas L. Markey, 27–50. Ann Arbor: Karoma.
Field, Fredric W. 2002. Linguistic Borrowing in Bilingual Contexts. Amsterdam: Benjamins.
Haspelmath, Martin & Uri Tadmor, eds. 2009a. Loanwords in the World’s Languages: A compara-
tive handbook. Berlin: De Gruyter Mouton.
Haspelmath, Martin & Uri Tadmor, eds. 2009b. World Loanword Database. Munich: Max Planck
Digital Library. http://wold.livingsources.org/
Holman, Eric W., Søren Wichmann, Cecil H. Brown, Viveka Velupillai, André Müller & Dik
Bakker. 2008. “Explorations in automated language classification”. Folia Linguistica
42:2.331–354.
Hymes, Dell. 2006 [1971]. “From the First Yale School to World Prehistory”. The Origin and
Diversification of Language ed. by Sherzer Joel, 228–270.
Lohr, Marisa. 1998. Methods for the Genetic Classification of Languages. PhD dissertation, Uni-
versity of Cambridge.
Matras, Yaron, 1998. “Utterance modifiers and universals of grammatical borrowing”. Linguistics
36:2.281–331.
Matras, Yaron, 2007. “The borrowability of structural categories”. Grammatical Borrowing in
Cross-Linguistic Perspective ed. by Yaron Matras & Jeanette Sakel, 31–73. Berlin: Mouton
de Gruyter.
Matras, Yaron & Jeanette Sakel, eds. 2007. Grammatical Borrowing in Cross-Linguistic Perspec-
tive. Berlin: Mouton de Gruyter.
McCarthy, Michael. 1999. “What constitutes a basic vocabulary for spoken communication?”
English Language and Literature 1.233–249.
Swadesh, Morris, 1950. “Salish internal relationships”. International Journal of American Linguis-
tics 16:4.155–167.
 Borrowability and the notion of basic vocabulary 245

Swadesh, Morris, 1952. “Lexico-statistic dating of prehistoric ethnic contacts: With special ref-
erence to North American Indians and Eskimos”. Proceedings of the American Philosophical
Society 96:4.452–463.
Swadesh, Morris, 1955. “Towards greater accuracy in lexicostatistic dating”. International Jour-
nal of American Linguistics 21:2.121–137.
Swadesh, Morris, 2006 [1971]. The Origin and Diversification of Language. New Brunswick, New
Jersey: Transaction Publishers.
Tadmor, Uri, 2009. “Loanwords in the world’s Languages: Findings and results”. Loanwords in
the World’s Languages: A comparative handbook ed. by Martin Haspelmath & Uri Tadmor,
55–75. Berlin: De Gruyter Mouton.
Wohlgemuth, Jan. 2009. A Typology of Verbal Borrowings. (= Trends in Linguistics 211.) Berlin:
Mouton de Gruyter.

Résumé

Cet article présente une étude en collaboration concernant des emprunts lexicaux dans 41 lan-
gues dans le but d’identifier des signifiés ou des groupes de signifiés qui résistent à l’emprunt.
Nous constatons que les noms s’empruntent plus facilement que les verbes ou les adjectifs, que
des mots lexicaux s’empruntent plus facilement que des mots à fonction grammaticale et que
différents champs sémantiques contiennent différentes proportions d’emprunts lexicaux. Si on
essaie de dresser une liste des signifiés le moins souvent empruntés différents problèmes surgis-
sent : nos données fournissent des informations sur le degré de probabilité d’un emprunt, cer-
tains signifiés n’ont pas d’équivalents dans toutes les langues, beaucoup de mots sont des com-
posés ou des dérivés et, de ce fait, sont presque par définition des termes non-empruntés. Nous
disposons également d’informations concernant l’âge des mots. De multiples facteurs rentrent
donc en jeu et nous proposons une combinaison des facteurs dans une liste de cent signifiés du
vocabulaire de base : la liste Leipzig-Jakarta.

Zusammenfassung

In diesem Aufsatz berichten wir über ein Gemeinschaftsprojekt zur Erforschung von Lehnwörtern
in 41 Sprachen, mit dem Ziel, entlehnungsresistente Bedeutungen und Bedeutungsgruppen zu
identifizieren. Es stellt sich heraus, dass Substantive entlehnbarer als Adjektive oder Verben sind,
dass Inhaltswörter entlehnbarer als Funktionswörter sind, und dass verschiedene semantische
Felder verschiedene Anteile an Lehnwörtern zeigen. Bei dem Versuch, eine Liste der entlehnungs-
resistentesten Bedeutungen zu erstellen, ergeben sich einige Probleme: Wir haben Informationen
über mehr oder wenig wahrscheinliche Entlehnung, nicht alle Bedeutungen haben Entsprechun-
gen in allen sprachen, viele Wörter sind Komposita oder Ableitungen und daher fast per definitio-
nem Nichtlehnwörter, und wir haben auch Daten über das Alter der Wörter. Es spielen also viele
Faktoren eine Rolle, und wir schlagen eine bestimmte Kombination der Faktoren vor, die eine
neue Liste von 100 Bedeutungen für Grundvokabular ergibt, die Leipzig-Jakarta-Liste.
246 Uri Tadmor, Martin Haspelmath and Bradley Taylor

Authors’ addresses
Uri Tadmor Bradley Taylor
Max Planck Institute for Evolutionary Max Planck Institute for Evolutionary
Anthropology Anthropology
Department of Linguistics Department of Linguistics
Deutscher Platz 6 Deutscher Platz 6
04103 Leipzig, Germany 04103 Leipzig, Germany
[email protected] [email protected]
Martin Haspelmath
Max Planck Institute for Evolutionary
Anthropology
Department of Linguistics
Deutscher Platz 6
04103 Leipzig, Germany
[email protected]