Proceedings of the Third ACL-SIGSEM Workshop on Prepositions, pages 51–56,
Trento, Italy, April 2006. c©2006 Association for Computational Linguistics
A Multilingual Analysis of the Notion of Instrumentality
Asanee Kawtrakul, Mukda Suktarachan (Kasetsart univ. Bangkok, Thailand),
Bali Ranaivo-Malancon, Pek Kuan Ng, (Univ. Sains Malaysia, Penang, Malaysia),
Achla Raina (IIT Kanpur, India),
Sudeshna Sarkar (IIT Kharagpur, India),
Alda Mari (Enst-Cnrs, Paris, France),
Sina Zarriess (Universit¨at Potsdam, Germany),
Elixabete Murguia (Univ. Deusto, Bilbao, Spain),
Patrick Saint-Dizier (Irit-Cnrs, Toulouse, France)
Abstract
Instruments are expressed in language by
various means: prepositions, postposi-
tions, affixes including case marks, nonfi-
nite verbs, etc. We consider here 12 lan-
guages from five families in order to be
able to identify the different meaning com-
ponents that structure instrumentality.
1 Credits
This work has been made possible partly via the
STIC-Asia cooperation framework.
2 Introduction
It is difficult to give a comprehensive definition of
what instrumentality is. In WordNet it is defined
as ’an artifact, or a set of artifacts, that are instru-
mental (i.e. behave as instruments) in accomplish-
ing some end’, i.e. reaching a certain goal. In
this definition, the triple relation agent-instrument-
goal (as in: John cuts the bread with a knife, where
John is agent, knife is instrument that does the cut-
ting, and bread cut is the goal), is left vague in
what concerns the exact involvement of the agent
and the instrument in the action, and the control
the agent has on the instrument and on the action
(Mari and St-Dizier 01).
If almost anything can be an instrument, we can
nevertheless formulate a few criteria that we will
try to elaborate in this paper. First, an instrument
is basically non volitional. When humans play the
role of instruments, they are obviously volitional,
but the action is controlled by another agent who
acts as an ’initiator agent’, taking the initiative of
the action. Instruments cannot be easily associated
with traditional thematic roles, in fact this is not
of much interest, because this is too superficial a
notion and also because instruments are generally
modifiers, not arguments.
In this paper, we address instrumentality as con-
veyed by prepositions or equivalent means (e.g.
postpositions, affixes). Our aim in this study
is twofold: (1) to identify the conceptual facets
of instrumentality so that a conceptual seman-
tics can be defined in the spirit of (Talmy, 01,
03), (Wierzbicka 92, 96) and (2) to elaborate an
accurate enough model for answering questions
about instruments within a cooperative question-
answering system. Instead of focusing on a spe-
cific language to elaborate all possible forms of
instrumentality, we found it more adequate to de-
velop a multilingual approach, considering lan-
guages from various families.
We consider for Europe: German, Spanish,
French, Italian; for India: Kashmiri, Urdu, Hindi
and Bengali; for the far-east: Thai and Malay;
for Northern Africa and the Middle East: Ara-
bic, and Berber dialects (in the group of Amazigh
languages). In this paper, we use the first let-
ter of each language to identify it: Thai, Malay,
Hindi, Urdu, Kashmiri, Bengali, German, Span-
ish, French, Italian, Arabic, BeRber). We have
also more or less adequately transcribed characters
into latin characters. The upper case A is equiva-
lent to aa.
3 An overview of preposition structures
3.1 European languages
German, Spanish, Italian and French, like most
European languages have prepositions that in-
troduce instrumental PPs. The most current
prepositions are:
- German: mit, mit Hilfe von,
mittels, durch, anhand, kraft,
51
dank, per.
- French: avec, par, au moyen de,
grˆace `a, `a l’aide de, `a travers.
- Spanish: con, en, por, a trav´es de,
mediante, por medio de, a base
de, con la ayuda de, gracias a.
- Italian: con, per mezzo di, tramite,
per, grazie a, con l’aiuto di.
3.2 Arabic and Berber
Arabic essentially has the preposition bi, used as
a prefix of the noun it heads:
Aktoub bi al kalami (I write with a pen).
bi can also be associated with a specific noun
or deverbal form (e.g. ’by applying’) to char-
acterize in more depth the instrumental relation.
This entails generic forms such as bi-tarika,
bi-istemali (by means of, the first form rein-
forces the importance of the instrument, while the
latter is more formal) and bi-fadli (thanks to).
For example, we have a kind of nominal form is-
tikhdam (= using or with the use of, constructed
from the root ’use’) in:
Fasser el massala bi istikhdam mithel (explain a
problem ’with the use of’ example). Arabic makes
explicit the metonymy we have in European lan-
guages.
There are a few other prepositions such as min
khilal (through) and min a to express the du-
ality source + instrument of the argument (as in
drink with a bottle, litterally). In the spatial con-
text, au can be used (to reach the top by this trail),
and ala is used to express a channel of commu-
nication.
Berber is composed of a large number of di-
alects, some just spoken in small ’tribes’, others
in larger communities. It is basically dialectal, but
a number of common elements can be identified.
We consider here Berber from the Moroccan Rif,
and the Algerian Kabyle. The main instrumental
preposition is `eg, but prefixes are also used s-,
ge-, th-, kh- which affect the morphology
of the noun they are attached to. The prefix s-
is widely used, with some variants (si, sei,
so in Kabyle), `eg (also realized as g’ or ge in
some places), is appropriate only when the action
is under full control of the agent. It is not em-
ployed when the instrument is abstract. In the Rif
area, s-focusses on the instrument contributing to
the action, it makes a kind of fusion between the
action verb and the instrumental noun. ge is only
used in spatial contexts involving means of trans-
portation (travelling by plane), kh is used only in
spatial contexts involving paths.
3.3 The Indic language family
The four languages considered, Kashmiri (Raina,
02), Urdu, Hindi and Bengali, share some simil-
itudes due to their common origins but also
contrasts that result from independent evolutions
which are useful to our analysis. They all have suf-
fixes and postpositions. Case suffixes (vibakhti)
are added to nouns or pronouns, but case suf-
fixes do not correspond strictly to thematic roles
(kAraka). Postpositions may also be used instead
of vibhaktis or in conjunction with them, as a kind
or re-inforcement. Basically, but with some nu-
ances, Hindi, Urdu, Bengali and Kashmiri lan-
guages have the following kernels:
• direct instrument: se (H/U), sity (K), -e/-te,
diye after null, dwArA after -r (B). Example:
H: raam ne chaabi se taala khola (ram - erg.
key with lock open-past= Ram openened the
lock with a key).
• means instrument: me (H/U), ke zariye (U),
manz (K), -e/-te (B). Example: U: raam
gaadi ke zariye daftar gayaa (ram car by
means of office go-past = Ram went to work
by car).
• causal instrument: ke kaaran (H), ki vajah se
(U), kiny (K), kArANe, kripAya (B). Exam-
ple: K: dil chi sayaahat kiny amir (Delhi be-
present tourism with rich = Delhi is rich with
tourism),
B: dillI paryatan-er khAtire samriddha or dillI
paryatan-er kripAya samriddha (Delhi be-
rich-er thanks to tourism)
• agentive instrument: ke dwaraa (H) ke zariye
(U) zariy, desi (K), diye (B) after -ke when
the nominal form is animate and specific,
diye after null when the nominal form is an-
imate and general (B). Example: H: raam ne
shyaam dwara apna kaam karvaaya (ram erg.
shyam by self’s work do-cause-past = Ram
got his work done by Shyam)
• action instrumentalised: kar (H/U), kerith
(K). Example: U: raam kuud kar ghar ke an-
dar daakhil huaa (ram jump participle house
52
into enter-past = Ram entered the house by
jumping in it)
There are less important cases that capture no-
tions like containment, which will be detailed in
section 5. Postpositions may vary also depending
on the semantic type of the NP they head.
In Bengali, case is indicated in several ways.
Vibhaktis are suffixes that are added to the stems
to form surface forms of words. In Bengali the
nominal stems take one of the following suffixes:
null or shunya vibhakti, -e, -te, -r, etc. Case is also
indicated in Bengali by the use of postpositions.
We have, for example: Cut bread with a knife, re-
alized as either:
1. chhuri diye ruti kATa (knife diye bread cut), or:
2. chhuri-te ruti kAta (knife-te bread cut)
Postpositions in Bengali are derived from certain
inflected forms of nouns, and also certain verbs
in participle form. When these words are used
as postpositions they are often not considered in
their original sense but define a specific type of re-
lationship of the noun phrase with the finite verb
phrase in the sentence. These words are appear
in a fixed form (indeclinables) as postpositions.
When a postposition is used to denote the case,
the nominal word preceding the postposition takes
on a vibhakti that is determined by the particu-
lar postposition. ’diye’ is used after null vibhakti,
’dwArA’ after -r vibhakti, etc.
3.4 Thai and Malay
Thai and Malay, although spoken in neighbour
countries, are substantially different.
Thai, from the Thai-Kadai family, has 6 prepo-
sitions (Silapasarn, 98) to denote instruments, the
most common being doi and duai which are
used for concrete instruments, means of trans-
portation, instruments close to manners, etc. kap
is used when the instrument is a part of the body,
while thang is used for means of transportation
only. tam characterizes control of the agent, and
chak is restricted to the instruments that convey
an idea of source. Examples:
khian - duai - din so (Write - with - pencil)
pai - pa ris - doi - khrueang bin (Go - Paris - by -
plane)
These semantic distinctions are, however, often vi-
olated in colloquial Thai.
Malay, from the Malayo-polynesian family, has
three ways to introduce instruments: preposition +
NP, affixes and compounding. Affixed words are
built from stems which are instrumental nouns,
this allows for the construction of the equivalent
of PPs, based on the prototypical use of the instru-
mental noun. The most common being: prefixes:
beR- (from kuda, horse, berkuda, on horseback),
meN- (from kunci, key, mengunci, lock with key),
prefix + suffix: meN- + -kan (from paku, nail,
memakukan, to fasten with nails), and with suffix
-i (from ubat, medicine, mengubati, by means of
medicine). Prepositions occur as the head of PPs,
and in verb particle constructions. PPs may also
be subject complements, avoiding the use of verbs
(dia di rumah, she at home). Besides affixes,
Malay has 6 prepositions that denote instrumen-
tality: dengan, melalui, mengikut,
menerusi, dengan menggunakan,
secara.
A simple example is:
berhubung - melalui - telefon (communicate - by -
telephone).
4 The meaning components of
instrumentality
Let us now consider the different meaning compo-
nents that emerge from our multilingual analysis.
The results presented below are still exploratory
due to the complexity of the notion. The distinc-
tions made (e.g. between concrete and abstract in-
struments) may seem arbitrary: they are just meant
to structure the presentation.
4.1 Concrete instruments
All languages studied have at least one basic in-
strumental mark operating over concrete objects
(T: duai, M: dengan, H: se, U: se, K: sity, B: diye,
-e, -te, G: mit, S: con, F: avec, A: bi, BR: `eg). Sev-
eral refinements are identified, for specific types of
NPs, or to denote a specific intention:
• the instrument is a recipient (S: en) or, more
generally, conveys an idea of container (e.g.
spoon) (B: -e kare), the idea behing is that
the container is used to carry the object along
a certain trajectory,
• the instrument is a part of the body (e.g.
hand): T: kap. In this case, the instrument
is not strictly artifactual.
• the goal is difficult to reach, it requires some
efforts from the agent (S: a base de),
53
• the focus can be emphasized by using ded-
icated marks (G: mit Hilfe (von), Mittels
(more fomal: Das Gericht hat mittels einst-
weiliger Verf¨ugung den Drogenhandel unter-
sagt (the court has with provisional ordinance
the drug traffic prohibited))).
The second major difficulty is prototypicality
(Rosch, 78). When the instrument used is not very
prototypical of the action, several languages re-
inforce the instrumental prepositions to, sort of,
coerce the type of the noun so that it can become
an acceptable instrument. We have examples in S:
por medio de,B:sAhAjye, sahojoge,
I: per mezzo di,F:au moyen de, par
le biais de (biais= bias which directly ex-
presses this idea), as in:
F: Il a ouvert la porte au moyen d’un cric (he
opened the door by means of a jack).
At a conceptual level, it is quite difficult to
characterize what is a prototypical instrument
for a given action (characterized by subject-verb-
object: John opens the door). Each event has its
own prototypical instrument, making corpus stud-
ies extremely large, probably unfeasable. When
searching on the web, we find an incredible variety
of instruments to open a door, almost impossible
to classify. Next, prototypicality is not a boolean
notion: instruments are more or less prototypical.
Since the instrument is very much dependent on
the verb and on the object, we cannot foresee any
form of incorporation in the verb that would give
us indications. A direction could be to assume
Qualia structures (Pustejovsky 91) associated with
each potential instrument that describes the func-
tion of the object in the telic role. For example,
key(X) would have open(X, door), with door be-
ing quite generic. This approach could work via a
large lexical development for concrete nouns, it is
much more risky when terms are abstract.
4.2 Abstract instruments
Abstract instruments (theorems, regulations, ex-
amples, etc.) are realized identically to concrete
instruments, but with some typical marks such
as: T: tam, H: dwAra, K: zariyi, B: dwArA,
M. mengikut. At this stage, it is difficult to ex-
plain why marks are different from concrete in-
struments. An hypothesis could be that abstract in-
struments are closer to causes (see 5.5), or to more
formal situations for which specific terms were de-
veloped (e.g. for G: kraft).
There are additional marks dedicated to partic-
ular fields: B: sahajoge, and A: min khilal
when instruments are of type ’example’ (explain
with an example). U: -ke zariye,S:por
medio de and G: Anhand, Kraft are more
formal, stronger for Kraft and apply particularly
to areas like juridical or psychological domains.
People and organizations can be seen as appro-
priate intermediaries for reaching a goal. They
may be conceived as metaphorical instruments.
Investigations show that people can get controlled
much in the same way as concrete objects:
F: Elle a inform´e Paul de son d´epart par Pauline
(She informed Paul of her leaving ’by’ Pauline).
If we now consider: S: Juan env´ıo este paquete
por correo (John sent this parcel ’by’ post)
Since post is the by-default medium to send pack-
ages, por is the only choice. Using more precise
services, like FedEx, is considered to be an alter-
native way, in that case F: par, avec S: por,
con are both acceptable.
4.3 Metaphorical instruments
Both concrete and abstract objects can be used
metaphorically as instruments. Examples abound
in the literature and on the Web. In 5.6 we ex-
amine the path metaphor which is very produc-
tive. Besides this case, we have a number of
metaphors, such as: write with your heart, fight
with your head, etc. These are not essentially dif-
ferent from metaphors observed in other situations
(Lakoff and Johnson 99).
4.4 The overlap instrument-manner
In a number of cases, it is not very easy to make
a distinction between instrument and manner. It
seems there is a continuum between these two no-
tions or even some form of overlap, where the ob-
ject is both an instrument and a manner at various
degrees, which may depend on context. A vari-
ety of marks contribute to characterize this over-
lap, manners at stake being quite diverse, but we
will not go into the study of manners. Specific
marks dealing with the manner/instrument am-
biguity are: T: doi,G:durch (which is also
used for metaphorical spatial uses), M: dengan
menggunakan,S:en, con, a as in S: es-
cribir en/con rojo (write in red),
T: khian - duai - muek - daeng (write - with - red -
ink)
BR: t`ete s-´ef`essen (She-eats with-hands).
54
4.5 Causality
It is clear that, a priori, instruments can be viewed
at various degrees as causes of an event. There is
a kind of overlap between these two notions. In-
struments are not volitional, so they are under the
partial or full control of an agent (humans playing
the role of instruments are also controlled by an
agent). Typically I: a causa di,F:a cause
de,S:a causa de signal that the instrument
has brought about an event:
I:Il castello e distrutto a causa di un violento in-
cendio. (The castle has been destroyed ’because
of’ a violent fire.)
Causality (e.g. Talmy, 01) being a complex no-
tion, it is not surprising that instruments, viewed
as intermediaries at various degrees, share some
features with causes. For example in cut the bread
with a knife, the cause of the bread being cut is
the action of the agent, but also the use of a pro-
totypical property of the knife: the knife does the
cutting. In (Talmy 01), the instrument is embed-
ded into the causing event:
(caused event) RESULTS FROM (causing event)
where the causing event has the structure:
Instrument ACT ON object, where object is bound
or related in some way to the object in the caused
event.
As analyzed in (Mari and Saint-Dizier, 01), in-
strumentality is the convergence of several factors:
• the degree of involvement of the instrument
in the action, therefore, the fact that the in-
strument causes the action or is just a means
managed by the agent who is the main cause,
• the type of control the agent has on the instru-
ment for the action at stake, from full control
to lack of control,
• the control the agent has over the action as a
whole.
Indic languages and Thai are particular explicit on
these matters. They have specific marks for two
major cases:
1. agentive instrument, action not controlled by
the agent: H: ke dwAra,U:ke zariye
K: zariy, desi,T:doi,
2. causal instrument that does most of the ac-
tion, under the control of the agent: H ke
kAran,U:ki vajah se,K.kiny,T:
duai,
Berber allows `eg only when the agent controls the
instrument. The other cases are expressed by non
prepositional forms.
4.6 Instruments and paths
Another productive situation is the use of spatial
metaphors to express instrumentality. The use of
F: par and other marks (e.g. in B., U.), show
that there is a close link between instrumentality
and path descriptions (spatial as well as temporal
paths). This is a kind of metaphorical use of paths
viewed as instruments (as can be seen in (Lakoff
et al. 99): ’action is motion, goals are paths, actors
are travellers’). Using an instrument parallels the
use of paths in the domain of space.
Marks denoting paths or sources are of much
interest. Some have really restricted uses, whereas
others are more flexible. We observe the following
main components:
• paths: T: tam, A: min khilal, S: por, a trav´es
de(por correo, by post), de, G: durch, F:
`a travers, note the distinctions, e.g. in M:
melalui (metaphorical paths: M : berhubung
melalui telefon (communicate by telephone)),
menerusi (channel of transmission), H, U:
me, se, T: thang. In B, -e and -te denote
paths where the agent that does the action has
no control, whereas diye and dhare involve at
least a partial control from the agent. In M,
metaphorical passages require melalui.
• sources: F: `a, A: min a, T: chak (for con-
crete and abstract sources). Example: A:
Achroubou mina Karoura (I am drinking with
bottle), which is also a kind of manner.
The duality path/instrument is particularly visi-
ble in, e.g.:
K: raam vot tshochi vati kiny gari (ram reach-past
short route via home = Ram reached home by the
short route).
Another interesting phenomenon occurs when
an argument is both an instrument and a path,
as in look at the moon in a telescope. Tele-
scope is indeed the instrument used and also
the path through which one looks, or which the
light traverses. This double facet of the argu-
ment is visible in surface realizations, where the
preposition used is ambiguous between instrument
(first preposition) and path (second one) readings:
G: mit, durch,S:con, por,M:dengan,
melalui,F:avec, dans. When one wants
55
to strongly stress the path interpretation, then a
more path-oriented preposition is used, e.g. S: a
trav`es de.
4.7 Means of transportation as instruments
Means of transportation (trains, spoons, boxes, en-
velopes, etc.), sometimes viewed as containers,
and mediums of transportation (by air) receive a
special treatment in a number of languages: T:
doi, thang, M: menerusi, melalui (for metaphorical
mediums and passages), H: me, U: ke zariye, K:
manz, zariy, B: kare, -e kare, -ya kare, A: ala, BR:
ge-, kh-, G: per, S: por, en, F: par. We have, for ex-
ample: U: raam gaadi ke zariye daftar gayaa (ram
car by means of office go-past = ram went to post
office by car)
T: pai - pa ris - doi - khrueang bin (Go to - Paris -
by - plane)
B: Nouko-ya kare phuketa jAo or Nouko-ya
phuketa jAo (boat-e kare phuket go or boat-e
phuket go = go by boat to Phuket)
A distinction is made between the medium and
the means as for: M: secara, which is used for
means of communication such as email or letters.
If the agent has effective control over the means,
then, for example, S uses con.
4.8 Language levels
Some marks are proper to formal discourse: G:
Mittels, Kraft, Anhand, H: dwAra.
4.9 Positive or negative orientation
The languages we studied also abound in positive-
oriented marks that express in a certain way the
idea of ’thanks to’: T: khop khun (+kah for fem-
inime and krup for masculine), H: ke kAran, U:
ki vajah, K: kiny, B: -er khAtire, (-er) kripAya, G:
dank, S: gracias a, F: grˆace `a.
There are also several negative-oriented marks
such as the following prepositional compounds:
F: de la faute de,I:per colpa di,S:
por colpa de (by the fault of), where the term
’fault’ conveys a negative orientation.
4.10 Metonymies
In most languages, the prototypical action denoted
by the instrument is implicit, it is analyzed as a
metonymy: object for action. Action is inferred
from the instrument and the verb in the given con-
text. In a number of situations, A and M need
to make explicit the action. For particular cases,
gerundive forms may be prefered to PPs (but not
to be confused with manners, e.g. ’by swim-
ming’), so that the verb that lexicalizes the action
is present.
For example, in M, ’by the trail’ in to reach the
top of a mountain by the trail requires to make ex-
plicit how the trail is used: dengan mengikuti de-
nai itu (litt.: with follow trail DET (same cases
in Arabic and German)). Another case is: G:
Mit Flugzeugen l¨asst sich Geld verdienen, (With
planes you can money earn), where a concrete ob-
ject replaces the whole procedure.
The metonymy could be reconstructed, for sim-
ple cases, by the Generative Lexicon (Pustejovsky
86), whose role is precisely to make explicit pro-
totypical functions of objects via their telic role, as
advocated above.

References
Dowty, D., 1989, On the Semantic Content of the No-
tion of Thematic Role, in G. Cherchia, B. Partee,
R. Turner (eds), Properties, Types and meaning,
Kluwer Academic.
Dowty, D., 1991, Thematic Proto-roles and Argument
Selection, Language, vol. 67-3.
Lakoff G., Johnson M., 1999. Philosophy in the Flesh.
Basic books, NY, USA.
Mari, A., Saint-Dizier, P., 2001, A Conceptual Se-
mantics for Prepositions Denoting Instrumentality,
in proc. 1st workshop on prepositions, Toulouse, and
in Syntax and semantics of prepositions, P. Saint-
Dizier (ed), Kluwer academic, 2006.
Pustejovsky, J., 1991, The Generative Lexicon, Com-
putational Linguistics, vol. 17, MIT Press.
Raina, Achla M., 2002. The Verb Second Phe-
nomenon, O.N. Koul and K Wali (eds.), Topics in
Kashmiri Linguistics. Creative Books, New Delhi,
India.
Rosch, E., 1978. Principles of Categorization. In E.
Rosch and B.B. Lloyd (eds.), Cognition and Catego-
rization. Hillsdale : Lawrence Erlbaum Associates
Publishers.
Sinlapasarn, Upakitt. 1998. Thai Grammar. Thai
Watthana Panich, Bangkok, Thailand.
Talmy L., 2001, 2003. Towards a Cognitive Seman-
tics, vol. 1 and 2. MIT Press.
Wiezbicka, A.,1996. Semantics primes and universals.
Oxford: Oxford University Press.
Wierzbicka, A., 1992, Semantic Primitives and Se-
mantic Fields, in A. Lehrer and E.F. Kittay (eds.),
Frames, Fields and Contrasts. Hillsdale: Lawrence
Erlbaum Associates, pp. 208-227.
