Predicative Lexical Units in Terminology: Marie-Claude L'Homme
Predicative Lexical Units in Terminology: Marie-Claude L'Homme
Predicative Lexical Units in Terminology: Marie-Claude L'Homme
Marie-Claude LHomme
Abstract Predicative lexical units have been largely ignored in terminology for a
number of reasons: one of them is the focus on entities viewed as the nodes of
knowledge representations; another is the lack of linguistic perspective on the data
to be represented. Things are changing though and an increasing number of re-
searchers in the field of terminology and other areas interested in processing spe-
cialized corpora recognize that predicative units (verbs, adjectives and many
nouns) play a key role in the expression and organization of specialized knowl-
edge. However, the models traditionally used in terminology to describe terms are
not equipped to capture the properties of predicative units adequately. In this con-
tribution, I review a selection of works in the area and discuss how they aim at
unveiling the contribution of predicative terms to the expression of specialized
knowledge. I also show how two specific lexical semantics frameworks (Explana-
tory Combinatorial Lexicology, ECL (Meluk et al. 1995; Meluk et al. 1984-
1999) and Frame Semantics (Fillmore 1977, 1982, 1985)) can be applied to the
description of predicative terms and help us represent their linguistic properties. I
will refer to data taken from the specialized fields of cycling, environment and
computing.
1. Introduction
Figure 1 shows what a knowledge structure (in this case, an ontology) can look
like in a specific field: i.e. the bicycle (Theory Simple-Bikes,
http://www.ksl.stanford.edu/htw/dme/thermal-kb-tour/simple-bikes.html).
In this example, items of knowledge (concepts or classes) that represent the parts
of the bicycle are organized in a hierarchy.1 The ontology concentrates on parts
and it can be argued that important concepts are missing, such as types of bicycles.
But we would be dealing with entities in both cases, as do many ontologies or
other types of knowledge representations. I will argue below that other equally
1 Much more can be said about this type of knowledge structure and the way it is defined: at-
tributes can be associated with each class; in some cases, definitions of classes are provided. Ex-
perts must define and agree on criteria according to which items can be organized in specific
fields (for instance, in the bicycle example, other organization principles could have been cho-
sen). Hence a consensus must necessarily be reached for such representations to be usable
(Guarino 2009).
3
important concepts are still missing and that the field of cycling is not only a col-
lection of entities. The example chosen is simple, but still representative of the
kind of reasoning applied when building knowledge structures.
This kind of work is certainly interesting from
the point of view of organizing knowledge in
specialized fields since it can later be used as a
reference for explaining that knowledge and
making conceptual distinctions. However, it
only considers a portion of what is necessary to
convey that knowledge. In the bicycle example,
one can discover the relationship between a
wheel and a bicycle, but cannot learn how
to express the installation of a wheel on a bicy-
cle or what the specific functions of a wheel or a
bicycle are. In addition, the linguistic items used
in the field are considered after concepts are de-
fined and are often viewed as labels that are su-
perimposed on knowledge structures. In the bi-
cycle example, the labeling of concepts is
secondary and some labels do not even corre-
spond to real terms (e.g. Wheel-Assy).2 The
purpose is to represent concepts (linguistic ex-
pressions are listed somewhere else when nec-
essary).
Hence, these methodologies (that I will refer to Fig. 1. A simple ontology of the
as knowledge-driven methodologies) raise a bicycle
number of questions with regard to the way
knowledge is expressed in texts: 3
Often, only nouns are considered. This is a consequence of the focus on
entities. Even in cases where activity concepts need to be taken into ac-
count (linguistically expressed by nouns or verbs), nouns are still pre-
ferred. One can easily think of activities associated with the bicycle (cy-
cling, riding, shifting speeds, etc.) or even properties, but
incorporating them in a structure such as that shown in Figure 1 is not
straightforward.
Even when other types of concepts (processes, events and properties) ap-
pear in knowledge structures, the specific nature of the units that express
2 The linguistic label may be chosen among a list of possible labels used be experts (in some
ontologies, a unique identifier is used to represent an item of knowledge unambiguously);
3 Other linguistic consequences are listed in LHomme (2014).
4
them, namely the fact that they require arguments (e.g., X rides Y, Y cy-
cles, X shifts Y), is not taken into account.
Terminologists resort to less formal methods when defining the contents of spe-
cialized dictionaries. However, they are still guided by a reasoning similar to the
one described above. In a small dictionary on cycling ( vos vlos 2013), one can
find terms linked to the following topics: types of bicycles (city bicycle, electric
bicycle), bicycle parts (back wheel, cable), accessories (dynamo, fender), tools, as
well as repair and maintenance equipment (file, patch), and infrastructure (bike
route, paved shoulder). All entries are devoted to noun terms. A small portion of
these nouns denote activities (e.g., two-way cycling), but they are treated exactly
as the other ones that refer to entities.
Let us know consider matters from a different perspective. Appendix 1 shows the
first results of term extraction (by TermoStat, Drouin 2003) applied to a corpus of
texts on cycling of approx. 150,000 words. Among the 50 most specific items
identified by the term extractor, 8 are verbs (one verb appears in third position)
and 2 are adjectives. Some might argue that some of these units do not correspond
to valid terms in the field of cycling. Although this can be said for some units (and
for some nouns, for that matter), it would be difficult to discard cycle, ride or
brake (especially if their nominal counterparts are already defined as terms). In
addition, 4 nouns refer to activities and can thus be defined as predicative units
(e.g., adjustment, cycling, ride, race). Eight others refer to entities, while still be-
ing truly predicative (e.g., intersection, rider).4 If a knowledge-based analysis such
as that described in the previous section is applied, more than half the items that
could be identified as relevant terms in the field of cycling would either be ne-
glected or described in a way that does not fully capture their semantic properties.
I argue that predicative units are not only necessary to communicate knowledge
they are also an important part of the structure of this knowledge. However, the
structures such as the ontology discussed in section 2 (and other similar ones) are
not well suited to represent them. Other methods need to be (and have been) de-
vised. We mention some of them in the following sections.
4Other items in the list could be considered as predicative units (frame of x, lever of x, etc.),
but we did not consider them in this quick count.
5
A view strongly held a few decades ago was that terms belonging to parts of
speech other than nouns had to correspond to a noun to become relevant for termi-
nology.5
Concepts represented in terminological dictionaries are predominantly expressed in the
form of nouns; concepts which are linguistically expressed as adjectives and verbs in
technical languages are frequently found only in the corresponding noun form and some
theorists deny the existence of adjective and verb concepts (Sager 1990, 58).
This allows us to validate the verbs brake and cycle as terms since they are related
to noun forms. Linguistically though it is difficult to argue that all nouns precede
the other parts of speech: the nouns ride and rider are defined on the basis of the
meaning of the verb ride. This also applies to cycling and race: the meaning of cy-
cling is explained according to that of the verb cycle; the meaning of the noun
race is based on that of the verb race.
Intuitively, it is also difficult to deny the important role played by some predica-
tive units, even those that are not related to noun terms. Corpus data confirms this
as discussed above. We mentioned the problem raised by some verbs in the list of
candidate terms produced after processing a corpus on cycling. Similar examples
can be found in many other fields of knowledge: configure, develop, write (com-
puting); code, replicate, synthetize (genetics); pollute, recycle, warm (environ-
ment). Some adjectives also convey a meaning that it would be difficult not to
consider as specialized: dynamic, static, virtual (computing); clean, green, sus-
tainable (environment).
This was pointed out by a number of scholars (who discussed the importance of
verbs in terminology; other parts of speech have not attracted as much attention):
Condamines (1993), in the field of banking, LHomme (1998) in the field of com-
puting, Lerat (2002), in the field of law, Lorente (2007) in various fields of
knowledge (see LHomme (2012a), for a more detailed account). Three types of
verbs can be found in specialized texts: 1. Specialized verbs (that are specific to a
given field); 2. Verbs that acquire a new meaning in the field; 6 3. General lan-
guage verbs (that can be further divided into support verbs (Lerat 2002) and dis-
cursive verbs (Lorente 2007)). Verbs in groups 1 and 2 would potentially corre-
spond to terms.
5 This view was also shared by Rey (1979). Traces can still be found in more recent work
(e.g., Lorente 2007) for whom a verb can become a quasi-term when it is morphologically re-
lated to a noun term.
6 In some cases, new meanings viewed from the perspective of a special subject field do not
Predicative units are not only defined as terms using criteria that differ from those
that apply to other types of terms, they must also be described in order to fully
capture their meaning. Authors have devised different methods to achieve this, but
they appear to agree on the fact that their argument structure should be repre-
sented, ideally in a way that explains the meaning of the predicative unit when
used in a specialized field. Again, verbs have attracted the most attention, but what
is said in this section also applies to other predicative units.
For example, assuming that ride has two arguments (x, y), they can be represented
using classes of objects (classes dobjets, Lerat 2002), semantic classes defined
specifically for a given subject field (Tellier 2007; Wandji et al. 2013), frame ele-
ments labels (Pimentel 2013), semantic roles (LHomme 2012b).7 In the examples
given in Table 1, typical terms (i.e., terms that are considered to be representative
7 The definition of argument structures (i.e. the number of arguments; the determination of
arguments vs. adjuncts; the nature of arguments) can differ substantially according to the theoret-
ical perspective taken. Often in terminology, arguments are defined at an abstract and semantic
level (i.e. those participants that are necessary to understand the meaning of the predicative unit).
This differs from perspectives based on syntactic evidence. Of course, as will be shown in this
paper, there is a relation between semantics and syntax. However, at the semantic level, argu-
ments may be considered obligatory and be realized in specific cases in syntax (this is the case
for nominalization of verbs, for instance).
7
of what can be expected in a given argument position) are used. 8 Other examples
are given for terms in the subject fields of computing and the environment.
Assuming like I did that predicative units should be considered as valid terms and
that their argument structures should be considered as important parts of their de-
scriptions, some principles based on lexical semantics can guide us when adding
these units to lexicons. In what follows, I show how this can be carried out using
two different lexical frameworks, namely Explanatory Combinatorial Lexicology,
ECL (Meluk et al. 1995; Meluk et al. 1984-1999) and Frame Semantics (Fill-
more 1976, 1982, 2985; Fillmore et al. 2003). In addition to a few aforementioned
examples in the field of cycling, I will refer to two specialized lexical resources:
one is dedicated to the fields of computing and the Internet (DiCoInfo. Dictionnai-
re fondamental de linformatique et de lInternet) and the other to the environment
(DiCoEnviro. Dictionnaire fondamental de lenvironnement). (A short description
of these resources is given in Section 4.1.)
DiCoInfo and DiCoEnviro are freely available online resources that contain terms
in English, French and Spanish (the coverage, however, differs from one language
8 In this kind of approach, authors implicitly assume that predicative units can be defined re-
gardless of their possible uses in other fields or situations. For instance, the arguments defined
for ride are valid from the point of view of cycling but not for ride in general. The same applies
to impact (verb and noun) and effect.
8
to the other). The resources describe terms that belong to different parts of speech
such as nouns (configuration, data, environment, biodiversity), verbs (to config-
ure, to download, to pollute, to warm), adjectives (configurable, environmental,
sustainable, virtual) and adverbs (e.g., dynamically, environmentally), part of
which are predicative units.
The theoretical and methodological principles used to compile these resources are
based chiefly on ECL. (However, as will be seen below, other principles were bor-
rowed from Frame Semantics.) They adhere to ECL principles in the following
ways:
9In the DiCoInfo, only specialized meanings are considered. Hence, even if the corpus does
contain occurrences of general meanings, only those related to computing or the Internet are in-
cluded in the word list of the DicoInfo and those meanings linked to the environment appear in
the DiCoEnviro.
10 This is beyond the scope of the article, but all lexical relationships are described using the
formal system of lexical functions (a system devised within ECL) that provides a formal basis to
specify the structure and meaning of relationships (paradigmatic as well as syntagmatic) (More
information about the application of ECL principles to the specialized resources can be found in
LHomme 2012b).
9
gument structure states the obligatory participants, and labels them in terms of
general semantic roles (Agent, Patient, Instrument11) and typical terms (the term
that is the most representative of those that instantiate a given semantic role in
corpora).
These systems (and especially semantic roles) enable us to capture the relation-
ships between semantically related predicative units.12 Table 2 gives a few exam-
ples of relationships that can be captured using this system.
11 The role Instrument as an argument has been questioned by a number of authors. Howev-
er, in specialized subject fields, it can be an obligatory participant. In addition, it often appears in
regular alternations (e.g. the user prints a document with a laser printer; the printer prints the
file).
12 In fact, the system is also extended to quasi-predicative units, but we do not deal with this
in this contribution.
10
When you[Agent] INSTALL a game[Patient], it generally also copies a huge array of samples on to
your hard disk.
You[Agent] can even run Linux directly from a CD, without having to INSTALL anything[Patient]
on your PC[Destination].
The operating system[Patient] IS INSTALLED on your hard disk[Destination].
13 This is the terminology used in ECL and in our resources. Frame Semantics refers to core
frame elements and non-core frame elements. Although the examples do not show this, circum-
stants (i.e. non-obligatory participants) are also annotated.
11
It so happens that the IMPACT of climate change[Cause] on world heritage sites[Patient] will be the
subject of a meeting taking place at UNESCO just two months from now, on 16 and 17 March.
Theories concerning the climatic[Patient] IMPACT of such emissions[Cause] vary.
In general, these scientific workshops focused on the drivers of global change and the IMPACT
of those changes[Cause] on ecosystems[Patient],
Attach, vt.
Agent{user} ~ Patient{file} to Destination{email}
Lexical relationships:
A file[Patient] that has been a. attached
Annotations
You[Agent] can also ATTACH files[Patient] to email messages[Destination]
Non-ASCII files, known as binary files[Patient], may BE ATTACHED to e-mail
messages[Destination]
You[Agent] can ATTACH more than one file[Patient] to a single email message[Destination] ..
14
The graph is generated with a tool developed by Robichaud (2012). In the original graph,
arcs are colored.
13
Fig. 2. Lexical relationships between pollute and other terms in the field of the environment
15 Our methodology differs from that applied in FrameNet in the sense that frames are dis-
covered afterwards based on the descriptions already available in our databases. In the FrameNet
project, frames are defined and then validated on corpus data (Fillmore et al. 2003).
14
Objective_influence
Note: A Frame based on Objective_influence in FrameNet
Definition:
An Agent or a Cause has an influence (that is often negative) on a Patient. The Patient can
undergo changes due to this influence.
Participants (1):
1. Agent | Cause
2. Patient
Participants (2):
Degree, Duration, Expanse, Location, Manner, Means, Method, Mode, Reason, Time,
Value
Lexical Units:
English: French:
affect 1 affecter 1
effect 1 effet 1
impact 1 impact 1
impact 2 influence 1
influence 1 influencer 1
influence 2 influer 1
Fig. 3. A frame with lexical units evoking it in the field of the environment
15
5 Concluding remarks
In this contribution, I emphasized the importance of the notion of structure in
terminology and the role it plays in knowledge understanding. Capturing and rep-
resenting concepts is an ontology or other types of knowledge repositories is or
should be an important part of terminology work. Unfortunately, most efforts car-
ried out up to now have focused on representations that allow us to take into ac-
count entity concepts and, indirectly, terms that refer to these entities. However, in
order to convey knowledge in text, other types of units are necessary as evidenced
by the small term extraction on a corpus of cycling.
I argued and tried to show that predicative units are an essential part in the con-
veyance of knowledge and that they must appear in terminological descriptions. I
also underlined the fact that, even though they are important, they cannot be prop-
erly described using the tools devised for terms that refer to entities. Their argu-
ment structure must be stated in such a way as to capture their specialized nature
(for instance, by specifying the type of unit that would instantiate an argument).
Finally, I showed that predicative terms can also participate in two different kinds
of structures. The first one is based on the relationships terms share with other
terms. This can be achieved if the links between terms have been properly identi-
fied and distinguished from one another. The second one is based on the notion of
frame; terms can be said to evoke frames and a first subset of data seems to con-
firm this. There might be cognitive implications in the ideas reported in this chap-
ter, but I will let experts such as Michael Zock dwell on them.
One question, however, remains unanswered. Would it be possible to connect
knowledge structures (at least ones that also represent terms) to descriptions of
predicative units such as those that we proposed in this contribution? Intuitively,
we can certainly think of some kind of connection. As shown in Figure 4, the ar-
gument structure stated for ride in which bicycle appears could be linked to a
knowledge representation in which types of bicycles are listed (the classes for
bicycle are based on those given in Wikipedia 2013). From this, we could infer
that we can ride hybrid bicycles, upright bicycles, touring bicycles, etc.
16
References
Fillmore, C.J., M.R.L. Petruck, J. Ruppenhofer & A. Wright. 2003. FrameNet in Action: The
case of attaching. International Journal of Lexicography 16(3): 297-332.
FrameNet 2014. FrameNet (https://framenet.icsi.berkeley.edu/fndrupal/). Accessed January 24
January 2014.
Guarino, N., D. Oberle and S. Staab. 2009. What is an Ontology? , In Staab, S. and R. Studer
(eds). Handbook on Ontologies, 11-17, Berlin: Springer.
LHomme, M.C. 1998. Le statut du verbe en langue de spcialit et sa description lexicographi-
que. Cahiers de lexicologie 73(2): 61-84.
L'Homme, M.C. 2012a. Le verbe terminologique : un portrait de travaux rcents. In Neveu, F.
et al. (d). Actes du 3e Congrs mondial de linguistique franaise, Lyon, France, EDP Scien-
ces.
LHomme, M.C. 2012b. Adding Syntactico-semantic Information to Specialized Dictionaries:
An Application of the FrameNet Methodology. In Gouws, R. et al. (eds.). Lexicographica 28:
233-252.
LHomme, M.C. 2014. Terminologies and Taxonomies. In Taylor. J. (ed.). The Handbook of
the Word, Oxford: Oxford University Press.
LHomme, M.C. (dir.) 2014. DiCoEnviro. Dictionnaire fondamental de lenvironnement
(http://olst.ling.umontreal.ca/cgi-bin/dicoenviro/search-enviro.cgi). Accessed 6 January 2014.
LHomme, M.C. (dir.) 2014. DiCoInfo. Dictionnaire fondamental de linformatique et de
lInternet (http://olst.ling.umontreal.ca/cgi-bin/dicoinfo/search.cgi). Accessed 6 January 2014
LHomme, M.C., B. Robichaud and C. Subirats. 2014. Discovering Frames in Specialized Do-
mains, Language Resources and Evaluation, LREC 2014, Reykjavik, Iceland.
Lerat, P. 2002. Quest-ce que le verbe spcialis ? Le cas du droit. Cahiers de Lexicologie 80:
201-211.
List of bicycle types. In Wikipedia (http://en.wikipedia.org/wiki/List_of_bicycle_types). Acces-
sed 13 December 2013.
Lorente, M. 2007. Les unitats lxiques verbals dels textos especialitzats. Redefinici d'una pro-
posta de classificaci. Lorente, M. et al. (ed.) Estudis de lingstics i de lingstica aplicada
en honor de M. Teresa Cabr Catellv. Volum II: De deixebles. 365-380, Barcelona: Institut
Universitari de Lingstica Aplicada de la Universitat Pompeu Fabra, Srie Monografies 11-
12.
Meluk, I. et al. 1984-1999. Dictionnaire explicatif et combinatoire du franais contemporain.
Recherches lexico-smantiques I-IV, Montral: Presses de lUniversit de Montral.
Meluk, I., A. Clas, and A. Polgure. 1995. Introduction la lexicologie explicative et combina-
toire, Louvain-la-Neuve: Duculot / Aupelf-UREF.
Pimentel, J. 2013. Methodological Bases for Assigning Terminological Equivalents. A Contri-
bution. Terminology 19(2): 237-257.
PropBank 2014. PropBank (http://verbs.colorado.edu/propbank/). Accessed 11 April 2014.
Rey, A. 1979. La terminologie : noms et notions, Paris : Presses universitaires de France.
Robichaud. B. 2012. Logic Based Methods for Terminological Assessment, In Language Re-
sources and Evaluation (LREC 2012). Istanbul, Turkey.
Ruppenhofer, J., M. Ellsworth, M. R. L. Petruck, C. Johnson, and J. Scheffczyk. 2010. FrameNet
II: Extended Theory and Practice, http://framenet.icsi.berkeley.edu/. Accessed 12 November
2013.
Sager, J.C. 1990. A Practical Course in Terminology, Amsterdam/Philadelphia : John Benja-
mins.
Tellier, C. 2008. Verbes spcialiss en corpus mdical : une mthode de description pour la r-
daction darticles terminographiques, Travail dirig prsent au Dpartement de linguistique
et de traduction, Universit de Montral.
Theory Simple Bikes, http://www.ksl.stanford.edu/htw/dme/thermal-kb-tour/simple-bikes.html.
Accessed 11 December 2013.
VerbNet 2014. VerbNet (http://verbs.colorado.edu/~mpalmer/projects/verbnet/downloads.html).
Accessed 11 April 2014.
18
Wandji, O., LHomme, M.-C. and Grabar, N. 2013. Discovering Semantic Frames for a Contras-
tive Study of Verbs in Medical Corpora. In Actes de la 9e confrence internationale Termino-
logy and Artificial Intelligence (TIA13), Paris, France.
Score Part of
Lemma Frequency Inflected forms
(Specificity) Speech
Brake, brakes,
brake 95 57.18 Verb
Braking
Adjust, adjusted,
adjust 197 56.53 Verb
Adjusting
bolt 107 55.11 Bolt, bolts Noun
reflector 78 54.63 Reflector, reflectors Noun
helmet 128 54.08 Helmet, helmets Noun
Pedal, pedals,
pedal 80 53.66 Verb
pedaled, pedaling
Pedestrian,
pedestrian 94 52.79 Noun
pedestrians