=Paper=
{{Paper
|id=None
|storemode=property
|title=Combining Formal Concept Analysis and Translation to Assign Frames and Thematic Role Sets to French Verbs
|pdfUrl=https://ceur-ws.org/Vol-959/paper15.pdf
|volume=Vol-959
|dblpUrl=https://dblp.org/rec/conf/cla/FalkG11
}}
==Combining Formal Concept Analysis and Translation to Assign Frames and Thematic Role Sets to French Verbs==
https://ceur-ws.org/Vol-959/paper15.pdf
Combining Formal Concept Analysis and
Translation to Assign Frames and Thematic
Role Sets to French Verbs
Ingrid Falk1 , Claire Gardent2
1
INRIA/Nancy Universités, Nancy (France)
2
CNRS/LORIA, Nancy (France)
Abstract. We present an application of Formal Concept Analysis in the
domain of Natural Language Processing: We give a general overview of
the framework, describe its goals, the data it is based on, the way it works
and we illustrate the kind of data we expect as a result. More specifically,
we examine the ability of the stability, separation and probability indices
to select the most relevant concepts with respect to our FCA application.
We show that the sum of stability and separation gives results close to
those obtained when using the entire lattice.
1 Introduction
Ideally natural language processing (NLP) applications need to analyse texts to
answer the question of “Who did What to Whom”. For computers to effectively
extract this information from texts, it is essential that they be able to detect
the events that are being described and the event participants. Because events
are mostly lexicalised using verbs, one ingredient that is essential for such sys-
tems is detailed knowledge about their syntactic and semantic behaviour. It has
been shown (Briscoe and Carroll (1993), Carroll and Fang (2004)) that detailed
subcategorisation information (that is, information about the number and the
syntactic type of verb complements) is crucial in enhancing their linguistic cover-
age and theoretical accuracy. However this syntactic information is not sufficient
to specify “Who did what to Whom” because it does not allow to identify the
thematic roles participating in the event described by the verb. For example in
John threw a ball to Mary the syntactic analysis of the sentence would not allow
to identify John which is the syntactic subject of the sentence as the Agent or
Causer of the throwing event, Mary, syntactically the prepositional object as the
Destination and ball (the object) as the item being thrown.
To help computer systems in this task of understanding and representing the
full meaning of a text, verb classifications have been proposed which group to-
gether verbs with similar syntactic and semantic behaviour, ie. which associate
groups of verbs with subcategorisation frames showing the syntactic construc-
tions the verbs may appear in and sets of thematic roles which represent the
participants in an event described by the verbs in the group.
For English, there exist several large scale resources providing verb classes
(eg. Framenet Baker et al. (1998) and VerbNet Schuler (2006), the classifica-
tion we use in our framework) in a format that is amenable for use by natural
language processing systems. For example for the verb throw the corresponding
VerbNet class shows that the participants in a throwing event are an Agent, a
Theme (the thing being thrown), a Source and a Destination. In addition, the
VerbNet class provides the syntactic constructions the verb can occur in (eg.
Subject(John) V(throws) Object(a ball ) PrepObject(to Mary)) and shows
how the participant roles can be realised as syntactic arguments: In the exam-
ple above the Agent (John) is realised syntactically as Subject, the Theme
(the ball ) as Object and the Destination (to Mary) as prepositional object
(PrepObject).
For French however, existing verb classes are either too restricted in scope
(Volem Saint-Dizier (1999)) or not sufficiently structured (the LADL tables
Gross (1975)) to be directly useful for NLP. Even though recently other large cov-
erage syntactic-semantic resources for French have been made available (Tolone
(2011) as well as further processed versions of Dubois and Dubois-Charlier
(1997), Hadouche and Lapalme (2010)) the terminology and linguistic formalisms
they are based on is often still hardly compatible with the methods and tools
currently used in the NLP community.
In this paper we present a method for providing a VerbNet style classifica-
tion of French verbs which associates verbs with syntactic constructions on the
one hand and sets of semantic role sets (the set of semantic roles participating
in the event described by the verb) on the other. To obtain this classification,
we build and combine two independent classifications. The first is semantic and
is obtained from the English VerbNet (VN) by translation, the second is syn-
tactic and is obtained by building an FCA (Formal Concept Analysis) lattice
from three, manually validated syntactic lexicons for French. The first asso-
ciates groups of French verbs with the semantic roles of the English VN class.
The second associates groups of French verbs (the concept extent) with syntactic
constructions (concept intent). We then merge both classifications by associating
with each translated VN class, the FCA concept whose verb set yields the best
F-measure with respect to the verb sets contained in each translated VN class.
We thus effectively associate the set of semantic roles of the VN class to the
group of French verbs and the syntactic information given by the FCA concept.
In the past several linguistic FCA applications have been presented, as Priss
(2005) shows in her overview. For example, Sporleder (2002) describes an FCA
based approach to build structured class hierarchies starting from unstructured
lexicon entries while the features used for building classes in the approach pre-
sented in (Cimiano et al., 2003) are collected from a corpus. Our approach (based
on earlier work presented in Falk et al. (2010), Falk and Gardent (2010)) is con-
cerned with building a lexical resource based on lexicons and is therefore related
to the FCA approach in (Sporleder, 2002). However, the features we use are
different. In addition we explore the use of concept selection indices to filter the
concept lattices and finally relate the formal concepts we obtain to other classes
obtained by a clustering approach based on different numeric features extracted
from lexicons and English-French dictionaries.
In the following we first introduce the terminology and data used in our
application domain. Next we describe how we associate groups of French verbs
with syntactic information using Formal Concept Analysis (Section 3). As the
resulting concept lattice has a very large number of concepts which are mostly
not useful verb classes we explore methods to select the concepts most relevant
to our application (Section 4). We show in particular that selecting only ∼ 10%
of the concepts of the lattice using indices proposed in Klimushkin et al. (2010)
gives results close to those obtained when using the entire lattice. We then show
how we build the translated VerbNet classes and how they are mapped to the
previously pre-selected FCA concepts (Section 5). Finally in Section 6 we present
the kind of associations we obtain by our method.
2 Linguistic Concepts and Resources
Our aim is to build a lexicon associating groups of French verbs with:
1) the syntactic constructions the verbs of this group may appear in,
2) the semantic roles participating in an event described by a verb of this group.
Syntactic constructions a verb may occur in are described using subcategorisation
frames (SCF) and are usually part of a lexical entry describing the verb. A
subcategorisation frame (SCF) characterises the number and the type of the
syntactic arguments expected by a verb. Each frame describes a set of syntactic
arguments and each argument is characterised by a grammatical function (eg.
SUJ - subject, OBJ - direct object etc.) and a syntactic category (NP indicates
a noun phrase, PP a prepositional phrase, etc.). For example John throws a
ball to Mary. is a possible realisation of the subcategorisation frame SUJ:NP V
OBJ:NP POBJ:PP.
The semantic (thematic) roles are the participants in an event described by
a particular verb. To date there is no consensus about a set of semantic roles
or a set of tests determining them. There may be a general agreement on a
set of Semantic Roles (eg. Agent, Patient, Theme, Instrument, Location, etc.)
but there is substantial disagreement on when and where they can be assigned
(Palmer et al., 2010). Thus each of the well known resources (FrameNet (Baker
et al., 1998), PropBank (Palmer et al., 2005), VerbNet (Schuler, 2006), LVF
(Dubois and Dubois-Charlier, 1997)) providing semantic role information have
their own semantic role inventory. In our work we chose the VerbNet semantic
role inventory for several reasons:
1. VN semantic roles provide a compromise between generalisation and speci-
ficity in that they are common across all verbs3 but are still able to capture
specificities of particular classes.
3
in contrast to FrameNet Baker et al. (1998) and PropBankPalmer et al. (2005) roles.
2. VN roles are among those generally agreed upon in the community.
3. None of the other resources provide the link between syntactic arguments
and semantic roles across different verbs.
4. Semantic roles are expected to be valid across languages and by using the
same role inventory as for English we hope to leverage some of the substantial
research done for English and link syntactic information for French with
semantic information provided by the English classes. Our method allows
us to detect groups of French verbs with the same role set as some English
VerbNet class and gives information about how these semantic roles are
realised syntactically in French.
Figure 1 shows an excerpt of the throw-17.1 VerbNet class, with its verbs,
thematic roles and subcategorisation frames.
verbs (32): kick, launch, throw, tip, toss, ...
sem. roles: Agent, Theme, Source, Destination
SCFs sem. roles
Subject V Object Agent V Theme
John throws a ball
Subject V Object PrepObject Agent V Theme Destination
frames (8):
John throws a ball to Mary
Subject V Object Object Agent V Destination Theme
John throws Mary a ball
etc.
Fig. 1: Simplified VerbNet class throw-17.1.
Thus, from this data an English NLP system analysing the sentence John threw
a ball to Mary could infer the semantic roles involved in the event, namely
those given by the VerbNet class. It could also detect the possible semantic
roles realised by the syntactic arguments: It would know that the subject is a
realisation of the Agent semantic role, the object of the Theme or Destination
semantic roles, etc.
3 Associating French Verbs with Subcategorisation
Frames
To associate French verbs with syntactic frames, we use the FCA classification
approach where the objects are verbs and the attributes are the subcategorisa-
tion frames associated with these verbs by the subcategorisation lexicon to be
described below.
3.1 Subcategorisation Lexicons
Subcategorisation information is retrieved from three existing lexicons for French:
Dicovalence van den Eynde and Mertens (2003), the LADL tables Gross (1975),
Guillet and Leclère (1992) and finally TreeLex Kupść and Abeillé (2008). Each
of these was constructed manually or with an important manual validation by
linguists. The combined lexicon covers 5918 verbs, 345 SCFs and has a total of
20443 hverb, framei pairs. Table 1 shows sample entries in this lexicon for the
verb expédier (send). Using the Galicia Lattice Builder software4 , we first build
Verb: expédier
SCF Source info
SUJ:NP,DUMMY:REFL DV:41640,41650
SUJ:NP,OBJ:NP DV:41640,41650;TL
SUJ:NP,OBJ:NP,AOBJ:PP TL
SUJ:NP,OBJ:NP,POBJ:PP,POBJ:PP LA:38L
Table 1: Sample entries in subcategorisation lexicon for verb expédier (send).
a concept lattice based on the formal context hV, F, Ri such that:
– V is the set of verbs in our subcategorisation lexicon. We ignore verbs with
only one SCF as they will result in classes associating verbs with a unique
frame.
– F is the set of subcategorisation frames (SCFs) present in the subcategori-
sation lexicon,
– R is the mapping such that (v, f ) ∈ R iff the subcategorisation lexicon
associates the verb v with the SCF f .
The resulting formal context is made of 2091 objects (verbs) and 238 attributes
(frames), giving rise to a lattice of 12802 concepts. Clearly however not all these
concepts are interesting verb classes. Classes aim to factorise information and
express generalisations about verbs. Hence, concepts with few (1 or 2) verbs
can hardly be viewed as classes and similarly, concepts with few frames are less
interesting.
To select from this lattice those concepts which are most likely to provide
the most relevant verb-frame associations, we explore the use of three indices for
concept selection: concept stability, separation and probability which have been
proposed and analysed in (Klimushkin et al., 2010). In Section 4.2 we investigate
which of these indices performs best in the context of our application. We then
use the best performing concept filtering method to select the most relevant
concepts with respect to our data. For each translated VN class we then identify
among the selected FCA concepts the one(s) with best f-measure between preci-
sion and recall. For a translated VN class CV N (consisting of French verbs) and
the extent (verb set) of an FCA concept CF CA precision, recall and f-measure
|CV N ∩ CF CA | |CV N ∩ CF CA | 2RP
are computed as follows: R = ,P = ,F =
|CV N | |CF CA | R+P
The translated VN class is then associated with the FCA concept(s) with best
F-measure. Thus the verbs in the FCA concept are effectively associated with
the thematic roles of the translated class and at the same time with the syntactic
subcategorisation frames in the intent (attribute set) of the FCA concept.
4
http://www.iro.umontreal.ca/~galicia/
4 Filtering Concept Lattices
The lattices we have to deal with are very large and many of the concepts do not
represent valid verb classes. To select those concepts which are most relevant in
the context of our application the concept lattice needs to be filtered. Klimushkin
et al. (2010) propose three indices for selecting relevant concepts in concept
lattices built from noisy data: concept stability, separation and probability. In
this section, we investigate which of these indices works best for our data.
Concept stability is a measure which helps discriminating potentially interesting
patterns from irrelevant information in a concept lattice based on possibly noisy
data. The stability of a concept C = (V, F ) is the proportion of subsets of the
extent V which have the same attribute set F as V :
|{A ⊆ V | A0 = F }| 5
σ((V, F )) = . (1)
2|V |
Intuitively, a more stable concept is less dependant on any individual object in
its extent and is therefore more resistant to outliers or other noisy data items.
Concept separation indicates the significance of the difference between the ob-
jects covered by a given concept from other objects and, simultaneously, between
its attributes and other attributes:
|V | |F |
s((V, F )) = P . (2)
|{v}0 | + 0
P
v∈V f ∈F |{f } | − |V | |F |
Intuitively we expect a concept with high separation index to better sort out the
verbs it covers from other verbs and simultaneously the frames it covers from
other frames. Whereas concept stability is a measure concerned with either ob-
jects or attributes, separation gives information about objects and attributes at
the same time.
Concept probability. For an attribute a ∈ A, the attribute set, we denote by pa
the probability of an object to have the attribute a. In practise it is the propor-
0
tion of objects having a: pa = |{a} |
|O| , where O denotes the set of objects.
For B ⊆ A, we define pBYas the probability of an arbitrary object having all
attributes from B: pB = pa . This formulation assumes the mutual indepen-
a∈B
dence of attributes. Based on this, and denoting n = |O| we obtain the following
formula for the probability of B being closed:
n
X
p(B = B 00 ) = p(|B 0 | = k, B = B 00 ) (3)
k=0
n
" #
X n k Y
= p (1 − pB )n−k (1 − pka ) (4)
k B
k=0 a∈B
/
5
Here and in the following 0 represents the operator on the power sets of objects:
0
: 2O → 2A , X 0 = {a ∈ A | ∀o ∈ X. (o, a) ∈ R} and dually on that of attributes.
A small p(B = B 00 ) suggests a small probability of the attribute combination B
to be a concept intent by chance only (and p(B = B 00 ) ≈ 1 that there is a high
probability that the combination is a concept intent by chance). However, this
reasoning is based on the independence of the attributes, which in our particular
case can not be warranted.
4.1 Computing Stability, Separation and Probability Indices.
Stability. Calculating stability is known to be NP-complete (Kuznetsov, 2007),
however Jay et al. (2008) show that when the concept lattice is known it can
be computed efficiently by a bottom-up traversal algorithm introduced in (Roth
et al., 2006). This is the algorithm we used to compute concept stability.
Separation can be computed in O(|O| + |A|) time, where O and A are the object
and attribute sets respectively. Computing separation is the least prohibitive of
the three indices.
Probability. Klimushkin et al. (2010) show that computing probability of only one
concept involves O(|O|2 ·|A|) multiplication operations which is computationally
very costly. With the computational means at our disposal it was not possible for
us to compute the concept probabilities. We therefore computed approximations
derived as follows: Y
First, we consider (1 − pka ) ≈ 1 for k > 40. In view of this, Equation (4)
a∈B
becomes:
40
" #
00
X n k n−k
Y
k
p(B = B ) = p (1 − pB ) (1 − pa ) (5)
k B
k=0 a∈B
/
n
X n k n−k
+ p (1 − pB ) (6)
k B
k=41
n
X n
As pk (1 − p)n−k = 1, Term (6) can be rewritten as:
k
k=0
40
X n
1− pkB (1 − pB )n−k = (7)
k
k=0
1 − F (40; n, pB ). (8)
k
X n i
F (k; n, p) = p (1 − p)n−1 is the cumulative distribution function of the
i=0
i
binomial distribution6 and can be computed using various statistical software
packages. Term (5) can also be computed more easily considering that nk pkB (1−
pB )n−k are binomial densities the computation of which is also provided by
statistics software7
6
Source Wikipedia: http://en.wikipedia.org/wiki/Binomial_distribution
7
We used the R software environment for statistical computing (http://www.
r-project.org/).
4.2 Evaluating the Concept Selection Indices
In the following we measure the performance of the three concept selection in-
dices with respect to our data. The experimental setting is as follows:
We first select a number of N (1500) concepts with best selection index. The
selected concepts are aligned with the classes translated from VerbNet (see Sec-
tion 5): For each translated class, we select the concept with best precision/recall
f-measure. Then we associate to the concept with best f-measure the thematic
roles of the translated VN class. Next we compare the obtained hverb, thematic
role seti associations with those given by a reference. As for our task recall is
more important than precision, we use the F 2 measure, which gives more weight
to recall, for comparison.
As reference we use the data used for training the classifier for learning
the translated VN classes (see Section 5): we are checking which index selects
the most relevant concepts, that is those best matching the translated classes.
The reference consists of the hverb, semantic role seti pairs marked as positive
examples in the training set, ie. those for which we considered that the French
verbs could have the semantic roles given by the English VN class. Table 2 shows
cov. prec. rec. F2
stab only 39.88 18.96 32.55 26.27
sep only 34.25 28.37 21.52 23.41
prob only 35.53 26.60 20.73 22.38
w/o filtering 100 12.30 60.96 26.30
Table 2: F2 scores and coverage for stability, separation and the 6th probability 10-
quantile.
the F2 scores and coverage when using only one index at a time. For stability
and separation we applied the method above on the top ranking 1500 concepts.
Regarding probability, at first sight, we should consider best the concepts with
lowest probability – because the probability of their intents of being closed by
chance only is accordingly low. However, looking at the data we found that these
concepts have very few verbs and large intent (frame) sets - which rather suggest
improbable or rare verb groups. On the other hand, the interpretation of concept
probability suggests that a concept with a probability close to 1 could occur by
chance only. For these reasons, to assess probability separately we settled on the
6th 10 quantile. The results confirm the observations of Klimushkin et al. (2010):
stability alone gives F2 scores close to an upper bound – the results obtained
without filtering, ie. aligning the translated classes with all the concepts of the
lattice. The results for separation and probability are several points lower.
As we only select ∼ 10% of the total number of concepts we also have to
make sure that the selected concepts cover at least a reasonable amount of
verbs. The cov column gives the percentage of verbs in the lattice covered by the
selected concepts. It shows that using only one index at a time the pre-selected
concepts would contain only 35% − 40% of the verbs in the entire lattice, which
is unsatisfactory.
Klimushkin et al. (2010) investigate the performance of the stability, sepa-
ration and probability indices at finding the original concepts in lattices pro-
duced from contexts which were previously altered by introducing two types
of noise: Type I noise is obtained by altering every cell in the context with
some probability, Type II noise is obtained by adding a given number or pro-
portion of random objects or attributes. According to this, our contexts are
affected by Type I noise rather than Type II. Klimushkin et al. (2010) found
that stability was most effective at sorting out Type II noise, but also proved
helpful in the case of Type I noise. In contrast, they suggest that separation
and probability can not be used on their own but should rather serve as a nor-
malising measure for stability. The most promising combination seemed to be:
stability + ksep · separation − kprob · probability.
In the following we start from the assumption that the most effective index
for selecting relevant concepts is given by a linear combination of stability, sepa-
ration and probability: kstab · stability + ksep · separation − kprob · probability, and
empirically determine the coefficients kstab , ksep and kprob such that the selected
concepts perform best with respect to our task.
We proceed as follows: We choose kstab , ksep and kprob . We then compute the
corresponding linear combination for the concepts and select the 1500 concepts
ranking highest. As in the previous experiments, we measure the relevance of the
selected concepts by aligning the concepts with the translated VN classes and
by comparing the alignments with the same reference as before. We consider the
“best” kstab , ksep , kprob combination the one giving highest F2 scores and good
coverage.
Table 3a shows the results for a first series of experiments where kstab and
ksep were assigned the values 0.5 and 1 and kprob 0.25 and 0.5 (The lines are
sorted by decreasing F2 score). They suggest that the stability and separation
coefficients had less impact on coverage and F2 score than the probability coef-
ficient. Interestingly the coverage is correlated with the F2 score.
In the second series of experiments, shown in Table 3b, we kept the stability
and separation coefficients fixed and varied only the probability coefficient. These
results suggest that the probability coefficient may not help at selecting the
most relevant concepts in our setting. This may be due first to the fact that our
attributes are not independent (we assumed independence of attributes when
setting up the formula for computing the probability index) and second to the
fact that we had to approximate the probability index and this approximation
may not be accurate enough.
In the next series of experiments we investigated the impact of the number
of preselected concepts (500). The results showed that with this smaller num-
ber of concepts the selected concepts reached a slightly smaller F2 score but a
substantially lower coverage. Also, in this configuration the probability index
did seem to be helpful. Preselecting 1000 concepts confirmed the previously ob-
served tendencies: The F2 score and coverage were only slightly lower than when
preselecting 1500 concepts and again the probability index seemed to have only
a small impact on the overall results.
(a) F2 and coverage when kstab , ksep ∈(b) F2 and coverage when kstab and ksep
{0.5, 1}, kprob ∈ {0.25, 0.5}. are kept fixed and kprob varies.
kstab ksep kprob cov. prec. rec. F2 kstab ksep kprob cov. prec. rec. F2
1 1 0.25 98.04 11.87 55.19 24.89 1 1 0 98.04 12.05 55.12 25.16
1 0.5 0.25 98.04 11.87 55.19 24.89 1 1 0.05 98.04 12.05 55.12 25.16
1 0.5 0.5 57.69 17.08 30.18 24.04 1 1 0.005 98.04 12.05 55.12 25.16
1 1 0.5 56.15 17.45 29.13 23.82 1 1 0.0005 98.04 12.05 55.12 25.16
0.5 0.5 0.25 56.15 17.45 29.13 23.82 1 1 0.1 98.00 11.91 55.38 25.00
0.5 1 0.25 53.81 18.03 27.82 23.36 1 1 0.2 98.08 11.88 55.12 24.91
0.5 0.5 0.5 49.72 18.55 26.25 23.06 1 1 0.25 98.04 11.87 55.12 24.89
0.5 1 0.5 49.90 18.61 25.98 22.95 1 1 0.3 98.00 11.79 55.38 24.80
1 1 0.4 59.95 16.27 31.23 23.91
1 1 0.5 56.16 17.45 29.13 23.82
w/o filtering 100 12.30 60.96 26.30
Table 3: F2 scores and coverage for various kstab , ksep , kprob combinations.
From these experiments we conclude the following: First they suggest that
the best linear combination is the sum of the stability and separation indices
as the F2 measure and the coverage for this combination are similar to those of
an upper bound, ie. the alignment obtained without filtering. They show that
selecting only ∼ 10% of the original lattice gives a verb, frame, semantic role set
alignment which is close to the alignment obtained when using the entire lattice
and that the pre-selected concepts also have a similar coverage.
Second, it does not seem evident that probability has a positive effect on
the selected concepts. However, it does improve f-measure when the number of
selected concepts is lower (500 or 1000 vs. 1500 in our experiments). Hence, for
our application we concluded that it is a better strategy to select a larger number
of concepts (1500) and not take probability into account. This is even more so as
the probability index in our case should be taken with caution because first we
had to use an approximation to compute it which may be too rough, and second
the computation of probability is based on the independence of attributes which
is not warranted in our case.
5 Associating French Verbs with Thematic Role Sets.
We associate French verbs with thematic role sets by translating the English
VerbNet classes to French using 3 English-French dictionaries. In the following
we first briefly describe the relevant resources, ie. VerbNet and the dictionaries
before giving the translation methodology. As for this paper only the translated
classes, but not the method to produce them is relevant8 we only very briefly
sketch the methodology.
8
Of course better translated classes will result in a better performance of our method,
but it is not straight forward to evaluate the quality of the translated classes.
VerbNet (Schuler (2006)) is the largest electronic verb classification for En-
glish. It was created manually and classifies 3626 verbs using 411 classes. Each
VN class includes among other things a set of verbs, a set of subcategorisation
frames and a set of thematic roles. Figure 2 shows an excerpt of the amuse-31.1
class, with its verbs, thematic roles and subcategorisation frames.
verbs (242): abash, affect, afflict, amuse, annoy, . . .
thematic roles: Experiencer, Cause
NP V NP Experiencer V Cause
NP V ADV-Middle Experiencer V Adv
frames (6):
NP V NP-PRO-ARB Cause V
...
Fig. 2: Simplified VerbNet class amuse-31.1.
English-French dictionaries. We use the following resources to translate the verbs
in the English VN classes to French: Sci-Fran-Euradic, a French-English bilingual
dictionary, built and improved by linguists , Google dictionary9 and Dicovalence
van den Eynde and Mertens (2003)10 . The merged dictionary contains 51242
French-English verb pairs.
In the following we describe our method for translating the English VerbNet
classes to French.
The translation of VerbNet classes is bound to be very noisy because verbs are
polysemous and the dictionaries typically give translations for several readings
of the verb: Thus the dictionary may give several translations vf r which do not
correspond to the meaning given by the hven , classi pair or this meaning may
even not be covered at all by the dictionary. To get more accurate translated
VN classes we use a machine learning method, namely Support Vector Machines
(SVM)11 . We follow a straight forward SVM application scenario: we build all
the French verb, VN class pairs hvf r , CV N i where vf r is a translation of an
English verb in CV N . The classifier has to give a probability estimate about
whether this association is correct or not.
For training the classifier we use the 160 verbs appearing in the gold standard
proposed by Sun et al. (2010)12 . We build the pairs hvf r , CV N i where vf r is a
verb in the gold standard which is a translation of a verb in CV N . For each
of these pairs we assessed whether or not there was a meaning of vf r where
the semantic roles involved in the event described by the verb were those given
by CV N . The features associated to the hverb, classi pairs are numeric and are
extracted from the dictionaries and VerbNet.
9
http://www.google.com/dictionary. We obtained 13824 French-English verb pairs.
10
The number of French-English verb pairs we obtained is 11351
11
We used libsvm, the software package and methodology presented on http://www.
csie.ntu.edu.tw/~cjlin/libsvm/, Chang and Lin (2011).
12
In fact this is the only existing gold standard for French VerbNet style classes and
we also use it for the overall evaluation of our system (not presented in this paper).
The trained classifier is then used to produce probability estimates for all
verb, class instances. We select the 6000 pairs with highest probability esti-
mates13 and finally obtain the translated classes by assigning each verb in a
selected pair to the corresponding class.
To give an idea of the quality of the obtained classes: The accuracy of the
classifier on the held out test set was 90%, compared to a maximum accuracy
of 93.84% for five fold cross-validation on the development set. The frequency
distribution of the translated classes obtained this way is much closer to the
distribution of verbs in VerbNet classes as when using an approach based only
on translation frequencies, thus providing more accurate verb groups to guide
the FCA concept - thematic roles associations.
6 The French Verb ↔ Thematic Role Sets ↔ Syntactic
Frame Associations
As a detailed and thorough evaluation of the verb, thematic role sets and syntac-
tic frames associations would be out of the scope of this paper we only give here
an intuition of the type of information provided by our method. Following the
preliminary investigations in the previous sections we associated French verbs
with subcategorisation frames and thematic role sets according to the scheme
listed below:
– We group the VerbNet thematic roles and assign to one class all the VN
verbs whose class have the same role set. We then translate the obtained
classes using the methods described in Section 5.
– We use FCA to group French verbs and syntactic frames associated to these
verbs by the lexicons described in Section 3. The concept lattices we create
are based on the formal contexts consisting of French verbs as objects and
SCFs as attributes.
– We then select the 1500 concepts where the sum of the stability and sepa-
ration indices is highest because in Section 4 we found this combination of
concept selection indices to work best for our application.
– For each translated VN class we identify among the 1500 filtered FCA con-
cepts the one(s) with best f-measure between precision and recall.
The translated VerbNet class is then associated with this FCA concept(s). Thus
the verbs in the FCA concept are effectively associated with the thematic role
set of the translated class and at the same time with the syntactic frames in
the intent (attribute set) of the FCA concept. Figure 3 shows the associations
between concepts, thematic role sets and frames generated by our method for
some VN classes14 . The figure shows the concepts associated to these thematic
role sets and for each of these concepts: their attribute set (syntactic frames),
13
In VerbNet there are 5726 verb, class pairs
14
These are the classes occuring in the gold standard proposed by Sun et al. (2010),
mentioned in Section 5.
1248 5022
32 7191 5312 617
SUJ:NP,OBJ:NP SUJ:NP,OBJ:NP,DEOBJ:PP
SUJ:NP SUJ:NP,OBJ:Ssub SUJ:NP,OBJ:NP,POBJ:PP,POBJ:PP SUJ:NP,DEOBJ:Ssub,POBJ:PP
SUJ:NP,OBJ:NP,POBJ:PP
AgExp-End-Theme
AgExp-Location-Theme AgExp-PredAtt-Theme AgExp-End-Start-Theme AgentSym-Theme
AgExp-Instrument-Patient AgExp-Start-Theme
verb set: 977 verbs verb set: 343 verbs verb set: 52 verbs verb set: 33 verbs
verb set: 1706 verbs verb set: 300 verbs
4584
SUJ:NP
18868 7190
1227 SUJ:NP,AOBJ:PP
SUJ:NP,OBJ:NP SUJ:NP,DEOBJ:PP SUJ:NP,OBJ:NP
SUJ:NP,OBJ:NP
SUJ:Ssub,OBJ:NP SUJ:NP,OBJ:NP SUJ:NP,OBJ:Ssub
AgExp-PatientSym SUJ:NP,OBJ:NP,DEOBJ:PP
AgExp-Cause AgExp-Theme
SUJ:NP,OBJ:NP,POBJ:PP
verb set: 122 verbs
verb set: 354 verbs verb set: 326 verbs
AgExp-Beneficiary-Extent-Start-Theme
verb set: 17 verbs
Fig. 3: French verb ↔ synt. frames ↔ thematic role set associations.
the associated thematic role set(s), the number of verbs in the concept and the
hierarchical relations between the concepts as given by the concept lattice.
Thus for example the following 11 verbs (occuring in the gold standard)
bouger, déplacer, emporter, passer, promener, envoyer, expédier, jeter, porter,
transmettre, transporter are in concept 5312 and thereby may be used in the
construction SUJ:NP,OBJ:NP,POBJ:PP,POBJ:PP15 (according to our lexical re-
sources). When they occur in this construction they are associated with the
thematic role set AgExp, End, Start, Theme, i.e. the semantic roles involved
are an Agent or Experiencer, a Start point, an End point and a Theme.
The listed verbs are all verbs of movement where an agent may move a theme
from a start point to an end point – therefore in this case the associations with
the syntactic frame and thematic role set seem to be correct. An NLP sys-
tem which encounters the verb déplacer for example, used in the construction
SUJ:NP,OBJ:NP,POBJ:PP,POBJ:PP could infer that possible thematic roles in-
volved in the described event are an Agent (or Experiencer), a Theme, an
End point and a Start point. However, it still would not know which thematic
role is realised by which syntactic argument.
There are also some problems with these associations. As can be seen in
Figure 3, there is one case where the classification maps the same concept to two
distinct VerbNet classes (AgExp-End-Theme and AgExp-Instrument-Patient).
In addition, verbs in sub-concepts inherit the class label of the super-concept.
Although there are verbs which belong to several VN classes, in many cases this
multiple mapping was not warranted. Improving the precision of these mappings
requires further investigations.
7 Conclusion
We introduced a new approach to verb clustering which involves the combined
use of the English VerbNet, a bilingual English-French lexicon and a merged
subcategorisation lexicon for French. Using these resources, we built two classi-
fications, one derived from the English VN by translation and the other, from
the subcategorisation lexicons via the construction of a formal concept lattice.
We then use the translated VN to associate FCA concepts with VN classes
15
a transitive construction with two additional prepositional objects
and thereby associate verbs with both syntactic frames and a thematic role
set. We explored the performance of the concept selection indices introduced
by Klimushkin et al. (2010) which are stability, separation and probability at
selecting most relevant concepts with respect to our data and found that the
sum of stability and separation gave best results in the setting of our appli-
cation. These results were similar to those obtained without filtering, showing
that this combination of the indices did indeed allow to select the most relevant
concepts with respect to our data. Finally we showed the French verb, syntactic
constructions and semantic role sets associations we obtained and briefly illus-
trated their potential use. Thus Formal Concept Analysis in combination with
the concept selection indices, translation and set mapping methods proved an
adequate method in this knowledge acquisition process.
Bibliography
Baker, C. F., Fillmore, C. J., and Lowe, J. B. (1998). The berkeley FrameNet
project. In Proceedings of the 17th International Conference on Computational
Linguistics, volume 1, pages 86–90, Montreal, Quebec, Canada. Association
for Computational Linguistics.
Briscoe, T. and Carroll, J. (1993). Generalized probabilistic lr parsing of natu-
ral language (corpora) with unification-based grammars. Comput. Linguist.,
19(1):25–59.
Carroll, J. and Fang, A. C. (2004). The automatic acquisition of verb sub-
categorisations and their impact on the performance of an hpsg parser. In
IJCNLP, pages 646–654.
Chang, C.-C. and Lin, C.-J. (2011). LIBSVM: A library for support vector
machines. ACM Transactions on Intelligent Systems and Technology, 2:27:1–
27:27. Software available at http://www.csie.ntu.edu.tw/~cjlin/libsvm.
Cimiano, P., S.Staab, and Tane, J. (2003). Automatic Acquisition of Taxonomies
from Text: FCA meets NLP. In Proceedings of the PKDD/ECML’03 Inter-
national Workshop on Adaptive Text Extraction and Mining (ATEM), pages
10–17.
Dubois, J. and Dubois-Charlier, F. (1997). Les verbes français. Larousse.
Falk, I. and Gardent, C. (2010). Bootstrapping a Classification of French Verbs
Using Formal Concept Analysis. In Interdisciplinary Workshop on Verbs In-
terdisciplinary Workshop on Verbs, page 6, Pisa Italy.
Falk, I., Gardent, C., and Lorenzo, A. (2010). Using Formal Concept Analysis to
Acquire Knowledge about Verbs. In Concept Lattices and their applications,
page 12, Sevilla, Spain.
Gross, M. (1975). Méthodes en syntaxe. Hermann, Paris.
Guillet, A. and Leclère, C. (1992). La structure des phrases simples en français.
2 : Constructions transitives locatives. Droz, Geneva.
Hadouche, F. and Lapalme, G. (2010). Une version électronique du LVF com-
parée avec d’autres ressources lexicales. Langages, pages 193–220. Mise en
page différente que celle parue dans la revue.
Jay, N., Kohler, F., and Napoli, A. (2008). Analysis of social communities with
iceberg and stability-based concept lattices. In ICFCA’08: Proceedings of
the 6th international conference on Formal concept analysis, pages 258–272,
Berlin, Heidelberg. Springer-Verlag.
Klimushkin, M., Obiedkov, S., and Roth, C. (2010). Approaches to the selection
of relevant concepts in the case of noisy data. In Kwuida, L. and Sertkaya,
B., editors, Formal Concept Analysis, volume 5986 of Lecture Notes in Com-
puter Science, chapter 18, pages 255–266. Springer Berlin / Heidelberg, Berlin,
Heidelberg.
Kupść, A. and Abeillé, A. (2008). Growing treelex. In Gelbkuh, A., editor,
Computational Linguistics and Intelligent Text Processing, volume 4919 of
Lecture Notes in Computer Science, pages 28–39. Springer Berlin / Heidelberg.
Kuznetsov, S. O. (2007). On stability of a formal concept. Annals of Mathematics
and Artificial Intelligence, 49(1-4):101–115.
Palmer, M., Gildea, D., and Xue, N. (2010). Semantic Role Labeling. Synthesis
lectures on human language technologies. Morgan & Claypool Publishers.
Palmer, M., Kingsbury, P., and Gildea, D. (2005). The proposition bank: An
annotated corpus of semantic roles. Computational Linguistics, 31(1):71–106.
Priss, U. (2005). Linguistic Applications of Formal Concept Analysis. In Ganter,
B., Stumme, G., and Wille, R., editors, Formal Concept Analysis, volume 3626
of Lecture Notes in Computer Science, pages 149–160–160. Springer Berlin /
Heidelberg.
Roth, C., Obiedkov, S. A., and Kourie, D. G. (2006). Towards concise represen-
tation for taxonomies of epistemic communities. In CLA, pages 240–255.
Saint-Dizier, P. (1999). Alternation and verb semantic classes for french: Analysis
and class formation. In Predicative forms in natural language and in lexical
knowledge bases. Kluwer Academic Publishers.
Schuler, K. K. (2006). VerbNet: A Broad-Coverage, Comprehensive Verb Lexi-
con. PhD thesis, University of Pennsylvania.
Sporleder, C. (2002). A Galois Lattice based Approach to Lexical Inheritance
Hierarchy Learning. In 15th European Conference on Artificial Intelligence
(ECAI’02): Workshop on Machine Learning and Natural Language Processing
for Ontology Engineering, Lyon, France.
Sun, L., Korhonen, A., Poibeau, T., and Messiant, C. (2010). Investigating
the cross-linguistic potential of VerbNet-style classification. In Proceedings
of the 23rd International Conference on Computational Linguistics, COLING
’10, pages 1056–1064, Stroudsburg, PA, USA. Association for Computational
Linguistics.
Tolone, E. (2011). Analyse syntaxique à l’aide des tables du Lexique-Grammaire
du français. PhD thesis, LIGM, Université Paris-Est, France, Laboratoire
d’Informatique Gaspard-Monge, Université Paris-Est Marne-la-Vallée, France.
(326 pp.).
van den Eynde, K. and Mertens, P. (2003). La valence : l’approche pronominale
et son application au lexique verbal. Journal of French Language Studies,
13:63–104.