DOMAIN SPECIFIC LANGUAGE IN TECHNICAL SOLUTION
DOCUMENTS
Discussion of Two Approaches to Improve the Semi-automated Annotation
Helena X. Schmidt, Andreas Kohn and Udo Lindemann
Institute of Product Development, Technische Universität München, Boltzmannstraße 15, 85748, Garching, Germany
Keywords: Solution knowledge, Ontology-based knowledge system, Mechanical engineering, Domain specific
language, Sublanguage, Annotation.
Abstract: The efficient search for existing solutions in mechanical engineering is a key-factor for successful product
development. Ontology-based knowledge systems can support the semi-automated annotation of documents
about existing solutions and enable the retrieval of those documents. However, the use of different wordings
for similar products and a generally heterogeneous domain-specific language hinder the efficient annotation
process. In this paper, two approaches to improve the semi-automatic annotation of documents by adding
terms to the ontology are described. We evaluate the two approaches by analysing the industry sector-
specific and company-specific languages used in documents in the field of mechanical engineering.
1 INTRODUCTION
In engineering, an important part of the product
development process is the research for existing
solutions. The discovery of existing solutions to a
technical problem can shorten the product
development process significantly. Certain barriers,
such as unstructured data and the different use of
language, hinder the access to existing solutions and
increase the product developer’s effort necessary for
the search (Gaag et al., 2009); (Kohn et al., 2010).
These problems in solution research are
addressed by the use case PROCESSUS which is
part of the German research project THESEUS.
Based on sales publications from the automation
industry an ontology for representing knowledge
about technical solutions has been developed.
Solution documents from different industry sectors
and different companies can be integrated into the
ontology structure. An ontology-based prototype
supports the semi-automated annotation of solution
documents to integrate them into the ontology and
the subsequent retrieval of relevant solution
documents (Gaag et al., 2009).
With the prototype, the user can annotate
solution documents semi-automatically. This can
reduce the time needed for the annotation
significantly. The text document is imported into the
prototype. In combination with linguistic algorithms,
such as word stemming and syntactic analysis, terms
that already exist in the ontology as instances are
recognized and suggested to the user. Therefore, the
semi-automated annotation’s success depends on the
completeness of the instances contained in the
ontology.
This paper focuses on improving the semi-
automated annotation of solution documents by
adding the appropriate instances for describing
technical products within a solution to the ontology.
As mentioned above, a factor for the identification
of terms is the language used to describe them in
solution documents. The language can differ from
industry sector to industry sector and from company
to company. The use of different terms and language
in specific domains and its importance for
knowledge management is a general challenge and
has been addressed by scientists in different areas
such as linguistics, information technology and
engineering.
Therefore, this paper will first provide an
overview of the research on language in specific
domains. In the next step, challenges for the semi-
automated annotation of technical solution
documents are shown. Then, two approaches are
explained and evaluated with an exemplary analysis
of the language used in solution documents.
159
Schmidt H., Kohn A. and Lindemann U..
DOMAIN SPECIFIC LANGUAGE IN TECHNICAL SOLUTION DOCUMENTS - Discussion of Two Approaches to Improve the Semi-automated
Annotation.
DOI: 10.5220/0003630301590166
In Proceedings of the International Conference on Knowledge Engineering and Ontology Development (KEOD-2011), pages 159-166
ISBN: 978-989-8425-80-5
Copyright
c
 2011 SCITEPRESS (Science and Technology Publications, Lda.)

2 LANGUAGE IN SPECIFIC
DOMAINS AND ITS ROLE FOR
KNOWLEDGE MANAGEMENT
In this section different approaches to the use of
language in specific domains are presented. Then,
the methods used for the analysis of language in this
work are explained.
2.1 Research on Language in Specific
Domains
Why is the examination of domain specific language
important for knowledge management and – in
particular – for ontology-based information retrieval
systems? According to Thellefsen (2003), in systems
for knowledge management the use of special
language in documents is not considered
sufficiently. He states that today, instead of adapting
the system, the documents have to be adjusted to the
system which leads to unefficient annotation and
retrieval of knowledge.
To point out the importance of the context in
which language is used several authors cite
Wittgenstein. He introduced the term “language
game” describing patterns in which a meaning of a
word is explained by its use (Petras, 2006;
Thellefsen, 2010).
This definition implies that one term can have
different meanings in different contexts or as Petras
states, ”there is no one-to-one mapping between a
sign (term) and a concept (meaning)” (2006, p. 15).
For knowledge management, this means that the
same term can have different meanings for users
with a different context.
Further semantic relations of terms play a role,
such as synonymy (two terms with the same
meaning) and hyponymy (one term is subordinate to
another) (Vossen, 2003).
In linguistics there are different approaches to
analyse the use of language in specific domains.
One approach is the study of sublanguages. It
focuses on the syntatic and lexical characteristics of
the language. According to Grisham and Kittredge
(1986) a sublanguage is the language used by a
community of speakers in a specific domain.
Kittredge (2003) lists a number of characteristics of
sublanguage such as a restricted lexicon and limited
term co-occurrence structures. Losee and Haas
(1995) establish statistic measures to evaluate the
degree of specification of different sublanguages.
They analyse abstracts from different scientific
fields such as history and electric engineering and
compare their occurrence and their meaning in
specialized and general dictionaries. Another
approach is the study of language for special
purposes (LSP). The emphasis lies on the semantic
characteristics (Petras, 2006). The characteristics
analyzed are stylistic, such as the use of conditional
sentences and the discursive line of the text
(Evangelisti Allori, 2001).
In product development, linguistic analyses have
been employed to facilitate the research for
analogies from biology. Cheong et al. (2008)
translate engineering terms into biological keywords
to draw analogies from biology for solution
research. The most promising biology terms are
selected by their occurrence in biology dictionaries.
Terms that occur very frequently are considered
insignificant because they are too general, terms that
occur very rarely because they are too specific.
Another research linked to product development
was conducted by Bohm and Stone (2009). They
propose an approach to map terms from a
component database to terms of a component
taxonomy. By comparing the similarity of the
component’s naming terms and their function, they
determine synonyms for the terms of the component
taxonomy.
Summing up, most of the presented approaches
focus on scientific texts. Bohm and Stone’s work is
an exception, as their scope are documents provided
by product developers for product developers. The
research presented in this paper is focused on
another type of documents: sales publications which
describe technical solutions for the customer.
2.2 Methods for Analysing Language in
Specific Domains
In linguistics, the term count is a parameter used for
the analysis of language in text documents. The term
count is the number of occurrences of a term in one
or several documents. A theory states that the
distribution of the term count in a number of
documents approximately follows a Poisson
distribution (Losee, 1995). The Poisson distribution
is a statistic law for events that occur with a known
average rate independently from each other. If the
distribution of an event follows a Poisson
distribution, its probability P that it occurs k times
can be calculated by equation (1). l is the expected
value, i.e. the arithmetic mean of occurrence
(Härtter, 1974).
() =



!
∗

(1)
Applied on term counts the occurrence k equals the
KEOD 2011 - International Conference on Knowledge Engineering and Ontology Development
160

term count. If term counts follow a Poisson
distribution, the probability of a certain term count
in a single document can be calculated (Losee,
1995).
3 CHALLENGES OF
SEMI-AUTOMATED
ANNOTATION
Solution documents for sales publications from the
automation industry served as a basis for the
development of the ontology (Gaag et al., 2009).
Figure 1 shows the mapping of knowledge from the
solution documents to the ontology structure. The
boxes contain concepts which are connected by
relations. The concrete instances are assigned to the
concepts.
Figure 1: Mapping of solution knowledge to the ontology
structure.
In common design theory, the research for
solutions is usually based on the function the
solution should perform (Ponn and Lindemann,
2008). Therefore, the function is set as the central
concept in the ontology. A technical solution, e.g.
the instance “robot 534”, realises a function. A
function consists of the concepts operation and
object. Figure 1 shows an exemplary sentence from
a solution document. Here, the operation is “stack”
and the object is “bottles”. The function is “stack
bottles”. A function is conducted by a function
owner. In this case the function owner is “system”.
As to the use of terms for function owners
different concretization levels and the use of
synonyms have been observed. As an example, we
examine the function owner “system” from Figure 1.
The term “system” does not describe the function
owner’s characteristics. Instead of naming the
function owner “system”, it can be concretised as
”robot”. It can be further concretized as “packing
robot”, i.e. a term that concretises the function of the
robot. There are more possibilities to concretize the
term “robot”, for example a property of the robot
can be described. A robot with articulations can be
described as an “articulated robot”.
On the same concretization level a function
owner can be described by a synonymous term.
Instead of “system”, “equipment” or “machine” can
be used. It depends on the context if the terms are
synonymous. For example “machine” can be used as
a synonym for “system” in the context of an
automation system composed of several machine
components. It seems less feasible to use it for
systems with different system boundaries, i.e. bigger
or smaller systems, for example - in the context of a
valve system that is part of a machine itself. The
synonyms can lead to other synonyms at a different
concretization level. Instead of naming the function
owner “packing robot” it can be referred to as
“packing machine”, for example. In conclusion, to
improve the semi-automated annotation the diverse
terms used for function owners have to be added.
The two approaches described and discussed in
this paper are the approach of noun extraction
(section 4) and embedding classifications (section
5). The usefulness of the two approaches is
evaluated for single companies and for several
companies from the same industry sector. The
applicability for different industry sectors is not
regarded in this paper because the probability that
different terms are used is higher. If an approach
proves to be useful for several companies from one
industry sector, its applicability for different industry
sectors can be evaluated in a next step.
4 NOUN EXTRACTION
The following section details the procedure to
identify function owners. The underlying
assumption of this approach is that function owners
found in the selection of solution documents also
occur in unknown solution documents. The
correctness of this assumption is evaluated in the
second section.
4.1 Procedure to Identify Function
Owners
Figure 2 shows the process of noun extraction. To
start with, solution documents are selected. They
have to cover different solutions the company offers
to provide a representative selection of documents
with a variety of function owners. As function
owners in the ontology are nouns, the nouns from
the solution documents are extracted. This step can
be automated. The next step is the manual
annotation performed by “experts”, i.e. users that are
familiar with annotation. Then, the function owners
The system stacks bottles .
function owner
objectoperation
function
executes
performs works on
technical
solution
realises
robot 534
DOMAIN SPECIFIC LANGUAGE IN TECHNICAL SOLUTION DOCUMENTS - Discussion of Two Approaches to
Improve the Semi-automated Annotation
161

Figure 2: Noun extraction.
are added to the ontology.
4.2 Evaluation of the Approach
Is the assumption that function owners from a
representative selection of solution documents also
occur in unknown solution documents correct?
This can be evaluated on different levels. In this
research, we analysed if
1. function owners extracted from one company
occur in the solution documents from this company
(company level).
2. function owners extracted from one company
occur in the solution documents of a different
company from the same industry sector (industry-
sector level).
For the evaluation, a sample of German solution
documents for sales publication of two companies
from the automation industry was used. From
company A 28 solution documents and from
company B nine solution documents were used.
Using a different number of solution documents
allowed to observe if there is a correlation between
the number of solution documents and the number of
function owners that were extracted.
From the solution documents of company A 168
function owners, for the solution documents of
company B 111 function owners were extracted
using the procedure explained in section 4.1.
For the extraction of the nouns from the solution
documents, the software tools TreeTagger
developed by Schmid (2011) and RapidMiner from
Rapid-I GmbH were used (Rapid-I GmbH, 2011)
were used. Two scientific assistants annotated the
function owners in the noun list.
The usefulness of the approach for the semi-
automated annotation of unknown documents from
the same company is evaluated in 4.2.1 (company
level). Then, in 4.2.2, its usefulness for documents
from the other company is evaluated (industry sector
level).
4.2.1 Evaluation on the Company Level
For the evaluation, the term count of the extracted
function owners in the solution documents is
analysed (see section 2.2). It has to be differentiated
between the term count of a function owner in all
documents and its term count in the single
documents. The distribution of the term count in the
single documents shows how often a function owner
occurs in how many documents. If a function owner
occurs in a significant number of documents and is
distributed regularly this is considered as an
indication that it will occur in unknown solution
documents as well. Both the significant number of
documents and the regular distribution depend on
the term count of the function owner in all
documents.
In the next step, to further evaluate if an
extracted function owner will occur in unknown
solution documents, the real distribution of the term
is compared to a Poisson distribution (see section
2.2).
Table 1 lists the nine function owners identified
by noun extraction with the highest term counts in
all documents for company A (Table 1). The
function owners with the highest term counts in all
documents are shown because a high term count is
needed to analyse if the function owners occur in a
significant number of documents and are regularly
distributed. On the right side of Table 1 the
distribution of the term count, i.e. the number of
documents in which the function owner occurs with
a certain term count is shown. To illustrate, the
numbers are coloured. A dark colour means a high
number of documents.
The most frequent function owner “robot” occurs
415 times in all documents (term count: 415). It
occurs in all 28 documents at an average 14,82
times. “Robot” has a minimum term count of nine in
one document and a maximum term count of 20 in
four documents. Between minimum and maximum,
every term count can be found in at least one
document except the term count 19. Therefore it is
concluded that “robot” has a relatively regular
distribution over the 28 documents.
The second frequent function owner “gripper”
occurs 66 times in all documents. It occurs in 24
documents which is considered a significant number
Solution
documents
robot
bottle
steel
Nouns Function owners
Automated
extraction
Manual
annotation
Addition
Ontology
concept 1
concept 3
concept 4
concept 5
concept 2
concept 6
robot
system
axis
KEOD 2011 - International Conference on Knowledge Engineering and Ontology Development
162

Table 1: Term counts of function owners from company A. Index: 1) Term count in all documents, 2) Number of
documents that contain the function owner, 3) arithmetic mean of the term count in the single documents.
Figure 3: Real distribution compared to Poisson distribution.
of documents. The minimum term count is zero in
four documents and the maximum term count is six
in one document. In between minimum and
maximum all term counts can be found in at least
one document indicating a regular distribution.
This is the case for most of the next function
owners, except “axis” and “roll conveyor”. Still,
these two function owners have similar term counts
in most documents.
As to company B, the term counts are lower than
in the solution documents of company A due to the
smaller number of documents. The function owners
occur in more than one document but have a less
regular distribution than function owners from
company A.
In the next step, the distribution of a function
owner’s term count is compared to a Poisson
distribution to further evaluate if the conclusion to
unknown solution documents is feasible.
As an example, the distribution of the function
owner “robot” in the solution documents of
company A is analysed. The arithmetic mean of its
term count is 14,82. Consequently the expected
value l of the adequate Poisson distribution is 14,82.
For the real distribution of the term counts the
portion of documents with a certain term count is
calculated. Figure 3 shows that the probability P and
the portion of solution documents that have a certain
term count are not equal. The real distribution of the
term counts does not strictly follow a Poisson
distribution. This result is in accordance with the
results obtained by Losee’s analysis of scientific
abstracts and the results of several other authors
(Losee, 1995).
Even though a Poisson distribution was not
found, both for company A and company B the
majority of the most frequent function owners occur
in a ”significant number of documents” and not only
in one document. To sum up, it is inferred that
extracted function owners that occur frequently in a
sample of documents will also occur in unknown
solution documents from the same company.
4.2.2 Evaluation on the Industry Sector
Level
In accordance to the previous analysis for a single
company, the term counts of the function owners
English te rm 1) 2) 3) Te rm count s ingle document
01234567891011121314151617181920
robot 415 2814,82 000000000112434151104
gripper 66 242,36 473742100000000000000
industry robot 59 282,11 0025300000000000000000
robot control 47 271,68 11014300000000000000000
palletiser 45 151,61 1362211120000000000000
axis 33 131,18 1582010100001000000000
robot cell 24 140,86 1483210000000000000000
facility 21 150,75 1396000000000000000000
roll conveyor 19 110,68 1764001000000000000000
numbe r of docume nts
0
0,02
0,04
0,06
0,08
0,1
0,12
0,14
0,16
0,18
0,2
5 6 7 8 9 1011121314151617181920
Term count k
probability P/ portion of solution documents
real distribution of the term count of „robot“
poisson distribution with l=14,82
DOMAIN SPECIFIC LANGUAGE IN TECHNICAL SOLUTION DOCUMENTS - Discussion of Two Approaches to
Improve the Semi-automated Annotation
163

from the two companies can be compared. Before
comparing the distribution of the term counts in
detail, it is examined how many function owners
from company A equal function owners from
company B. The result is that 14 function owners
from company A were also extracted from company
B. This amounts to 8 per cent of the function owners
extracted from company A or 13 per cent extracted
from company B. This low percentage of equal
function owners shows that the noun extraction from
documents from company A does not provide a
significant number of function owners that occur in
the solution documents from company B. Therefore,
a further analysis of the term counts is considered
unnecessary.
To analyse why the number of equal function
owners in the documents of the two companies is so
low, the function owners can be regarded with a
focus on their meaning.
As an example, the function owner “robot” is
examined. As explained in section 3, a function
owner can have different concretization levels. In
the solution documents from company A in addition
to “robot” eight more concretely defined “robots”
are used. In comparison, in the documents from
company B two concretizations of “robot” are used.
They are different from the concretizations of
company A. The concretizations of company A and
B are not synonymous. Examining the
concretizations from company B more closely, the
term count of “spot welding robot” is one and of
“swivel-arm robot” is two. Both occur in one
solution document respectively. This is an indication
that they are very specific and that company B refers
to most robots with the term “robot” without further
concretizing them (term count of 72). Even though
in documents from company A the term count of the
general term “robot” is 415, concretizations such as
“palletising robot” and “articulated robot” have a
relatively high term count of 18 and 17. Thus, in
documents from company A the function owner
“robot” is concretized more often than in documents
from company B.
This example shows how function owners are
used differently by the two companies. For other
function owners similar observations were made.
Even though both companies are from the same
industry sector and have similar products, they use
different terms for their function owners. The use of
language for function owners in solution documents
from company A and B is company-specific.
Therefore, the noun extraction of documents from
one company does not provide a significant number
of function owners relevant in documents from other
companies.
5 EMBEDDING
CLASSIFICATIONS
In this section the approach of embedding
classifications is explained and evaluated for three
different classifications (5.1). The results are
compared in section 5.2.
For this approach terms from three different
product classifications are added as function owners
to the ontology. This has been described by Hepp
(2005) for the eCl@ss classification.
5.1 Evaluation of the Approach
In this paper three classifications are regarded. The
function owners from company A and B are
compared to classes included in the classification.
To justify the effort to embed classifications into the
ontology, at least 40 % of the function owners
should be included in a classification. For the
comparison, the function owners obtained by noun
extraction from company A and B were used.
Hereafter, the three classifications are described and
the results of the comparison are shown in Table 2.
5.1.1 VDMA e-Market
The VDMA e-Market is a platform provided by the
VDMA, a German industry association with member
companies from the capital goods industry. It
contains approximately 200.000 product descriptions
embedded in a classification (VDMA Verlag GmbH,
2011).
The VDMA e-Market classification is structured
into eight industrial sectors. The sectors contain
1571 classes. The main product in a solution
document is identified and assigned as an instance to
a class (VDMA Verlag GmbH, 2011). As the objects
are very specific product names, for the evaluation
the function owners were compared to the classes in
the VDMA e-Market.
5.1.2 eCl@ss
eCl@ss is a hierarchic system that classifies
materials, products and services by standardized
properties. It has been developed within a project
funded by the German Ministry of Economy and
Technology (eCl@ss e.V., 2011).
The eCl@ss classification contains classes divi-
KEOD 2011 - International Conference on Knowledge Engineering and Ontology Development
164

ded into functional areas, main groups, groups and
sub-groups. Products and services are classified on
the subgroup level. Classes have properties with
values, for example the length defined in mm. Key-
words are assigned to classes. For the analysis,
version 6.0 was used which contains 32592 classes
and 51329 key words (eCl@ss e.V, 2011). For the
evaluation, the classes were compared to the
function owners.
5.1.3 UNSPSC®
UNSPSC® (United Nations Standard Products and
Services Code®) is a standard for the classification
of products and services developed within the
United Nations Development Programme.
The classification includes 31498 classes
numbered with a code (UNSPSC, 2011). For the
evaluation, they were compared to the function
owners.
5.2 Results
The results of the comparison are depicted in Table
2. The UNSPSC® classification includes 21 per cent
of the function owners from company A and 33 per
cent from company B. The other classifications
contain less than 20 per cent of the function owners
from company A and B. This is a relatively low
percentage.
Table 2: Number of function owners included in
classifications.
Company A Company B both
VDMA e-Market
22
(13 %)
8
(7 %)
6
(43 %)
eCl@ss
16
(10 %)
13
(12 %)
4
(29 %)
UNSPSC®
35
(21 %)
37
(33 %)
12
(86 %)
In addition, the number of classes from the
classifications is relatively high, especially for
eCl@ss and UNSPSC®. Both classifications contain
all types of products including food and plants.
The comparison of the function owners that
company A and B have in common leads to a
different result: Up to 86% of these 14 function
owners occur in the classifications.
In conclusion, embedding the product
classifications examined in this section does not
improve the semi-automated annotation of the two
companies significantly, because they contain less
than 40 % of the examined function owners. The
effort to include several thousands of classes into the
ontology seems to be high in comparison. The
function owners that are used by both companies are
included with a significant percentage in the
classifications. Still, as their number is relatively low
this is no feasible improvement. With regard to the
VDMA E-Market classification, the effort is lower
which can justify embedding it, even though the
improvement of the semi-automated annotation is
slight.
6 DISCUSSION
The evaluation of the two approaches was performed
with a small number of documents and companies.
For the analysis of the term counts’ distribution
exact numbers for the “significant number of
documents” and the “regular distribution” could not
be stated. Consequently, the evaluation provides
indications for the usefulness of the approaches for
single companies and for different companies from
the same industry sector. For general assumptions, a
bigger document basis is needed which was not
available at this point.
In addition to the evaluation of the two
approaches, the conduction of the noun extraction
provided insights about the strengths and
weaknesses of noun extraction. Noun extraction can
only extract nouns that consist of one word.
Composite nouns such as “robot 75”and nouns
containing hyphens such as “XY-robot” are not
identified as one term. In German this poses fewer
problems than in many other languages, because
terms are often merged from several words. For
example, “industry robot” is “Industrieroboter”.
During the manual annotation of the noun list, the
annotation of the two scientific assistants differed on
several function owners. A few function owners
were overlooked by one assistant but annotated by
the other. They disagreed if some nouns were
function owners or not. In addition, a number of
terms were annotated that are not used as function
owners but as objects in the documents. An example
is the term “work holding fixture”. In common
understanding this is a function owner, but in the
analysed documents the term stands for an object
which is assembled by a robot. Linguistic algorithms
that distinguish between subjects and objects can be
a solution to this problem. In most cases the function
owner is the subject of the sentence whereas the
object of a function is also “grammatically” an
object.
On the other hand, a strong point of the noun
extraction was the completeness of the list of
function owners extracted. In previous research
DOMAIN SPECIFIC LANGUAGE IN TECHNICAL SOLUTION DOCUMENTS - Discussion of Two Approaches to
Improve the Semi-automated Annotation
165

solution documents were manually annotated by
several persons (Kohn et al., 2010). With the
function owners identified by noun extraction of the
manually annotated function owners up to 80 per
cent of the manually annotated function owners are
covered.
7 CONCLUSIONS AND
OUTLOOK
In this work two approaches to improve the semi-
automated annotation of solution documents in
mechanical engineering were described and
evaluated exemplarily. The first approach, the noun
extraction, is promising if it is used to improve the
semi-automated annotation of documents from the
same company. For the annotation of documents
from other companies from the same industry sector
the results are not satisfying. This is due to
company-specific use of language to describe
function owners. The results for the approach of
embedding existing classifications are less
promising. The three regarded classifications
contained a relatively low number of function
owners.
This work discloses a number of starting points
for future research. The noun extraction can be
improved by applying linguistic algorithms to
identify terms composed of several words and to
distinguish between subjects and objects. For the
embedding of classifications, other classifications
can be regarded. As to the nature of function owners,
the different specification levels could be further
examined. In addition, synonyms can be added to
the ontology.
ACKNOWLEDGEMENTS
Part of this work has been funded by the German
Federal Ministry of Economy and Technology
(BMWi) through THESEUS.
REFERENCES
Bohm, M. R. and Stone, R. B. (2009). A Natural
Language to Component Term Methodology: Towards
a Form based Concept Generation Tool. ASME IDETC
2009. San Diego, USA.
Cheong, H.; Shu, L.; Stone, R. B. (2008). Translating
Terms of the Functional Basis into Biologically
Meaningful Keywords. In ASME IDETC‚08.
Brooklyn, USA.
eCl@ss e.V.(2011). eCl@ss – Das führende
Klassifikationssystem. Retrieved May 5, 2011 from
http://www.eclass.de.
Evangelisti Allori, P. (2001). Conceptual and genre-
specific constraints: How different disciplines select
their discoursal features. In: Mayer, F. (ed.): Language
for special purposes: perspectives for the new
millenium. (pp. 70-79). Tübingen: Gunter Narr
Verlag.
Gaag, A., Kohn, A. and Lindemann, U. (2009). Function-
based Solution Retrieval and Semantic Search in
Mechanical Engineering. In ICED’09. Stanford,
California, USA.
Grisham, R., Kittrege, R. (1986). Analysing language in
restricted domains. Hillsdale: Lawrence Erlbaum.
Härtter, E. (1974). Wahrscheinlichkeitsrechnung für
Wirtschafts- und Naturwissenschaftler. Göttingen:
Vandenhoeck und Ruprecht.
Hepp, M. (2005). A Methodology for Deriving OWL
Ontologies from Products and Services Categorization
Standards. In ECIS 2005. Regensburg, Germany.
Kittredge, R. I. (2003). Sublanguages and controlled
languages. In Mitkov, R. (Ed.), The Oxford handbook
of computational linguistics. (pp. 430-447). Oxford:
Oxford University Press.
Kohn, A., Peter, G. and Lindemann, U. (2010). The
Challenge of Automatically Annotating Solution
Documents. In KEOD’10. Valencia, Spain.
Losee, R. and Haas, S. (1995). Sublanguage Terms:
Dictionaries, Usage and Automatic Classification.
Journal of the American Society for Information
Science, 519-529.
Petras, V. (2006). Translating Dialects in Search: Mapping
between Specialized Languages of Discourse and
Documentary Languages. PhD thesis, University of
California, Berkeley.
Ponn, J., and Lindemann, U. (2008). Konzeptentwicklung
und Gestaltung technischer Produkte. (1rst ed.).
Berlin: Springer.
Rapid-I GmbH (2011). RapidMiner. Retrieved January 13,
2011 from http://rapid-i.com.
Schmid, H. TreeTagger (2011). Retrieved January 20
,
2011 from http://www.ims.uni-stuttgart.de
Thellefsen, M. (2003). The role of special language in
relation to knowledge organization. Proceedings of the
American Society for Information Science and
Technology, 206-212.
Thellefsen, M. (2010). Knowledge Organization,
Concepts, Signs: A Semeiotic Framework. PhD
Thesis, Royal School of Information and Library
Science, Denmark.
UNSPSC (2011). UNSPSC®. Retrieved May 5, 2011 from
http://www.unspsc.org/
Vossen, P. (2003). Ontologies. In: Mitkov, R. (Ed.): The
Oxford Handbook of Computational Linguistics.
(pp.464-482). Oxford: Oxford University Press.
VDMA Verlag GmbH (2011). Über den VDMA E-Market.
Retrieved May 5, 2011 from http://www.vdma-e-
market.com/de/ueberEmarket.
KEOD 2011 - International Conference on Knowledge Engineering and Ontology Development
166