SemanticHPST is a project in which interacts ICT (especially Semantic Web) with history and philosophy of science and technology (HPST). Main di culties in HPST are the large diversity of sources and points of view and a large volume of data. So, HPST scholars need to use new tools devoted to digital humanities based on semantic web. To ensure a certain level of genericity, this project is initially based on three sub-projects: the rst one to the port-arsenal of Brest, the second one is dedicated to the correspondence of Henri Poincare and the third one to the concept of energy. The aim of this paper is to present the project, its issues and goals and the rst results and objectives in the eld of harvesting distributed corpora, in advanced search in HPST corpora. Finally, we want to point out some issues about epistemological aspects about this project.
The application of computer science to research in history has existed for a long
time [
The paper is organized as follows. Section 2 presents the main goals of the SemanticHPST project and its three French HPST sub-projects for which semantic web technologies are useful. Section 3 presents some requirements and corresponding tools supporting di erent resource retrieval processes according to the researchers' practices. Section 4 presents some issues from an epistemological viewpoint. Section 5 concludes the paper. 2
In November 2010, the main topic of a European workshop was the uses of ICT
and history of science and technology in education.2 To improve research in
HPST on one hand, and to promote dissemination of the HPST in the eld of
education on the other hand, some participants were convinced by the necessity
to use new ICT tools [
In 2012, some historians of science and technology and computer scientists have created a consortium called SemanticHPST.3
The main goal of SemanticHPST project is to enrich the practices of researchers and communities in HPST. According to the speci city of the practice as historians of science, three main issues were tackled: 1. The management of large quantities of data especially for the most recent periods (xixth, xxth centuries up to the present day). Knowing that the historical approach involves to integrate relevant elements from the context of production of these data into metadata. 2. The heterogeneity of sources and corpora constituted from these sources. 3. The production of new relevant digital corpora from several available digital historical collections.
To address our main goal and the three previous issues, our project is based on the Semantic Web principles and technologies. Thus, it has three main sub-goals: (i) Building intelligent digital corpora, that is to say corpora with primary and secondary sources having semantic metadata and their corresponding ontologies; (ii) Designing tools to access and enrich existing corpora and to create new ones; (iii) Evaluating the resulting practices and building an epistemological viewpoint about the use of TIC in HPST.
To achieve these goals, it is necessary to ensure a certain level of genericity for metadata, ontology, computer-based tools and practices.
To deal with genericity and the diversity of sources, the project is applied in three di erent use cases or sub-projects with the aim to cover di erent methods and approaches that are typical in the domain of HPST. Those approaches are covering only partially the methods used in history and archaeology. These subprojects are described in the following paragraph. 2.1
This sub-project takes part in the research programs \History of marine science
and technology" and \Digital Humanities for History of Science and Technology"
developed in Brest in the Centre F. Viete. One topic concerns the comprehension
of the scienti c and technological evolution of the port-arsenal in Brest (France)
on a large period (xviith to xxth century) with a methodological approach
considering this military-industrial complex dedicated to shipbuilding as a large
technological system [
The hypothesis is to consider the large technological system of the
portarsenal as a large spatiotemporal and multi-scale artifact which is possible to
decompose in elements of smaller scale (which are also artifacts) like industrial
workshops, shipbuilding areas, storage areas, etc. Each of these elements are
themselves composed by elements/artifact of smaller scale. The system has to
be seen as the sum of all these artifacts and of all the relationships between them.
The research in Brest [
That \life cycle" involves the elaboration of ves categories of entities: time
entities, actors (individuals or social groups), concepts/theories, location and
artifacts. The analysis of the important ontology in the domain of cultural
heritage named CIDOC-CRM (that \provides de nitions and a formal structure for
describing the implicit and explicit concepts and relationships used in cultural
heritage documentation")5 shows that this ontology could be a rst reference
to help and build our own ontologies because some concepts and relationships
about \temporal entities" and \actors" can be reused. But if the concept of
\Thing" exists in the CIDOC-CRM, we consider that the concept of \Artifact"
and the associated relationships have to be elaborated rst from our historical
model and by considering of course the possibility of equivalent concepts in the
CIDOC-CRM. A work is in progress in Brest about this topic from concrete
examples of artifact as crane, quays and seawalls. A second step will be to examine
others methods to produce ontologies well-adapted to our HPST problems in
the domain of marine history [
This work is coupled with examples of typical requests (when and where were positioned all the cranes in the port of Brest since 1650 until 1970? In the 4 See http://brmdp.hypotheses.org/. 5 http://www.cidoc-crm.org/. port of Mar del Plata? Which rms were in charge of the construction of the quays/cranes in the port of Brest since 1800 until 1900? What are the engine power of all cranes in the world since 1850 until 1970? Etc.). 2.2
The platform Henri Poincare papers. In 1992, the laboratory of history of
science and philosophy Archives Henri Poincare was created to promote Henri
Poincare's manuscripts and to publish his correspondence. For more than 20
years, this long-term project has produced three volumes of letters: the rst
one is devoted to the Poincare - Mittag-Le er letters [
The corpus consists of more than 2000 letters, 1046 sent by Henri Poincare and 949 received by him.7 All known letters are digitalized8 and around 50% of them are in plain text (in LATEX and XML versions). Lots of letters contain mathematical and physical formulae. In Henri Poincare Papers website,9 the correspondence is available. In this platform, each known letter is indexed with Dublin Core extended metadata.10 This enables to query the corpus by e.g.
Q1 = \Letters sent by Henri Poincare in 1885"
Q2 = \Letters received by Eugenie Launois between 1882 and 1894" There is also the possibility of plain text search for the letters already transcribed.
Towards more HPST-adapted search. Now, consider the following queries: Q3 = \Letters from an astronomer" Q4 = \Letters in reply to a letter of Mittag-Le er" Q5 = \Letters about the n-body problem"
Q6 = \Letters of the late xixth century"
These queries cannot be executed in the current platform. They require
additional data and knowledge:
6 This correspondence is partly online http://henripoincarepapers.univ-lorraine.fr.
7 About 50% of this letters are with scientists. Original letters come from 63 di erent
archive centers and libraries from 14 countries.
8 Due to copyright laws, some are not available online.
9 http://henripoincarepapers.univ-lorraine.fr.
10 It exists di erent projects devoted to scienti c correspondences for example the
CKCC project (http://ckcc.huygens.knaw.nl) [
The possibility to take into account such queries using semantic web principles and technologies, are examined in the SemanticHPST consortium. 2.3
One part of the SemanticHPST project is dedicated to the concept of energy. Our aim is to create an ontology of energy for researchers working in the eld of HPST as well as for science teachers.
For researchers, the ontology aims at making available a methodical body of knowledge that allows previously unseen connections to be made. For example, correspondence between two authors or the presence of a speci c term or concept in a text will allow researchers to put forward hypotheses regarding the emergence of an idea or the cross-fertilization of ideas.
For teachers, the ontology aims at acting as a resource, allowing educators to nd historical information relevant to school curricula as well as ideas for speci c activities to carry out in the classroom.
The content consists of reference texts in the eld of HPST, contemporary scienti c texts and a database of historic scienti c instruments and documents. This content is currently being selected and developed and will be enhanced as the research progresses.
To date, the following three steps have been undertaken on the project: { The rst step was to identify the presumed ways the ontology will be used, for example, the type of requests that a researcher or teacher might make in a search. To this end, one `persona' for a researcher and one for a teacher have been created. Analyzing the theoretical queries from these two personas helps in the selection of a relevant body of work and is also a useful guide for indexing. { The second step was to begin indexing the reference texts. Duhem, Poincare, Mach and Meyerson have been selected for a rst approach in order to produce keywords and common references and to outline an embryonic model. Using the shared scienti c knowledge of the physicists involved in the project, a sort of `cloud' of concepts related to describing energy was de ned and classi ed. These elements led to the structure of an initial mind map. { Finally, based on this mind map (created with Docear), we used Protege software to create a rst draft overview of the project. The next steps require documenting these three steps in detail to re ne the data and then build the ontology.
During the stages of the project carried out so far, various problems have been identi ed that must be resolved. One of the main problems concerns the modeling of time. How can an event be modeled? Moreover, how can knowledge be modeled in a way that avoids immobilizing the knowledge? How should knowledge be contextualized? What approach should be adopted when modeling concerns a concept or an object? How can a coherent and logical body of content be created and how can its coherence be assessed? It is clear that the question of time as well as how to approach the treatment of objects and works are issues to be investigated in the semanticHPST project. 3
According to the three described sub-projects, the main goals of researchers in HPST are to access and retrieve relevant resources in existing primary and secondary sources or corpora, to produce new resources in existing corpora, to enrich existing digital corpora or to create new ones, for answering research questions in the history of science and technology. Existing digital corpora come from libraries, information holdings, digital libraries or others like Gallica (http://gallica.bnf.fr), Internet Archive (http://archive.org), Google Books (http://books.google.com), etc., and CMS (Content Management System) (blogs, wikis, Drupal, Omeka, etc. more generally social media tools) have been used by the community11 and digital AHP (http://www.ahp-numerique.fr/). Some heritage and bibliographic resources have already been described by several institutions, associations and/or project (BNF, Gallica, British Museum, Europeana, Amsterdam Museum, LODLAM, ...). The creation of new corpora or resources can be made on social media tools distributed on Internet (as well as other digital corpora).
The design of tools for HPST researchers has to integrate and/or aggregate the existing heterogeneous tools and to ensure interoperability among them. Thus, the goal is not to build a single new environment, but to design a platform which integrates existing tools selected for their relevance according to the practices of researchers and provide an agile architecture able to model and/or support the processes involved in the research work and enrichment.
This platform will be mainly based on the Semantic Web and Linked Data approaches (RDF Triple Store, ontologies, OWL 2, RDFS, SPARQL, etc.). Nevertheless, the platform will also provide access to non-semantic resources. A network of ontologies dedicated to HPST will be designed to meet the interoperability and open access requirements for corpora. Some existing ontologies and standards will be reused and integrated in the ontology network, like CIDOC-CRM, FRBRoo, FRSAD, Dublin Core, etc. and those available at LOV (http://lov.okfn.org/dataset/lov/).
In this paper, we focus our attention on the resource retrieval problem that we can divide into two di erent aspects : advanced search in HPST corpora and harvesting distributed corpora. The former focuses on advanced search function11 The alambic numerique (http://alambic.hypotheses.org/4924) is based on Omeka. alities in a single corpus. The latter studies the resource retrieval on distributed corpora. These two aspects will be integrated. 3.1
In order to perform advanced searches in a HPST corpus, we have to build intelligent digital corpus: corpus with primary and secondary sources having semantic metadata (RDF Triples) and their corresponding ontologies using a fragment of OWL (actually, RDFS will be su cient for the following examples). These ontologies are domain ontologies related to the corpus. A domain ontology for Henri Poincare letters has already been designed. Finally, some tools will have to be developed for answering some of the queries.
This section presents the advanced search using the query examples Q3-Q6 introduced in Section 2.2.
Q3 requires some additional data and knowledge to get satisfactory answers, as stated in Section 2.2. In particular, if the annotation le contains the following RDFS triples: (letter1 isSentBy rodolphe radau) (rodolphe radau rdf:type Geodesist) (Geodesist rdfs:subClassOf Astronomer) then the execution of the following SPARQL query on an engine supporting RDFS
Q3 = select ?` where f?` isSentBy ?a . ?a rdf:type Astronomerg will return letter1.
Q4, similarly, can be answered by a SPARQL engine supporting RDFS with the following query:
Q4 = select ?` where ?` isAnAnswerTo ?`2 .
?`2 isSentBy mittag-leffler It can be noticed that this query can give a letter of the corpus that answers a lost letter: the missing letter cannot be found, but its answer can.
Q5, for being executed, requires the use of annotations about the scienti c content of the letter:
Q5 = select ?` where ?` hasForTopic ?t .
?t rdf:type N-body-problem The n-body problem is a topic having sub-topics, in particular, the 3-body problem is a problem more speci c than the n-body problem. For this reason, we have chosen to model these two problems by two classes, the former being more general than the latter. Therefore, a letter of the corpus about the 3-body problem will be returned by the execution of this query.12 12 We could also have chosen to model the 3-body problem as an instance of the nbody problem, but rst, it is more homogeneous to consider every topic as a class,
Q6 can be modeled by a SPARQL query based on the assumption that \the late xixth century" corresponds to the interval 1881 1900:
Q6 = select ?` where ?` sentDuringYear ?y .
filter(?y >= 1881 && ?y <= 1900) However, this solution is debatable: the modeling of the fuzzy period of time by a crisp interval raises the problem of the choice of the boundaries. Indeed, some events before 1881 or after 1900 can be considered by historians to be related to the end of the xixth century. In order to address this issue, some approximate search is planned. How to put this idea in practice is an ongoing work. 3.2
Harvesting distributed corpora at semantic level (according to Linked Data principles) require to solve two di erent problems. The rst one is to queries several triple store by means of federated queries to linked distributed sources. The second one is to get RDF triples from social media tools.
Most of social media applications are data silos. In other words, data are unavailable on the web. Only people may have access to data, not computers. Reuse and exchange of data among social media tools are only possible by means of API { that is to say manually by mean of one API per tool. Some social media tools like Drupal, Semantic media wiki may have their own triple store exposing data to others.
A toolkit, called SMOOPLE for Semantic Massive Open Online Pervasive
Learning Environment, has been designed to solve these two problems. It was
rstly dedicated to the technology-enhanced learning domain [
SMOOPLE has semantic services which are in charge of managing incorporated semantic models, extracting and storing the data produced on social media tools, making and answering to semantic queries against one or several distributed sources (federated queries). The Semantic Web server (semantic services) is based on Jena 2. When the social media tools do not have a triple store and a SPARQL endpoint, content and corresponding semantic metadata can be extracted on the y from social media applications, by means of plugin (similar to sioc export) and stored in a RDF repository. Several light ontologies (SIOC, FOAF, DC, RDF, RDFS, etc.) are used to acquire semantic metadata automatically. It will be necessary to de ne the interlinkage among distributed sources (triple stores) to support federated queries.
second, this way, it is always possible to consider a more speci c topic, e.g., the restricted 3-body problem for which the mass of one of the 3 bodies in considered to be negligible.
An aim of the SemanticHPST project is to focus on the epistemological issues raised by the development of these new tools based on semantic web. This work in progress takes part to epistemological questions in the domain of Digital Humanities.13 A rst series of questions concerns the modeling of knowledge, the main step in building ontologies so that researchers can easily identify and apprehend knowledge. Therefore the creation of e ective ontologies requires de ning concepts and elucidating certain tacit or implicit knowledge. So the initial questions are: How to approach these de nitions? How to ensure that indexing does not immobilize knowledge? How can the modeling anticipate how it will be used in order to ensure that the knowledge generated is contextualized to avoid anachronism and misinterpretation? Moreover, the wide range of works in the collection, including texts (manuscripts, books, letters, web pages), multimedia documents, 3D archaeological or historical objects and media from a variety of sources (photographs, original texts, maps, etc.), necessitate di erent approaches. This raises the question: How to approach a photograph, a scienti c instrument or a text and still obtain a uni ed ontology? How can the modeling enable relationships between objects yet avoid the pitfalls described above?
In the eld of HPST, the issue of modeling time is central and particularly
tricky. Modeling a long period of time, an event, a succession of events or events
that are juxtaposed requires making decisions that should be taken collectively.
Indeed, this emerging issue is shared by historians [
A second series of questions concerns the researcher's environment, which
has signi cantly changed with the rise of digitized data. Whatever the works
considered or their origin (libraries, archives, etc.), the massive volume of data,
its diversity and location are all part of this change. Yet this radical shift is
not exclusively the result of the accumulation of a large amount of data. The
fact that data can be `analyzed as well as communicated, represented, reused {
in short, mobilized for research { in a quantity and with an ease incomparable
with previous periods' [
The aim of this proposal is to contribute to the development of the research in the domain of digital humanities. Based on the Semantic Web principles and technologies, the SemanticHPST group proposes new methodologies in History and Philosophy of Science and Technology in the framework of a strong collaboration between labs working in the area of computer science and humanities (here HPST). The main goal is to enrich the practices of researcher and communities in HPST as well in science and technology heritage. To deal with such a goal, the project has to: i) Build intelligent digital corpora, that is to say corpora with primary and secondary sources having semantic metadata and their corresponding ontologies; ii) Design tools to access and enrich existing corpora and to create new ones; iii) Evaluate the resulting evolution of practices in historical science and build an epistemological viewpoint about the impact of new tools and practices in humanities based on knowledge modeling and semantic web.
Another important issue is to deal with the reuse of intelligent digital corpora. Thus, it is necessary to build representations of the entities, people and processes involved in producing the digital corpora. The \PROV Model Primer" from W3C (http://www.w3.org/TR/prov-primer/) can be used to address this issue.