Term Discovery in an Early Modern Latin Scientific Corpus

This paper presents the results of a pilot project aimed at
the development of automatic techniques for the discovery
of salient technical terminology in a corpus of Latin texts.
These texts belong to the domain of mechanics and date
from the last quarter of the sixteenth and the fi rst quarter
of the seventeenth century, a period of intense intellectual
activity in which engineers and scientists explored the limits
of the Aristotelian and Archimedean paradigms in mechanics.
The tensions that arose ultimately were resolved by the new
“classical mechanics” inaugurated by Newton’s Principia in
1687 (cf. Damerow et al. 2004).
The work presented here forms part of a larger research
project aimed at developing new computational techniques to
assist historians in studying fi ne-grained developments in longterm
intellectual traditions, such as the tradition of Western
mechanics that begins with the pseudo-Aristotelian Problemata
Mechanica (ca. 330 B.C.E.). This research is integrated with two
larger institutional projects: the working group “Mental Models
in the History of Mechanics” at the Max Planck Institute for
the History of Science in Berlin, and the German DFG-funded
Collaborative Research Center (CRC) 644 “Transformations
of Antiquity.”
The purpose of this paper is to present initial results regarding
the development of effi cient methods for technical term
discovery in Early Modern scientifi c Latin. The focus is on the
identifi cation of term variants and on term enrichment. The
methodology employed is inspired by Jacquemin (2001), whose
approach allows for the use of natural language processing
techniques without the need for full syntactic parsing, which is
currently not technically feasible for Latin.
The present paper extends prior work in term discovery
along two vectors. First, work in term discovery has primarily
addressed the languages of Western Europe (especially
English and French), with some work also in Chinese and
Japanese. Latin presents some typological features that require
modifi cations to established techniques. Chief among these is
the rich infl ectional morphology (both nominal and verbal)
of Latin, which is a language of the synthetic type. Latin also
exhibits non-projectivity, i.e. syntactic constituents may be
represented non-continuously (with the intrusion of elements
from foreign constituents). Although the non-projectivity of
Renaissance Latin is considerably less than what is found in the
artistic prose (and a fortiori poetry) of the Classical language
(Bamman and Crane 2006), term detection must proceed
within a framework that allows for both non-projectivity and
(relatively) free word order within constituents. Second, researchers in the fi eld of term discovery have focused
almost exclusively on contemporary scientifi c corpora in
domains such as biomedicine. In contemporary scientifi c
literature, technical terms are characterized by a particularly
high degree of denotative monosemicity, exhibit considerable
stability, and follow quite rigid morphological, syntactic, and
semantic templates. Although these characteristics are also
applicable to the terminology of Latin scientifi c texts, they are
applicable to a lesser degree. In other words, the distinction
between technical terminology and ordinary language
vocabulary is less clear cut than in the case of contemporary
scientifi c and technical language. The lesser degree of
monosemicity, stability, and structural rigidity of terminology
holds implications for automatic term discovery in corpora
earlier than the twentieth (or at least nineteenth) century.
The corpus of Early Modern mechanics texts in Latin is welldesigned
for carrying out experiments in adapting established
techniques of term discovery to historical corpora. Mechanics
is by this time a scientifi c discipline that possesses an extensive
repertoire of characteristic concepts and terminology. Thus
it is broadly comparable to contemporary scientifi c corpora,
while still presenting unique features that merit special
investigation. Several thousand pages of text are available in
XML format, which have been digitized by the Archimedes
Project, an international German/American digital library
venture jointly funded by the DFG and NSF. It will be possible
to extend future work to a multilingual context, by examining
in addition closely-related vernacular works (in Italian, Spanish,
and German) that are contemporary with the Latin corpus.
(Some of these are translations and commentaries.)
The set of technical terminology discovered by the
methods presented in this paper is intended to further the
computationally-assisted framework for exploring conceptual
change and knowledge transfer in the history of science
that has been described by Hyman (2007). This framework
employs latent semantic analysis (LSA) and techniques for
the visualization of semantic networks, allowing change in the
semantic associations of terms to be studied within a historical
corpus. The concluding section of the present paper will survey
the applications of technical term discovery within historical
corpora for the study of the confl ict, competition, evolution,
and replacement of concepts within a scientifi c discipline and
will suggest potential applications for other scholars who are
concerned with related problems.
References
Bamman, D. and G. Crane. 2006. The design and use of a Latin
dependency treebank. In Proceedings of the TLT 2006, edd. J.
Hajič and J. Nivre, Prague, pp. 67–78.
Damerow, P., G. Freudenthal, P. McLaughlin, J. Renn, eds.
2004. Exploring the Limits of Preclassical Mechanics: A Study
of Conceptual Development in Early Modern Science: Free Fall
and Compounded Motion in the Work of Descartes, Galileo, and
Beeckman. 2d ed. New York.
Hyman, M.D. 2007. Semantic networks: a tool for investigating
conceptual change and knowledge transfer in the history of
science. In Übersetzung und Transformation, edd. H. Böhme, C.
Rapp, and W. Rösler, Berlin, pp. 355–367.
Jacquemin, C. 2001. Spotting and Discovering Terms through
Natural Language Processing. Cambridge, MA.