Computer Assisted Conceptual Analysis of Text: the Concept of Mind in the Collected Papers of C.S. Peirce

Computer assisted reading and analysis of text (CARAT)
has recently explored many variants of what has become
fashionable to call “text mining” strategies. Text mining
strategies are theoretically robust on large corpus. However,
since they mainly operate at a macro textual level, their use is
still the object of resistance by the expert readers that aim at
fi ne and minute conceptual analysis. In this paper, we present
a computer assisted strategy for assisting conceptual analysis
based on automatic classifi cation and annotation strategies.
We also report on experiment using this strategy on a small
philosophical corpus.
Conceptual analysis is an expert interpretation methodology
for the systematic exploration of semantic and inferential
properties of set of predicates expressing a particular concept
in a text or in a discourse (Desclés, 1997; Fodor, 1998;
Brandom, 1994; Gardenfors, 2000; Rastier, 2005). Computer
assisted reading and analysis of text (CARAT) is the computer
assistance of this conceptual analysis.
The strategy of CACAT
Our text analysis strategy rests on the
following main hypothesis:
The expression of a canonical concept in a text presents
linguistics regularities some of which can be identifi ed using
classifi cation algorithms
This hypothesis itself unwraps into three sub hypothesis:
Hypothesis 1: conceptual analysis can be realized by the
contextual exploration of the canonical forms of a concept
This is realized through the classical concordance
strategy and variances on a pivotal term and its linguistic
variants (e.g. mind, mental, mentally, etc.) (Pincemin
et al., 2006; McCarthy, 2004; Rockwell, 2003). Hypothesis 2: the exploration of the contexts of a concept is itself
realized through some mathematical classifi cation strategy.
This second hypothesis postulates that contexts of
a concept present regularities that can be identifi ed
by mathematical clustering techniques that rest
upon similarities found among contextual segments
(Jain et al. 1999; Manning and Schütze, 1999).
Hypothesis 3: Classes of conceptual of similar conceptual
contexts can be annotated so as to categorize their semantic
content.
This last hypothesis allows to associate to each segment of a
class of contexts some formal description of their content be it
semantic, logical, pragmatic, rhetorical, etc. (Rastier et al., 2005;
Djioua and Desclés, 2007; Meyers, 2005; Palmer et al., 2005;
Teich et al., 2006). Some of these annotations can be realized
through algorithms; others can only be done manually.
Experiment
From these three hypotheses emerges an experiment which
unwraps in fi ve phases. This experiment was accomplished
using C.S. Peirce’s Collected Papers (volumes I-VIII) (Peirce,
1931, 1935, 1958). More specifi cally, this research aimed
at assisting conceptual analysis of the concept of “Mind” in
Peirce’s writings.
Phase 1: Text preparation
In this methodology, the fi rst phase is text pre-processing.
The aim of this fi rst phase is to transform the initial corpus
according to phases 2 and 3 requirements. Various operations
of selection, cleaning, tokenisation, and segmentation are
applied. In the experiment we report on, no lemmatisation or
stemming was used. The corpus so prepared was composed of
74 450 words (tokens) with a lexicon of 2 831 word types.
Phase 2: Key Word In Context (KWIC)
extraction (concordance)
Using the corpus pre-processed in phase 1, a concordance
is made with the pivotal word “Mind”. The KWIC algorithm
generated 1 798 contextual segments of an average of 7 lines
each. In order to be able to manually evaluate the results of the
computer-assisted conceptual analysis, we decided to select in
the project only a random sampling of the 1 798 contextual
segments. The sampling algorithm delivered 717 contextual
segments. This sample is composed of 3 071 words (tokens)
and 1 527 type words.
Phase 3: KWIC clustering
The concordance is in itself a subtext (of the initial corpus). A
clustering technique was applied to the concordance results. In
this project, a hierarchical agglomerative clustering algorithm
was applied. It generated 83 clusters with a mean 8.3 segments
per class. It is possible to represent spatially the set of words
in each class. Figure 1 illustrates such a regrouping for cluster
1.
Figure 1. Graphical representation of cluster 1 lexicon.
It is often on this type of representation that many numerical
analyses start their interpretation. One traditional critic
presented by expert analysts is their great generality and
ambiguity. This kind of analysis and representation give hints
on the content of documents, but as such it is diffi cult to use
for fi ne grained conceptual analysis. It must hence be refi ned.
It is here that the annotation phase comes into play.
Phase 4: Annotation
The annotation phase allows the expert reader to make more
explicit the type of information contained in each clusters
(generated in phase 3). For instance, the interpreter may indicate
if each cluster is a THEME, a DEFINITION, a DESCRIPTION,
an EXPLANATION, an ILLUSTRATION, an INFERENCE, or
what is it MODALITY (epistemic, epistemological, etc.). The
variety of annotation types is in itself a research object and
depends on various textual and linguistic theories.
Annotation results
In this abstract, size constraints do not allow us here to present
detailed results of classifi cation and annotation processes.
We shall only present a sample on a few segments of three
classes.
Annotations of cluster 1: The fi rst cluster contained 17 segments
all of which have received an annotation. Here are samples of
annotation for two segments of cluster 1. The annotation is
preceded by the citation itself from the original text.
[SEGMENT NO 512]
“Finally laws of mind divide themselves into laws of the universal
action of mind and laws of kinds of psychical manifestation.”
ANNOTATION: DEFINITION: the law of mind is a general
action of the mind and a psychological manifestation [SEGMENT NO 1457]
“But it differs essentially from materialism, in that, instead of
supposing mind to be governed by blind mechanical law, it
supposes the one original law to be the recognized law of mind,
the law of association, of which the laws of matter are regarded
as mere special results.”
ANNOTATION: EXPLICATION: The law of mind is not a
mechanical materialism.
Phase 5: Interpretation
The last phase is the interpretative reading of the annotations.
Here, the interpreter situates the annotated segments into
his own interpretative world. He may regroup the various
types of annotation (DEFINITIONS, EXPLANTIONS, etc.)
and hence build a specifi c personal data structure on what
he has annotated. From then on, he may rephrase these in his
own language and style but most of all situate them in some
theoretical, historical, analytical, hermeneutic, epistemological,
etc. perspective. It is the moment where the interpreter
generates his synthesis of the structure he believes underlies
the concept.
We present here a sample of the synthesis of conceptual
analysis assisted by the CARAT process on cluster 1 (the
concept of “mind” in C.S. Peirce’s writings – cluster 1).
The law of Mind: association
The Peircian theory of MIND postulates that a mind is governed
by laws. One of these laws, a fundamental one, is associative
(segment 512). This law describes a habitus acquired by the
mind when it functions (segment 436).
Association is connectivity
This functioning is one of relation building through connections.
The connectivity is of a specifi c nature. It realizes a synthesis (à
la Kant) which is a form of “intellectual” generalisation (segment
507).
It is physically realized
Such a law is also found in the biological world. It is a law that
can be understood as accommodation (segment 1436). In
fact, this law is the specifi c form of the Mind’s dynamic. It is a
fundamental law. But it is not easy for us to observe it because
we are victim of a interpretative tradition (segment 1330) that
understands the laws of mind as laws of nature. This is a typical
characteristic of an “objective idealism” (segments 1762 and
1382). The laws of mind do not belong to mechanist materialism
(segments 90 and 1382).
And there exist a variety of categories
There exist subdivisions of this law. They are related to the
generalisation process that is realised in infanthood, education,
and experience. They are intimately related to the growth of
consciousness (segments 375 and 325).
Conclusion
This research project explores a Computer-Assisted Reading
and Analysis of Text (CARAT) methodology. The classifi cation
and annotation strategies manage to regroup systematically
segments of text that present some content regularity. This
allows the interpreter to focus directly on the organized
content of the concept under study. It helps reveal its various
dimensions (defi nitions, illustrations, explanations, inferences,
etc.).
Still, this research is ongoing. More linguistic transformations
should be applied so as to fi nd synonymic expressions of a
concept. Also, various types of summarization, extraction and
formal representation of the regularities of each class are to
be explored in the future.
But the results obtained so far reinstate the pertinence of the
concordance as a tool for conceptual analysis. But it situates it
in a mathematical surrounding that aim at unveiling the various
dimensions of a conceptual structure. Most of all, we believe
that this methodology may possibly interest expert readers
and analysis for it gives a strong handle and control on their
interpretation process although assisting them throughout the
process.
References
Brandom, R.B. (1994). Making it Explicit. Cambridge: Harvard
University Press.
Desclés, Jean-Pierre (1997) “Schèmes, notions, predicats et
termes”. Logique, discours et pensée, Mélanges offerts à Jean-
Blaize Grize, Peter Lang, 9-36. 47.
Djioua B. and Desclés, J.P. (2007), “Indexing Documents
by Discourse and Semantic Contents from Automatic
Annotations of Texts”, FLAIRS 2007, Special Track “Automatic
Annotation and Information Retrieval : New Perspectives”,
Key West, Florida, May 9-11.
Fodor, J. (1998) Concepts: Where Cognitive Science Went Wrong.
Oxford: OUP.
Gardenfors, P. (2000) Conceptual Spaces. Cambridge (Mass.):
MIT Press.
Jain, et al. (1999). Data Clustering: A Review. ACM Computing
Surveys, 31(3):264–323.
Manning, C. and Schutze, H. (1999) Foundations of Statistical
Natural Language Processing, Cambridge Mass. : MIT Press.
Meyers, Adam (2005) Introduction to Frontiers in
CorpusAnnotation II Pie in the Sky Proceedings of the
Workshop on Frontiers in Corpus Annotation II: Pie in the Sky,
pages 1–4, New York University Ann Arbor, June 2005. McCarthy, W. (2004) Humanities Computing, Palgrave
MacMillan Blackwell Publishers.
Palmer, M., Kingsbury, P., Gildea, D. (2005) “The Proposition
Bank: An Annotated Corpus of Semantic Roles”.
Computational Linguistics, vol. 31, no 1, pp. 71-106.
Peirce, C.S. (1931-1935, 1958), Collected Papers of Charles
Sanders Peirce, vols. 1–6, Charles Hartshorne and Paul Weiss
(eds.), vols. 7–8, Arthur W. Burks (ed.), Harvard University
Press, Cambridge, MA, 1931–1935, 1958.
Pincemin, B. et al. (2006). Concordanciers: thème et
variations, in J.-M. VIPREY (éd.), Proc. of JADT 2006, pp. 773-
784.
Rastier, F. (2005) Pour une sémantique des textes théoriques.
Revue de sémantique et de pragmatique, 17, 2005, pp. 151-180.
Rastier, F. et al. (eds) (1995) L’analyse thématique des données
textuelles: l’exemple des sentiments. Paris: Didier Érudition.
Rockwell, G. (2003) What is text analysis, really? Literary and
Linguistic Computing, 18(2): 209–219.