Clustering and Relational Ambiguity: from Text Data to Natural Data

Text data is often seen as “take-away” materials with little noise and easy to process information. Main questions are how to get data and transform them into a good document format. But data can be sensitive to noise often called ambiguities. Ambiguities are aware from a long time, mainly because polysemy is obvious in language and context is required to remove uncertainty. One claim in this paper is that syntactic context is not sufficient to improve interpretation. This paper tries to explain that firstly noise can come from natural data themselves, even involving high technology, secondly texts, seen as verified but meaningless, can spoil content of a corpus; it may lead to contradictions and background noise. We used a set of papers in biology to identify ambiguous facts related to human interpretation and a nearest-neighbour display associated to a Zipfian distribution to compare structural content of a corpus. Four kinds of discourse technical, general, short-communication and artificial have been studied. keywords computational linguistics; paradox; contradiction; ambiguity; semantic relationship; domain; information extraction; corpus INTRODUCTION Human cognition refers to a diversity of concepts such as memory and brain anatomy, inference and reasoning, motivation, time and space, classification and clustering. Inference tries to identify good relations or properties associated to an object. In this sense it is also possible to test validity or consistency of a relation. Let be the proposition P = “a cat is a stone” is false or contradictory because a stone is not a living organism, though a cat is a living organism. P can be called paradoxal or contradictory. Sometimes, society lives with contradictions such as tolerance to many deaths on roads or in wars but intolerance for death from diseases. In this paper we more specifically focus on sources of potential contradictions that could spoil computation of information extraction. Formal semantics is attached to validate relations between a set of objects. Our study focuses on issues in managing complexity of a logical proposition and how to compute its truth value, but we also study how to extract relations and see how they are asserted as non contradictory with regard to other relations extracted in other texts. Thus texts are the primary material of discussion. Chapter 1 presents relational ambiguities we can find in text. We start by presenting a typology of logical relations. Given a type of relations, we explain how to extract such relations with markers in text files. But markers are not sufficient to detect a contradiction. A specialized language such as molecular corpus provides an example of ambiguous relations (contradictory) that cannot be detected with markers. Hence, we show that a global overview of words collocations in a corpuscan give a good signal about the structure. In our “publish or perish” new era of research and development system, production of literature is high but a non-negligible percent of papers becomes false over time. It is possible to compile from the