Akademie Greifswald
130
Arbeitsgruppe 3
The Concept of Maximum Entropy: Applications in
Language Processing and Trading
Leitung
Dr. Jan HendrikWitte
Mathematical Institute, University of Oxford
Teilnehmer
insbesondere Studierende der Fächer mit naturwissenschaftlicher
Ausrichtung; ein solides Verständnis von Schulmathematik und
ein Interesse an Finanzmärkten sind notwendig
In automatic language processing (or ‘translation’), probabilistic patterns have to be iden-
tified by a statistical algorithm; once a space of possible patterns has been constructed and
probabilities have been allocated, the most likely translation of a given word or phrase can
be chosen. (Hence, in a sense, a translation engine such as google-translate simply aims to
replicate the way in which also the human mind learns a new language: by steady observa-
tion and training).
We will develop the concept of maximum entropy and natural language processing based
on [1]. In a first step, we will aim to develop a step by step characterisation of a statistical
translation algorithm. Once accomplished, we will proceed to consider pattern identifica-
tion in financial markets and develop adaptations of our algorithm to trading.
