PhD Thesis on "Quantum Computation and Natural Language Processing"

[Joseph Chen]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Dear Connectionists,

The PhD thesis:  "Quantum Computation and Natural Language Processing"
is available at

http://www.sub.uni-hamburg.de/opus/volltexte/2002/769/pdf/ 
dissertation.pdf

This may be of interest to some of you.

Best Regards,

Joseph Chen

====== Abstract ======
In this thesis, a novel approach to natural language understanding  
inspired by quantum mechanical principle is proposed. It is based on an  
analogy between the physical objects at the quantum level and human's  
mental states. In this way, the physical and the mental phenomena are  
to be understood within the same framework. It is also proposed that  
the apparent differences between mind and matter do not lie in the  
fundamental differences of their properties, but in the different  
manifestation of macroscopic matter and macroscopic mind owing to their  
different composition of pure quantum eigenstates. The apparent  
differences are therefore quantitative rather than qualitative.

Specifically, symbols in various cognitive functions are to be treated  
as eigenstates with respect to a particular quantum experimental  
arrangement. Moreover, I claim that reasoning and inference can be  
treated as transformations of semiosis with symbols being the  
eigenstates of a particular formulation operator. The operator is the  
counterpart of an observable in quantum mechanics. A state of affairs  
(a superposition of these eigenstates) does not have well-defined  
physical properties until it is actually measured. Consequently the  
classical semantics (as classical symbols' referring to the classical  
physical reality) is also not well-defined and may be a misleading  
idea. Different from classical semantics, meaning in the quantum  
mechanical framework should be treated as an active measurement done on  
a state of affair.

Moreover, the ill-definedness also manifests itself in the cognition  
internal to a person if we regard memory as a language-like  
representational system. Nevertheless, memory, treated as a specific  
language system, is a largely quasi-classical phenomenon in that the  
chemical activities in the brain are an aggregate limiting case of  
quantum mechanics with a very large number of quanta. The classical  
``objective'' physical reality is therefore a limiting case of quantum  
reality as well.

The general language in which common sense logic is embedded is then  
investigated and the apparent evasiveness and ambiguity of language can  
be accommodated in a quantum framework. This is done by postulating an  
analogous Uncertainty Principle and observing the implication of it.   
An important implication is the ``concept-symbol'' duality. As  
applications, the quantum mechanical formalism is applied to cognitive  
processes. For instance, non-monotonicity and counterfactual  
conditionals can be accommodated and assimilated in this framework.  
Specifically, the time-asymmetric property and the genuine unknown  
state of non-monotonic reasoning can be easily explained in quantum  
mechanics. This is also the case for the potentiality and actuality,  
which are crucial ideas for explaining counterfactual reasoning.  
Furthermore, causality can be regarded as a disguise of counterfactual  
reasoning.

The second part of the thesis is devoted to simulations and technical  
applications of the aforementioned principle in natural language  
processing. First the preliminary experiments of common sense logic are  
presented. These show that the ``classicization'' of common sense logic  
can be implemented with very simple quantum mechanical systems.  
Moreover, the richness of the quantum framework goes well beyond what a  
classical system can offer. There can be ``fine-structures'' within  
seemingly simple logical arguments (XOR, for example). This is also the  
case for non-monotonic and counterfactual reasoning.

Simple natural language tasks are also simulated based on different  
natural language corpora. First the syllogistic arguments embedded in  
natural language are simulated with a quantum system, which delivers  
quite remarkable results. Secondly, a monolingual syntax manipulation  
is implemented with a quantum system, in which the quantum mechanical  
approach can achieve much better performance than connectionist one. In  
the last experiment, a quantum mechanical architecture is trained for  
bilingual translation between English and German, in which there are  
several thorny properties in the natural language corpus, for example  
lexical ambiguity, separable prefixes, complicated conjugation, and  
non-linear translational word mappings. Nevertheless, the quantum  
mechanic architecture can deliver very satisfactory results.

- --

Dr. Joseph C.H. Chen

Computer Science Department,
University of Hamburg

Vogt-Koelln-Str. 30
22527 Hamburg

http://nats-www.informatik.uni-hamburg.de/~joseph/
+49-40-42883-2523 (O)
+49-40-42883-2515 (FAX)

PGP Public Key:
http://nats-www.informatik.uni-hamburg.de/~joseph/joseph.pgp.asc

Printed on Planet Earth and with 100% recycled electrons.
- ----------------------------------------------------------------
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.1 (Darwin)

iD8DBQFAPv6ANE0t/Ns3SDgRAtPlAJ9NMIfBPlY+aLRYnD4xyxqcp5xdUQCeO1hK
NOwwv80fRi4TcJtqfdHIxQ0=
=XcFl
-----END PGP SIGNATURE-----