Natural language research announcements

[Margaret Fleck <mfleck AT cs.uiuc.edu>]

-------- Original Message --------
Subject: Fwd: LPBB next week
Date: Fri, 20 Apr 2007 10:21:33 -0500
From: Richard Sproat 
<rws AT XOBA.COM>
Reply-To: 
rws AT uiuc.edu
To: 
LING-DEPT-L AT LISTSERV.UIUC.EDU
References: 
<8CCFAB3F-696B-434B-99D7-A1229925AE7D AT uiuc.edu>

---------- Forwarded message ----------
From: Sarah Brown-Schmidt 
<brownsch AT uiuc.edu>
Date: Apr 20, 2007 10:00 AM
Subject: LPBB next week
To: Sarah Brown-Schmidt 
<brownsch AT uiuc.edu>

Dear Language Processing Brown Bag,

I hope you can all make it to next weeks FINAL LPBB of the semester!  Our
speaker will be Richard Sproat. The talk will be next Thurday, April 26th at
12:30 in room 4269 of the Beckman Institute. I hope to see you there.

Talk title and abstract are below.

-Sarah Brown-Schmidt


Decipherment, Pseudodecipherment and the Phaistos Disk

Richard Sproat

The Phaistos Disk, discovered in Crete in 1908 and thought to date to
between 1600 and 1850 BC, is one of the great outstanding
archaeological enigmas. It has a number of unique features, the most
puzzling of which is what appears to be a text of 241 glyphs composed
on both sides in an otherwise unknown script. Since its discovery
there have been no fewer than 20 published "decipherments", ranging in
quality from intriguing to truly bizarre.

In this talk I will present a variant of a proposal by Kevin Knight
for a computational model of script decipherment, and use this to
argue that in the absence of another sample of this writing system, it
is simply pointless to try to decipher this artifact or any artifact
with such a short text. In particular, when we consider the way that
known Aegean scripts such as Linear B encode language and if we make
similar assumptions for the Phaistos disk, there are simply too many
internally consistent possibilities, no matter what language we assume
it encoded. Using relatively simple computational models involving
Finite State Transducers, and standard language modeling techniques,
it is possible to generate large numbers of candidate "translations"
on short order.

As part of the presentation, I will give necessary background on what
is known about Aegean scripts and how they encoded Archaic Greek. This
should be of interest both to linguists and psycholinguistics
interested in reading since the way Linear B, for example, encoded
Greek involved some rather extreme phonological simplifications in
mapping between the phoneme sequences and the corresponding graphemes.

I will also examine and critique a couple of the more recent proposed
decipherments for the Phaistos disk in some detail.



--
Richard Sproat
Dept of Linguistics
  Dept of Electrical and Computer Engineering
   Adjunct, Dept of Computer Science
      Adjunct, Dept of Psychology
        Beckman Institute
University of Illinois at Urbana-Champaign

Mail Address:
Department of Linguistics
Foreign Languages Building 4016D,
707 South Mathews Avenue, MC-168
Urbana, IL, 61801
Phone: +1-217-244-4120 (Beckman, Rm. 2057)
Fax:   +1-217-244-8430 (Linguistics)
rws AT uiuc.edu
----------------http://www.linguistics.uiuc.edu/rws/ -----------------------

I do not mean everything's under control.
That would not be cool and froody.