What the heck is "Common Reference"?

Mon May 11 05:11:04 UTC 2009

To the list:
I've been asked in a couple of private posts where I got the terminology 
"Common Reference" and what kind of peer-reviewed source it came from.   Valid 
question.

I suspect it's not difficult to understand what "common reference" means in 
the context of a lot of linguistic terminology (though not all).   

"Reference" here means the object, property, process, in time and place, 
that a word or other aspect of language stands for, symbolizes or refers to.   
This is reference as in "sense and reference" (Sinn und Bedeutung) pretty 
much as Frege defined it back in the 1890s.   The reference a word (or other 
parts of language structure) makes corresponds to some thing or some process 
or situation in the world, the universe or alternative universes or 
imaginary worlds.   It can even be a reference to other words in the same sentence, 
but that's in no way a limitation on the broader sense in which "reference" 
is used here.

"Common" here means shared in common with other people.   It does not mean 
here frequently or commonly occuring or average.   Very simply put, when two 
people use the same word to stand for the same thing, that's Common 
Reference.

Now there are all kinds of little logic games that you can play to show 
that no two people can ever share the same reference, but for now I think most 
of us can confirm that common reference works pretty well for most of us 
most of the time.   If I say, "Bring me the book", chances are pretty good that 
the person I am talking to will pick up a physical book, carry it over and 
hand it to me -- without falling into a metaphysical vortex where an 
infinite of potential meanings collide.   The problem that shared references can 
only be approximations doesn't mean they don't work, but rather tell us how 
they work to change language.

You'll find "common reference" used in the way descrbed here, of course, in 
many disciplines and institutions that feel the need for standardized 
specialized terminology among its members, especially among doctors, engineers 
and computer scientists.   So that everybody uses the same terms to refer to 
the same things.   This usage of "common reference" simply takes the same 
sense and applies it beyond specialist terminology to all language.

The IMPORTANT point here is that common reference is NOT 100% present in 
any instances of use of language.   The idea is instead that it is the most 
fundamental objective in all language -- and in that it is the closest we come 
to a universal in language.  Among language users, common reference is not 
a framework, it is an on-going quest that creates frameworks. 

If I ask -- in English -- a person who only speaks Mandarin to "Bring me 
the book", we can expect an absence of common reference.   The sounds I've 
made do not correspond to objects or processes I am referring to, in the 
language of my listener.   If i wish to change that circumstance, I will need to 
find a way to share all or most of my references with that listener.   On a 
much less extreme level, that may be the task with every sentence we speak, 
even every utterance we make.

There is at least one field of research that is supporting the idea that 
finding common reference is a pervasive problem throughout language use.    If 
you try to build a robot that uses human language, and make syntactic   
structure what happens first, you get a very confused and confusing robot.   
But if you make a robot's first job finding common reference, you actually 
start to see a robot developing its own pragmatics. 

What's led me to the use of the "common reference" is my time to time 
involvement with businesses involved in interactive robotics.   In this field, 
there has been a very long but most convincing drift towards this idea, or 
ideas very much like it, to describe the first and recurring steps an 
artificial intelligence must take in the real world to use human language.   A lot of 
this trend is going on in a commercial context which does not generate a 
lot of publicly published findings, but some has made its way into the public 
dialogue.

An incredibly enlightening paper regarding robotic interactive language use 
-- and one that defines the term "common reference" operationally if not 
formally is:
Yoko YAMAKATA, Tatsuya KAWAHARA, Hiroshi G. OKUNO and Michihiko MINOH: “
Belief Network based Disambiguation of Object Reference in Spoken Dialogue 
System”, Transactions of the Japanese Society for Artificial Intelligence, 
Vol. 19, No. 1, pp.47-56 (2004).The paper can be downloaded at:
http://www.jstage.jst.go.jp/article/tjsai/19/1/19_47/_article/-char/en

This paper has been very influential among a number of human-robotic 
language developers because it provided a way of overcoming the hurdle of how 
robots should be designed to respond to the starting block problem of obviously 
non-human, pre-programmed language generation.   (The paper is one reason 
that I and some others came to use the term "common reference", although 
terminology in this field is hardly uniform, because a lot of the concept naming 
is private.)

Here's part of the abstract: "... In addition to the ambiguity of the 
object reference, the actual system must cope with two sources of uncertainty: 
speech and image recognition. We present the belief network based 
probabilistic reasoning system to determine the object reference. The resulting system 
demonstrates that the number of interactions needed to find a COMMON 
REFERENCE is reduced as the user model is refined." (caps added)    The word 
"belief" here is actually a quantitive system of measuring confidence levels by the 
robot, as it restructures its language based on human responses.   The 
robot's success in all interactions in these tests were enhanced by anything 
that increased common reference.

The term common reference appears in a number of other papers where 
robotics has attempted functioning use of human language.

Bauer A, Gonsior B, Wollherr D, Buss M. Heuristic Rules for Human-Robot 
Interaction Based on Principles from Linguistics - asking for directions, AISB 
(2009).   The paper can be downloaded at:
www.lsr.ei.tum.de/fileadmin/publications/aisb09_bauer.pdf

In this research, the term "common reference" turned from a question of 
spatial coordinates to one of linguistic rules.   The authors cite Karl 
Buhler's The Theory of Language: The Representational Function of Language (1934), 
but don't use Buhler's broad definition of Deixis -- following instead 
Levinson and others who limit deitic to a narrower category of words and 
constructions.   That the rules had to be heuristic simply reflects the fact that 
simple syntactic or semantic rules would not solve the real-world tasks the 
robots were given.

Finally, there's also the work of Luc Steels, who has been associated with 
the SONY Computer Science Lab in Paris, and has generated a good deal of 
research that relates to the idea of "common reference" as a driving objective 
in artificially reproducing human language, though not always under that 
name.   One of Steels' points is that the the need for shared connotation and 
adaptive processes puts the study of language at a higher order of 
rule-making than can be established by structural approaches or even those focusing on 
discourse.

A very informative article from a 2004 commercial applications blog on how 
Steels' approach relates to progress on information sharing technologies and 
a process that starts with dismbiguation through common reference (or 
"shared cognitions") and then moves on to the rest of language:
http://www.headshift.com/blog/2005/02/can-robot-learning-teach-us-ho.php

There are a number of Luc Steels' recent papers downloadable on the web, 
and they can be found by Googling his name.

I'll try to post something about how the idea of "common reference" relates 
to pragmatic concepts like deixis and indexicality in a future post.

Any questions or comments would be appreciated.

steve long

**************
An Excellent Credit Score is 750. See Yours in Just 2 Easy 
Steps! 
(http://pr.atwola.com/promoclk/100126575x1222585010x1201462743/aol?redir=http://www.freecreditreport.com/pm/default.aspx?sc=668072&hmpgID=62&
bcd=MayExcfooter51109NO62)