This document describes my ideas about how natural lnguage works and how we could build a computer system that can have a meaningful conversation with us.
It starts with the basic reality that our heads are finite structures that have finite extent in three space. The implications of this simple observation have a large consequences for being able to build an 'artifical intelligence'. This paper, however, will primarily discuss the function of language and how we can use these ideas to build an artifact the can talk with us about its understanding of our world. We leave to other papers how this understander might, in practice, be 'built'.
It is ofter imagined that sentences have 'meaning'. My understanding is
that the sentences themselves do not have meaning but
The meaning of a sentence is the effect/change it has in the head of a hearer.
There are several things to notice about this definition. First there is the explicit mention of 'a hearer' Fernando Flores pointed out to me that there is a difference between 'data' (from the latin 'do' - to give) which refers to what might be given and 'capta' (from the latin 'capto' - to get or capture), The former can be thaught of as what a speaker says, the latter what a listener hears.. Our definition of meaning focuses on 'capta' not 'data'. The point is clear. What matters, when a sentence is uuttered, is not the expectations of the speaker but rather how the hearer interpretes the utterance.
This is, of course, one of the main reasons people lie. Not because they intend to be deceptive, but rather they have some idea what the listener will hear and they adjust what they say to make their point. In the trivalist case an English speaker visiting Germany might try to express their desires in German, thus making it more likely they are saying something meaningful. Searle in his famous paper, ^^^ remarks that for most native English speakers, there is no meaning in a Chinese sentence. In my mind this is a simple direct consequence of the above definition.
Second, a sentence may have different 'meaning' for different hearers. It is everyday folklore that there can be misunderstandings during conversations. Again this is a simple direct consequence of my definition. Furthermore, much of natural language is about how, in a particular conversation, whe participants to understand the words being used.
It has been suggested that the 'context' of the conversation matters. The 'meaning' of the word 'batter' will likely be different in the kitchen than at the ball park. This is right but needs to be understood that the 'context' itself is an aspect of the understanding of the hearer - if we are discussing baseball (or cricket) in the kitchen, that awareness is embodied in the participants, not in the sentences themselves.
All this adds up to the fact that scouring over a sentence for its 'meaning' without thinking about how it effects its hearer is hopeless.
The idea that a sentence has 'meaning' independent of its effect on a hearer leads to the idea that there is an arbitor of meaning that is somehow outside of the participants in a conversation. I reject this and suggest instead that the notion of 'common knowledge' simply means that an individual believes that the mention of some notion in a sentence, will be interpreted by most hearers in a way that is compatable with the speakers idea of the notion. Can a speaker make life death decisions bsased on this belief? Of course not. No reasonable person believes that 'common knowledge' is absolute. They know from an early age the difference between data and capta. This is common sense justification for the above definition of meaning,
------------------------------------------------------------------------------ A long time ago I taught a course called 'Philosophy and Aritficial Intelligence' and I used to tell this story. "Suppose you are on a long drive and you get thirsty and you pull over to a roadhouse only to discover that its actually a neighborhood biker bar. As you enter you notice that the only people in the bar are very tough and strong looking men who, independent of their actual actual personalities, are intimidating to you by their very looks. Unfortunately, you are extremely thirsty and you decide to suppress your fears and buy a drink anyway. So you go up to the bar and happen to stand next to a very mean looking leather clad person. They take one look at you and say "What's the summ of 2 plus 2?" Here's the deal. If you heaitite for even a microsecond wondering what he ment then you have doubts about the meaning of sentences even when the refer to prima facie the most basic 'analytic' truth. Either the answer is 4, without hesitation because it's shared 'common knowledge' about arithmetic, or if you heasitate for just an instant your faith in the external validity of arithmetic is broken. If 'common knowledge' is really true for everyone then there can be no need for worndering what the person at the bar 'meant'. ------------------------------------------------------------------------------
The attempt to find the 'meaning' of a sentence outside the head of a listener is the root cause of endless difficulties in the traditional philosophy of language. Consider the debates about direct vs indirect speech acts. The issue here is whether we should accept the literal meaning of a sentence or instead attribute a contextual meaning to it. This boils down to is the meaning some abstract situation independent idea (wouldn't it be nice) or, once again is it somehow contextually defined. My definition of 'meaning' implies that all speach acts are indirect and that it is only accidental if sentence could be thaught of as having a 'direct' meaming.
A similiar discussion could be had about the distinction between 'de re' and 'de dicto' utterances. My definition implies that there are no 'de re' meanings but all phrases must be thought of as 'de dicto'.
Consider the sentence "John is in New Mexico" which is true now but may be false at some later date. Do we imagine that the correctness of this 'fact' has a structural differences in our heads. If it does then it would have the wierd consequence that remotely occurring events culd change our head. i.e., since their meaning 'changes' our heads would change. JMC worried about this in his discussions of what he called the 'frame problem' and how you would cope with such issues. In all the above cases, the root of the problem is the assumtion that meaning is indepent of an individual and as is truth. My definition rejects both these ideas.
Understand what is heard
The above discussion refocuses our attenton from trying to understand what a sentence means to trying to determine how a hearer inteprets what they hear. I believe this shift in perspective opens up a pathway both for natural language understanding and a blueprint for building a computer system that understands and can discuss its environment.