A Computational Model for Phonetic Interpretation

Motivation

``We listen to the beginnings and ends of things and just guess the rest,'' said Bill Idsardi in the pscholinguistics lab meeting.
``Double-Accent Phenomenon'' Mark Greenstreet explained why one of the grad students could not understand a faculty candidate.

Idea

Given a stream of sounds as input, the beginnings and ends of each word and also the beginning and end of the entire stream are more clear than the rest of the stream. So in order to interpret the sounds, we first map the clearer sounds to what we know already and then guess the stuff in between.

Steps Involved

Define the native inventory. A monolingual has one native inventory and a multilingual has one native inventory with the non-active inventory sounds suppressed.
If the listener is not a monolingual, then also define a working inventory. A working inventory is the inventory that holds the sounds of a learning language. Some sounds in the working inventory may be the same as some items in the native inventory.
INPUT = stream of phonetically marked sounds. Some sounds are more clear than others, and some are not clear at all.
Define the mapping and guessing functions. These two functions may interact so it is not so clear to me now what operations belong to which function. The necessary operations for both of these are:
1. Which sounds do we try to map first? Left->right? One pass only? In parallel?
2. Search strategy: Do some sounds have a shorter path to the inventory? Which inventory is "closer" to get to? Does there also exist a temporary inventory for each conversation?
3. When multiple candidates exist, how do we choose the right one?
4. Conformity: Obviously, at least phonology and semantics will interact here somehow. What exactly do they do?
5. If a mapping or a guess returns a "wrong" result, do we reanalyze? If so, how many times do we try this? When do we give up?
6. How are steps (a) to (e) ordered/combined with respect to each other?
OUTPUT = an interpreted sound sequence OR fail

Scenarios

Case 1: Native speaker (A) listens to Native speaker (B).

A hears stream of sounds; map to native inventory; guess the rest.
Nothing unusual.

Case 2: Native speaker (A) listens to Foreign speaker (B).

A hears stream of sounds; map to native inventory; more guesses.
Input more fuzzy; nothing unusual.

Case 3: Foreign speaker (A) listens to Native speaker (B).

A hears stream of sounds; map to working inventory; some guesses.
Interpretation depends on what is in the working inventory and their strengths (i.e., the amount of exposure A has)

Case 4: Foreign speaker (A) listens to Foreign speaker (B), where A is different from B.

A hears stream of sounds; map to working inventory; many more guesses.
Input extremely fuzzy.