Phonological Pivot Parsing

There are two basic mysteries about natura l language. The speed and ease with which i t is acquired by a ch i ld and the speed and ease with which i t is processed. S i m i l a r l y to language acqu is i t i on , language processing faces a strong inpu t -da ta -de f i c iency problem. When we speak we a l t e r a great l o t in the idea l ized phonological and phonetic representat ions. We dele te whole phonemes, we r a d i c a l l y change al lophones, we s h i f t stresses, we break up i n tona t i ona l pat terns, we inser t the pauses at the most unexpected places, etc. I f to th is c r ipp led 'phonologica l s t r i n g ' we add a l l the noise from the surroundings which does not help comprehension e i t h e r , i t is bewi lder ing tha t the parser is supposed to recognize anything at a l l . However, even in the most d i f f i c u l t circumstances ( fo r eign accent, loud environment, being drunk, e tc . ) we do comprehend speech qu ick ly and e f f i c i e n t l y . There must be then some signals in the phonetic s t r i ng which are p a r t i c u l a r l y easy to grasp and to process. I ca l l these s ignals ' p i v o t s ' and parsers working with these s ignals I ca l l ' p i v o t parsers ' .