Bottom-up Abduction by Model Generation

We invest igate two real izat ions of paral lel abduct ive reasoning systems using the model generat ion theorem prover M G T P . The f irst one, called the M G T P + M G T P method , is a co-operat ive problem-solv ing archi tecture in wh ich model generat ion and consistency checking communicate w i t h each other. There , paral lel ism is explo i ted by checking consistencies in paral le l . However, since th is system consists of two dif ferent components, the possibi l i ties for paral le l izat ion are l im i t ed . In contrast , the other me thod , called the Skip m e t h o d , does not separate the inference engine f r om consistency checking, bu t realizes bo th funct ions in on ly one M G T P that is used as a generateand-test mechanism. In th is me thod , mu l t i p le models can be kept in d is t r ibu ted memories, thus a great amount of paral le l ism can be obta ined. We also a t t emp t the upside-down metain te rp re ta t ion approach for abduc t ion , in which top-down reasoning is s imulated by a b o t t o m up reasoner.