ABSTRACT
Animal behaviourists are interested in the causal factors that determine behavioural sequences — i.e. when animals perform particular activities, and under which circumstances they switch to alternative activities. It is accepted that observed behaviour results from the nervous system integrating information regarding the physiological state of the animal, e.g. the levels of nutrients in the blood, with sensory inputs, e.g. concerning the levels of nutrients in a food (Barton Browne, 1993). The combined physiological and perceptual state of the animal is termed the ‘motivational state’ (McFarland, 1999). MacDonald and Raubenheimer (1995) modelled behaviour sequences using an HMM whose unobserved underlying states were interpreted as motivational states. Their model captures an important aspect of the causal structure of behaviour, since an animal in a given motivational state (e.g. hungry) might perform not only the most likely behaviour for that state (feed) but also other behaviours (groom, drink, walk, etc.). There is not a one-to-one correspondence between motivational state and behaviour. And it is the runlength distributions of the motivational states that are of interest, rather than those of the observed behaviours. HMMs do not, however, take into account the fact that, in many cases, behaviour also influences motivational state; feeding, for example, leads to satiation.
