A widespread phenomenon in languages noticed or proposed by the linguist Michael Silverstein in 1976. The actants in a sentence may be treated differently not only according to their traditional actant role (subject, object, etc.), but also according to their place in a scale of what may be described as agentivity or less accuracy animacy.

Some people and things are supposedly more likely to act on other things than to be acted upon: they are more active than passive. They are certainly more capable: sticks and stones seldom leap up and do things to people. A stronger form of the Silverstein hypothesis is that this reflects a mental interest in the world: it is more natural for us to be doing things to others than for them to be acting on us. Like the Sapir-Whorf hypothesis, this mental model version is not strongly backed up by the facts.

What is undeniable is that there is a split in the behaviour of case marking, either in the complexity of forms, or in so-called split ergativity. Many languages are ergative (see that node) but few are purely so (Dyirbal in Queensland is one that is). Most use the ergative only in some situations and an accusative system in others. The split is usually at some point on the Silverstein hierarchy:

first person pronouns
second person pronouns
third person pronouns
proper names
kin terms
human nouns
animate nouns
inanimate nouns

To illustrate it in detail we'd have to use examples from obscure Caucasian languages, but it can partly be noted in English. We say I bit the child and The child bit me, but The dog bit the child, The child bit the dog. The first person pronoun has separate words I/me to mark the distinction I do something versus Something is done to me.

In a number of languages, I see him and He sees me are structured differently. When I'm doing it to someone else, it's called direct, but when the "less natural" roles are used, a form called inverse marking is used. This occurs in many North American languages. Under the node obviative I give some examples from Cheyenne, which has one prefix when the second person is an actant (either subject or object), another prefix when the first person is involved as either actant otherwise, and another when neither is involved.

The hierarchy is merely a generalization, a trend. Many languages do partly conform to it in various ways. We can find counterexamples in English: the second person you has less marking than the third person she/her and he/him. In the ergative language Georgian the first and second persons are less distinctly marked than any third person nominals including nouns; but then Georgian's ergativity is notoriously superficial.

This animacy hierarchy is or was also called the Silverstein hierarchy, which is what I originally noded this as. But the same hierarchy also applies to other phenomena, and it seems that 'animacy hierarchy' is now the usual term. For example, terms high up on the hierarchy show greater marking of number (singular, plural, dual etc.) than lower ones.