Given a continuous acoustic/auditory input stream,
how are segments, syllables, words extracted from it?
Why it's hard
The input contains far more than what is needed.
For a given phonetic/phonological unit, there is great
variability: gender, dialect, other speaker differences, and rate
all make a difference.
For a given segment, there may be strong co-articulation effects:
segments may depend on their contexts.
A high level of noise in the input doesn't bother people.
Analysis (parsing, understanding)
Given a stretch of text (a sentence, a paragraph, a
conversation), produce a representation embodying the semantic
content of the text (what is this anyway?) and possibly also the
presumed intentions of the speaker/writer (and what are these?).
Given a sentence, produce a representation for its syntactic
structure.
Why it's hard
Hearers can understand sentences they have never heard before.
The input may be ill-formed; within limits, this does not bother people.
Many sources of knowledge (morphology, syntax, linguistic
semantics, linguistic pragmatics, world knowledge) need to brought to
bear.
The input may contain ambiguity (structural, lexical, pragmatic).
The input often underdetermines the intended output.
Speakers often make use of metaphor and metonymy.
The hearer needs to have (in some sense) a model of the speaker.
Generation (production)
Given some communicative intent (pragmatic and/or semantic),
produce some text.
Why it's hard
Speakers can talk about thoughts they've never talked about before.
The speaker needs to have (in some sense) a model of the hearer.
Speakers plan what they say based on mutual knowledge.
Machine translation
Given some text in one language, produce a corresponding text in
another.
Why it's hard
One word or structure in the source language may correspond to two
or more in the target language.
This may require a deep understanding of the source text.
Languages have different requirements about what is obligatory in
their grammars.
Language acquisition
Given samples of a language, learn a grammar.
Given examples of words used in context, learn the meanings
and structural properties of the words.