Evolution of Language and Creativity:
Evolutionary Precursors to Communicative Language:
Internal Languages

(Narrow and broad evo-devo theories of language.)

Aaron Sloman
http://www.cs.bham.ac.uk/~axs
School of Computer Science, University of Birmingham.

Installed: 11 Mar 2011
Last updated: 11 Mar 2011;13 Mar 2011; 25 Feb 2019 (formatting)

These notes are based on an email message I posted after the IRLAB presentation on 10th Mar 2011, on Evolution of Language, by
Rustam Stolkin
https://www.birmingham.ac.uk/staff/profiles/metallurgy/stolkin-rustam.aspx
His talk included some ideas of Derek Bickerton.

My original message was posted on Fri, 11 Mar 2011 11:33:33 GMT, but there have been several changes since then, some indicated with dates below.

At the end of the seminar, I suggested that most researchers on language and its evolution mistakenly ignore a host of related competences that are present in far more species. (I don't know enough about the work of Bickerton to comment directly on his claims. I was reacting to the subset of his work presented in the talk.)

Note Added 13 Mar 2011:

I have just found this paper, and it confirms my suspicion:

http://www.elsevier.com/authored_subject_sections/S06/S06_345/misc/languageevolution.pdf
Language evolution: A brief guide for linguists
Derek Bickerton
Lingua 117 (2007) 510-526 (Online since 2005)

    ...
    "All species ancestral to ours (plus our cousins, the Neanderthals) were
    conservative in the extreme, preserving their small and limited tool-sets
    virtually unchanged for hundreds of thousands of years. The emergence of our
    own species released a torrent of creativity that is still gathering speed.
    What caused this difference? Clearly, some startling increment in cognition."

This assumes that there was no creativity in the use of those tool-sets.
Compare the fact that the structure of human hands has been unchanged for
many thousands of years. That is not evidence for the lack of creativity of
humans. Evidence of creativity can be found in behaviours as well as products
of behaviours. Unfortunately, behaviours do not leave fossil records, so we
need to use much more indirect evidence.

A problem in many discussion of evolution is that a failure to take account of
some of the details of animal competences and to think deeply about the kinds
of mechanisms that are and are not capable of accounting for those competences
(which requires adoption of the designer stance) can lead to blindness
regarding some of the transitions that occurred in evolution and still occur in
human development. (Contrast A. Karmiloff-Smith's Beyond Modularity.)

    Bickerton continues:
    ...
    "A central, if never yet addressed, problem for Pinker-Bloom, Jackendoff and
    anyone else who takes a gradualist position on syntax lies in the
    relationship between language and cognition. If the achievement of full
    syntactic language was what sparked the creativity of our species noted in
    the previous section, and if syntax developed gradually, why did it not
    confer the least inkling of that creativity on species antecedent to our own?"

This seems to me to show ignorance of some of the results of observations of
animals as diverse as corvids, elephants, various sorts of apes, squirrels, and
octopuses. (Refs to be added.) Moreover, people who have not studied AI and
tried to design intelligent robots and other machines typically have no
understanding of the difficulties of giving them capabilities found in many
species that do not use language, and in pre-verbal children. It was thinking
about the requirements for such capabilities (e.g. nest-building in magpies and
other corvids) that fuelled the ideas about the requirement for GLs summarised
below.

Note that there are two interpretations of "gradualist": one implying
continuous change, the other allowing large numbers of discontinuous changes
including varying degrees and kinds of discontinuity. Understanding of
evolution is impossible if the latter interpretation is ignored, which
Bickerton -- like many other thinkers-- seems to do in this paper.

What we have learnt in the last six or seven decades about varieties of
information processing and what they can do makes a huge difference to our
ability to think about evolution and development, and more generally to think
about complex working systems. Unfortunately that is not yet a standard part of
most educational systems, and the consequences for scientific and
philosophical theorising about evolution, development, learning, perception,
are drastic.

(Perhaps this is an instance of the Singularity of Cognitive Catch-up (SOCC).)

The competences that are usually ignored include competences of many non-human
species, competences that evolved long before human use of language for
communication, and competences that develop before use of communicative
language in young human children.

Added: 13 Mar 2011
It is not usually noticed that creativity in behaviour and in perception
has to precede creativity in producing tools, art-works, and other physical
products of behaviour. That is because creative behaviours and perception
of structures and affordances that have not previously been noticed, are
required for the production of the more enduring types of evidence of
creativity: you cannot make a novel type of tool without first being able to
produce novel actions, partly on the basis of perceiving possibilities and
constraints on possibilities -- i.e. proto-affordances -- that had not
previously been noticed.

Early uses of tools may involve selection of by-products of actions previously
done for other purposes (e.g. noticing that stones that are thrown can break in
a way that produces sharp edges). But (a) that requires new types of perception
and (b) it requires new types of action -- as the newly noticed tools are used
for tasks that were previously not performed at all or performed in other ways,
and usually less well.

A later stage could include not merely using the tools but also creating them,
perhaps initially by modifying accidentally created tools (e.g. making them
sharper or easier to hold), then later creating them from scratch, starting
from mateials that were previously not used at all, or used for unrelated
purposes.

These non-linguistic competences appear to be inexplicable except as the
products of powerful information processing systems that manipulate
internal information structures that must have many of the features
regarded as unique to language --  even though they are used internally and not
for communication.

So, being used for communication should not be regarded as a defining
characteristic of "language" in the most useful, broad sense of that word.
    For more detailed arguments see
    http://www.cs.bham.ac.uk/research/projects/cogaff/talks/#glang

Those internal uses of more or less complex information structures include
perception, learning, motive formation, reasoning, consideration of alternative
actions, forming plans for sequences of actions, evaluating such plans,
executing them, revising them when they go wrong, etc. etc.

Note added: 13 Mar 2011

There is now a wide-spread, but seriously mistaken, fashion for claiming that
if only we understand the implications of embodiment we shall see that it is
unnecessary to postulate such internal information processing mechanisms to
explain human and animal competences, or to produce them in future robots.
A lengthy critique of Brooks' influential exposition of this fashionable view
is available in:
    Some Requirements for Human-like Robots:
          Why the recent over-emphasis on embodiment has held up progress,
    in Creating Brain-like Intelligence, Springer-Verlag, 2009, pp. 248--277,
    Eds. B. Sendhoff and E. Koerner and O. Sporns and H. Ritter and K. Doya,
    http://www.cs.bham.ac.uk/research/projects/cosy/papers/#tr0804

One of the reasons why most researchers on language ignore such non-linguistic
competences is that they focus attention mainly (and sometimes exclusively) on
the first of three major aspects of human language:

1. the formal properties of the overt expressive vehicles used
    (words, phrases, sentences, paragraphs, narratives, etc.)

2. the semantic contents expressed using those vehicles
    e.g. referring to or describing scenes, processes, hypotheses, problems,
    puzzles, goals, possibilities, constraints, opportunities, generalisations,
    questions, etc.

3. the multiple uses to which languages are put, including private
    uses such as wondering, remembering, hypothesising, predicting,
    explaining, planning, having internal debates, formulating
    questions, being puzzled about, including generation and
    interpretation of external linguistic utterances. ... etc.

When considering the evolution of language (or its development in
individuals) many researchers focus far too much on 1, and ask how that could
have evolved, ignoring the evolution of 2 and 3.

By focusing on 1 (including ignoring the internal mechanisms required for
production and interpretation of the vehicles of communication) they are
focusing on the (often unobvious) structures of overt behaviours.

That encourages people who think about evolution of language to concentrate on
questions about how overt behaviours evolved and that leads to a focus on all
the considerations that were presented in the seminar, including variations in
morphology, and variations in some kinds of non-linguistic actions such as
tool-making, hunting, consuming a carcass, etc.

If, instead, we focus on on items 2 and 3, which are also crucial features of
communicative language, and ask what sorts of internal mechanisms are required
to support them, and how the mechanisms may have evolved (including information
processing mechanisms required for production and comprehension, such as
parsing or generation of complex linguistic structures) that suggests a very
different view of both the evolution of language and its development in children.

Note added: 13 Mar 2011

Examples of mechanisms required obviously include (a) various kinds of short
term and intermediate term memory for dealing with complex and changing
percepts and controlled actions, and (b) various kinds of long term memory, for
ontologies, for discovered laws, theories, and useful generalisations, for
information about enduring individuals of various kinds, for information about
where various things are, including things out of sight, things that happened
in the past, and expected future events, processes and states of affairs.

Less obviously, garbage-collection mechanisms are required that enable new
information to replace old information that's no longer of any using, including
information in many transient memory stores, and information in longer term
stores that may need to be replaced or modified.

One of the conclusions I reached reluctantly and with some distress, in the
days when I was still teaching (including philosophy and programming) is that
humans can vary significantly in the details of their mechanisms for performing
these functions. For instance I suspect some researchers on evolution or
development of language may find it very hard to keep all of these diverse
issues in mind while thinking about the problems. This could be a limitation of
certain memory mechanisms.

That diversity, though distressing, should not be surprising, given all the
other significant differences between humans, including physical and
behavioural differences.

(The theological implications will not be discussed here!)

In particular, it suggests that long before human languages were used for
communication there were richly structured, highly manipulable,
multi-functional internal forms of representation supporting a host of
cognitive functions (including, for example, visual perception of complex scenes
and processes, and generation and control of actions).

Those pre-linguistic internal forms of representation were almost certainly
very different in detail from human spoken languages (though they may have more
commonalities with human sign languages). In particular the forms of
representation of visually perceived changing 3-D environments probably do not
consist entirely, or even mainly, of linear, discrete, sequences of symbols.
(Contrast trees, graphs, various kinds of maps, with or without metrical
information, etc.) Exactly what the intermediate forms of representation are
that humans and other animals use in perceiving complex scences, and what sorts
of architectures and mechanisms make that use possible, remain unsolved
problems.

But analysis of the achievements of those representations and mechanisms
suggests that they shared with human communicative languages at least:
 -- the use of re-usable components that can be assembled into larger
    components of varying complexity and varying structure,
 -- the use of structural relationships that can involve different sorts
    of terms in the relationships (e.g. contains, touches, between, before,
    after, overlaps, etc.)
 -- mechanisms for constructing, dis-assembling, rearranging, comparing,
    contrasting, and interpreting the forms making use of those components
    and relationships.
 -- use of compositional semantics in the sense that the semantic content of
    larger structures (including novel ones) are derived from
     (a) the semantic contents of the components and relationships, and
     (b) how the components are assembled
         (as in sentences and also maps, flow charts, computer programs,
         algebraic formulae, etc.).

NB: Compositional semantics is a requirement for coping with novelty.

(Actually it needs to be context-sensitive compositional semantics, but that's
a complication that we need not consider here. See
    http://www.cs.bham.ac.uk/research/projects/cosy/papers/#dp0605

I don't think there is anything in AI that comes close to meeting all those
requirements (and other requirements I've not mentioned).

Jackie Chappell and I have been calling these internal forms of representation
GLs, Generalised Languages, because in talking about them we generalise the
notion of language to include systems not used for communication, but sharing
many of the features of communicative languages.

Thinking about the evolution of these internal forms of representation and how
they could later have led in humans, and to a lesser extent other species, to
the development of communicative languages, leads to a very different
evolutionary story, one of whose consequences is that human brains evolved so
as to use sign languages as an extension of non-linguistic intelligent actions,
especially collaborative actions. Vocal languages only came later.

This may help to explain (among many other things) why the deaf children
brought to live together in Nicaragua were able spontaneously and rapidly to
develop a sophisticated new sign language that far exceeded the linguistic
scope of the sign language their teachers were trying to teach them.

There's lots of information online about this e.g.
    http://en.wikipedia.org/wiki/Nicaraguan_Sign_Language

The view expressed here is not to be confused with other theories that human
language started as gestures, where the original gestures had the primary
function of being communications rather than other actions. Several such
theories focus (mistakenly) on recent work on mirror neurons. (E.g. Michael
Arbib, I think.)

For more detail (but still with many gaps) see
    http://www.cs.bham.ac.uk/research/projects/cogaff/talks/#glang
    Evolution of minds and languages.
    What evolved first and develops first in children: Languages for
        communicating, or languages for thinking
        (Generalised Languages: GLs)?
    Informal presentation. Original version 2007, since updated.

An earlier paper making similar points:

    http://www.cs.bham.ac.uk/research/projects/cogaff/81-95.html#43
    Aaron Sloman (1979)
    The primacy of non-communicative language, in
    The analysis of Meaning: Informatics 5
       Proceedings ASLIB/BCS Conference, Oxford, March 1979,
    Eds. M. MacCafferty and K. Gray,
    ASLIB, London, pp. 1--15,

Comments, criticisms and suggestions welcome.

Maintained by Aaron Sloman
School of Computer Science
The University of Birmingham