Feedback on paper of Lee & Barnden.


FLAWS

This paper has an unusual main hypothesis: one that refers to research methodology.
Unfortunately, no convincing evidence is provided to support this hypothesis. 

No clear criteria are given as to what it means for a machine to 'understand' a
metaphor, i.e, some formalisations of metaphors are provided, but it is not
clear how these are to be evaluated or assessed to ensure that they are, in some
sense, 'correct'.

The notion of 'dead metaphor' is mentioned, but this is not taken seriously as a
rival explanation, especially of mixed metaphors. That is, 'dead metaphor' is
interpreted as an importation of terminology into a domain, that was originally
metaphorical, but, though long usage, has now become part of the new domain,
such that it can no longer be consider metaphorical. If this happens twice or
more from different source domains, then the 'mixing' is now only apparent.



HYPOTHESES

Below, I quote the main hypotheses of the paper and discuss any evidence provided
to support them.

1. "This article suggests that such an assumption [that mixed metaphor is
secondary straight metaphors] is detrimental to long-term research".

	This is an interesting hypothesis. It is about research methodology in
	the metaphor-understanding field, rather than a claim about a specific
	system. What kind of evidence might support it? I'd be looking for
	evidence that the attempt to emulate and understand 'straight' metaphor,
	in a way that was not readily extensible to mixed metaphor, had led the
	field in the wrong direction. But no rival, 'straight'-only, research is
	discussed at all, let alone an attempt to somehow show that its
	programme had failed. Rather, we are shown an exemplar of an approach
	that can deal uniformly with both 'straight' and mixed metaphor. Such an
	example is suggestive evidence for the hypothesis at best. It shows that
	uniform approach is possible, but neither that it is the best approach
	nor that an uncombined approach is detrimental to the field. 

2. "It is claimed that the same kinds of reasoning and knowledge structures
involved in understanding straight metaphors are also required in understanding
mixed metaphors and that mixing is a central phenomenon."

	This is a sub-hypothesis of #1. The key word here is "required". One
	exemplar of a system that /can/ handle both is not evidence that it is
	/necessary/ to handle both with a single approach. Of course, one can
	appeal to Occam's Razor to argue that a single general approach is more
	desirable, if possible, although such an appeal is not made
	explicitly. Neither is evidence provided that mixing is a /central/
	phenomenon, just that there are quite a few examples of it and that not
	all these examples can be regarded as examples of "defective speech".

3. "The reasoning processes and data structures involved in understanding mixed
metaphors are identical to those used in understanding straight metaphors".

	This is similar to 2, but clearly too strong, since they themselves have
	to extend ATT-Meta to allow nested cocoons to deal with mixed
	metaphor. However, by providing a uniform model to deal with both kinds of
	metaphor, this is the claim they come closest to showing.

4. "ATT-Meta should, if possible, be broadly consistent with experimental
results in psychology".

	This is a very weak claim. Note that they are /not/ claiming
	psychological validity for their system ATT-Meta, but only that it is
	'broadly consistent' with experimental results, i.e., that its
	psychological validity has not already been decisively refuted. It's an
	interesting discussion, as it analyses what might be thought to be such
	a refutation, namely that it is based on a discredited "literal first"
	approach, and argues that the situation is more subtle and complex than
	previous commentators have presumed. This was an interesting
	contribution, but somewhat peripheral to the main theme of the paper.

The methodology employed in the paper is an example of what philosophers call
'conceptual analysis', and which is used throughout cognitive science. It is a
detail reasoned argument for conceptualising, modelling or formalising a
cognitive phenomenon in a particular way, in this case the nested-cocoon model
of metaphor. It typically uses a series of well-chosen examples and non-examples
to differentiate and prefer one model over its rivals. One disadvantage of this
methodology is that any claims are necessarily based on a few (non-)examples, since
each one requires an extended discussion. To compensate, it is implicit that
these examples were chosen, with some care, to be widely representative. 


YOUR REVIEWS

* Some of you missed or de-emphasised hypothesis 1. Its prominent placing in the
abstract and repetition throughout the paper made it the main claim. 

* Some of you picked a lot of minor, passing or supportive claims and gave them
equal billing with the more major ones. You have to learn to identify which
claims are major and which minor. Only the former need to be discussed, although
they subsidiary ones might be mentioned in support.

* Many people using the ERA forms were tempted to tick a lot of boxes without
justifying commentary, leading to vague and unsupported criticisms.

* The potential significance was underestimated. If the main hypothesis had been
proven then it might have saved a lot of researchers wasting their time with
unsuccessful methodologies.

* Some people using the latex template ticked contribution 9 "Identifies and
motivates a new problem", usually citing the main hypothesis 1. This was a
misclassification.