W3C home > Mailing lists > Public > w3c-wai-gl@w3.org > July to September 2002

Re: 4.1 proposal for discussion

From: Avi Arditti <aardit@voa.gov>
Date: Thu, 05 Sep 2002 13:55:10 -0400
Message-ID: <3D779A7E.DB3D2300@voa.gov>
To: w3c-wai-gl@w3.org
CC: kgodden@atl.lmco.com

With permission, I am forwarding this note from computational linguist
Kurt Godden, Ph.D. I first met Kurt when he headed the Controlled
Automotive Service Language (CASL) program at General Motors.  He is now
Principal Member of the Engineering Staff, Advanced Technology Labs,
Lockheed Martin Corporation:

"Kurt S. Godden" wrote:
> 
> I agree completely about the difficulties in trying to give specific
> success criteria based on syntactic constructs such as passive etc,
> where languages can differ so much.
> 
> Given that you intend to cover content in any language, I think you are
> better off giving more abstract, higher-level success criteria, but you
> could use language-specific examples.  An example of higher-level
> 'universal' criteria would be something like 'To the extent
> reasonable/feasible etc try to have no more than one clause per
> sentence.'  That one may be a bit too restrictive, but it at least
> illustrates the principle.  This rule would entail, for example, that
> you break up conjoined sentences, and (here is where it's probably too
> restrictive) would reduce/eliminate subordinate clauses.
> 
> You'd also want to have some abstract guidelines that address issues of
> ambiguity and the need to reduce it, e.g. avoid noun phrases that
> consist of strings of noun modifiers.  This is intended to reduce
> strings of nouns, which are often used in English technical literature,
> such as 'powertrain control module', because as these noun strings get
> longer, more and more ambiguity is introduced.  Does the Nth noun modify
> its successor, or some other one?  As a specific instance of this
> guideline, in the CASL project we had a rule for service authors not to
> use what we called "noun names" that consisted of more than 4 nouns in a
> row.  We found sentences in published service manuals that had as many
> as NINE nouns strung together in a single NP.  Only expert mechanics had
> a chance at understanding those things, and even they probably
> struggled.  As an alternative, we gave examples in our author guidelines
> of how to break such long noun strings up into nouns with prepositional
> phrase modifiers.
> 
> To the extent that long sentences also tend to be more complex than
> short sentences, you may want to consider something like a guideline on
> limiting the length of a sentence to being <= some number of words.  In
> CASL, we had a rule to write sentences no longer than 25 words, which
> seemed to work well in practice, and automatically forced simpler
> sentence structures.  Of course, other languages won't work well for a
> guideline like this, e.g. Turkish or even German.
> 
> Finally, you may want to say something additional about the selection of
> words/terminology.  You do refer to giving explanations of new
> concepts/terms, but for non-technical terms you may want to say
> something about using commonly occurring words.  The general idea I'm
> getting at is that I never liked those metrics like Flesch or the Fog
> Index that penalized words for having many syllables.  By that criteria,
> the word 'automobile' is harder to understand than 'neem' (a kind of
> tree that grows in India, but the word appears in English
> dictionaries).  What's important is the frequency of occurence of words
> in a language.  Words with higher freqency of occurrence are more
> understandable, regardless of syllable length.
Received on Thursday, 5 September 2002 13:55:48 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Monday, 7 December 2009 10:47:20 GMT