[quote="Deinjur"]
There are a few different schools of thought on this. Some people feel that each word needs to sound like what it is, within the confines of their phonology. Meaning, you think about and create each word. This is very abstract, but you just might come out of it actually being able to remember a lot of your words, maybe even be able to speak your conlang 
[/quote]

I generally follow this method for artlangs, though I am willing to use word generators for engelangs.  I'm not sure why you describe it as "abstract" though; it seems more concrete, more embodied, having more psychic weight to it (I'm having trouble thinking of a word in English for whatever I'm thinking; gzb isn't helping either) than the opposite extreme method of generating random words and randomly assigning them to a preexisting list of meanings.

[quote="Deinjur"](Remember, VERY few conlangers are fluent in their language, and the ones that claim to be are suspect, because who can really judge them?).[/quote]

Indeed very few conlangers are fluent in their languages.  We may have a way to quantify how few, pretty soon; I'm working on analyzing the results of my conlang fluency survey.  Going through and marking up the email messages saved to text files so my Perl script can automatically convert them into a database or spreadsheeet is what's taking so long.  But my impression is that out of about 43 conlangers who've responded to my survey, only a handful claim to be fluent in speaking and thinking in their conlangs (though at least as many more say they can write read/text in their conlang at a reasonable speed without looking up words and grammar).  The sample is admittedly biased because, in addition to posting the survey on CONLANG and ZBB, I later sent it by email to a few specific people I'd heard of who are supposedly fluent in their conlangs but aren't active on those fora.

As for how we can judge their claims, well, I guess we can't, not with absolute objectivity.  But I trust people to be reasonably honest, and I think the highly specific questions I ask in the survey might help -- instead of just asking point-blank "are you fluent in your conlang?" I ask specifically whether they have spontaneous thoughts in it, whether they can think in it at will without consciously constructing sentences, whether they can write it as fast as they can type, etc.

http://bellsouthpwp.net/j/i/jimhenry1973/conlang/fluency-survey.html

[quote="Deinjur"]
I had quite a time figuring out a good word list to use; I started with Ogden's Basic English, which has about 850 words. However, it is a list generated for teaching basic English, not for creating a conlang. Some words in the Basic English list might be "covered" differently in the word list of another language.
[/quote]

Indeed, I wouldn't recommend that list.

[quote="Deinjur"]
But I found another list that I thought was better, mostly because it was much shorter: the Swadesh List. Only about 200 words there!
[/quote]

If I recall correctly, the Swadesh list was designed for etymological or sound change comparison across languages within families; it's supposed to list words that are resistant to being replaced by borrowing from another language, so each of those words in a given language's families will probably all be descended from the same root in the proto-language.  It's not an attempt at reducing to semantic primes [i]per se[/i].

[quote="Deinjur"]
In the second Language Creation Conference, John Clifford spoke a little about semantic primes, which aren't "words" so much as they are blocks of meaning. Its a different way of thinking, but a little reading here can also help you develop a word list of your own. I found a word list, called the Universal Language Dictionary, that groups words together according to concepts, which may help you if you want to create a derivational morphology or something. The ULD at least partially embraces the semantic prime idea, and can be another good resource for developing/building/copying a word list for lexicon generation.
[/quote]

I'm not sure Rick Harrison would describe the ULD as based on semantic primes.  It does make a good effort at listing concepts rather than simple English words though, but lots of its concepts are relatively complex or high-level, not prime.  Still, it's one of the best resources I know of for building your vocabulary one semantic field at a time.

[quote="Deinjur"]
So, with a short list, you CAN use the first abstract method, or you can randomly generate, and then change words as you determine better sounding ones, and add to the lexicon as you translate phrases. Long lists may be more cumbersome, but can be worth the time and headache if you plan on doing a lot of translating, as you won't have to stop to create a lot of new words each time.
[/quote]

If you're going to automatically generate a list of words to be matched to a list of concepts, I would recommend doing some work on the list of concepts first -- don't just take the ULD or Natural Semantic Metalanguage primes list or Swadesh list or whatever as-is, but go through carefully considering every line making it a unique definition of a word in your conlang, and adding concepts missing from your source list that you or speakers of your conlang will think and speak about frequently.  (When I pre-edited the ULD to generate part of the initial lexicon for [url=http://www.pobox.com/~jimhenry/conlang/conlang13/intro.htm]my engelang sb zjed'a[/url], I deleted most of the entries, combined some entries for related concepts into one, split up a few entries into multiple definitions, added a few concepts of my own, and -- crucially -- added a field with a numeric estimate of how often a word for such a concept was likely to occur relative to other words.)  Then sort the list of randomly generated words by length, from shortest to longest, and sort the list of concepts by the estimated frequency field, and combine them in that order, with the shortest words matched to the probably most frequent concepts.

Alternatively, you could do the same process of transforming your source concept list into the list of definitions for your conlang, sort by estimated frequency, and then go through it manually in that order devising words by hand according to your taste.

Leave the estimated frequency field in your dictionary, perhaps; you could use it as input to a flashcard program if you're trying to learn your conlang, so that it's more likely to quiz you more often on words for "sleep" and "clothing" than words for "statistics" and "markup language".  Later on, maybe after you build a large enough corpus, you could replace the estimated frequency data with actual frequency data.  (This is part of my long-term plans for sb zjed'a.)



Another method of preparing a definition list to be matched to a wordform list, which I've thought of for a possible future engelang but haven't yet implemented, is this; you might plan on some of your definitions being compounds or derivations of more basic words, without yet knowing what the actual phonological forms of the roots words and suffixes are going to be.  So you could maybe build a definition file with lines like this,

[code]
CAUS		causative suffix
GB		[adj] green, blue
__		cat, bobcat, mountain lion
BG-CAUS	to embarrass, cause someone to blush [the speakers of the conlang have copper-based blood]
[/code]

Then your script to match up wordforms with definitions would remember what wordform it associated with the codes CAUS and GB, and when it sees a hyphenated code, look up its component parts and see what wordforms it used earlier, and use those to form a word for this definition instead of assigning a random unanalyzable word.   __ is a placeholder for words you don't plan to use in compounds so don't need a unique code for.  So you'd get output like


[code]
-ta		causative suffix
suri		[adj] green, blue
xawi		cat, bobcat, mountain lion
surita		to embarrass, cause someone to blush
[/code]

(Much simplified; a real implementation would probably have an estimated-frequency field, a part of speech or argument structure code field, and things like that.)

