Here's a secret: I have never invented a word personally. It was always a product of an programmed engine.
My latest toy is something that literally mix words together. It was written as an attempt to automate lojban root from its sources.
The algorithm is...
well...
I'm too lazy to explain it, study the algorithm yourself by looking at the javascript: https://dl.dropboxusercontent.com/u/5517255/JS%20Tests/Mixer.html
Anyway, my first experiment is combining Japanese, Korean, and Chinese. For starter, here's the numeral (Cause everyone loves numeral)
My latest toy is something that literally mix words together. It was written as an attempt to automate lojban root from its sources.
The algorithm is...
well...
I'm too lazy to explain it, study the algorithm yourself by looking at the javascript: https://dl.dropboxusercontent.com/u/5517255/JS%20Tests/Mixer.html
Anyway, my first experiment is combining Japanese, Korean, and Chinese. For starter, here's the numeral (Cause everyone loves numeral)
CH JP KR CAN1 CAN2 CHOS 1 it pito hana itop hana itop 2 ni puta tur tur nip nip 3 sam mi seis sei sam sei 4 si io neis sio nei syo 5 ñu itu tases tase ñuit tase 6 riuk mu ieses ries iuki ryes 7 tshit nana irkop natti irkop natti 8 prat ia ieter prat iete yete 9 kiu kokono ahop kokiu ahoko kokyu 10 dsip too ier ieri toip yeri 100 braik momo on mono brai mono 1000 tshen ti tsëmën tshen sëmën chen 10000 mion iorodu tëmen ioron tëmen yoron
Second-order Markov chain. That's all you had to say!
Looks neat, though! Definitely has potential for koineization. I see you're using syllable structure from the whole dataset but restricting composition to the same row—how are you doing that without modifying the transition frequency table? Do you just throw out hits to bad characters, or are you rewriting the frequency table to exclude them afterwards?
Looks neat, though! Definitely has potential for koineization. I see you're using syllable structure from the whole dataset but restricting composition to the same row—how are you doing that without modifying the transition frequency table? Do you just throw out hits to bad characters, or are you rewriting the frequency table to exclude them afterwards?
The frequency table is reset after every row.
It's Markov-ey in a way but with a different... "frequencization".
Right, here's how the algorithm works.
See the number above each columns? That's the "weight" of each language. For the default it's set at equal 10. In Lojban the weight for each number is adjusted by the number of speakers.
Let's take the first row for example. First the engine concatenates each member of each column and give them a value by multiplying the weight of each column, like so:
And then for each concatenated string it takes a set of digraphs by taking a letter and all the letters to the right of it. For example the string "pitohana" will result in the following digraphs:
And then each digraphs are given a value by the following formula:
combined value / (location_of_first_leter + location_of_second_letter)
Like this:
Notice how some digraph appears twice or more? We just simply add the frequency. Do this for all other 8 combos and add them up!
After all is finished. We sort the digraphs with the largest value to the lowast.
Like dis:
["it","ha","pi","an","to","na","tp","ti","th","ah","op","oh","ai","ap","oi"]
["tu","ni","pu","ur","ut","ta","ip","it","in","rt","rp","rn","at","ap","an"]
["se","sa","mi","is","ei","am","ms","im","ss","mm","sm"]
Now... the engine will perform the wordbuilding.
Continues to the next reply.
It's Markov-ey in a way but with a different... "frequencization".
Right, here's how the algorithm works.
See the number above each columns? That's the "weight" of each language. For the default it's set at equal 10. In Lojban the weight for each number is adjusted by the number of speakers.
Let's take the first row for example. First the engine concatenates each member of each column and give them a value by multiplying the weight of each column, like so:
CONCAT VALUE itit 100 itpito 100 ithana 100 pitoit 100 pitopito 100 pitohana 100 hanait 100 hanapito 100 hanahana 100
And then for each concatenated string it takes a set of digraphs by taking a letter and all the letters to the right of it. For example the string "pitohana" will result in the following digraphs:
p i t o h a n a p - pi pt po ph pa pn pa i - - it io ih ia in ia t - - - to th ta tn ta o - - - - oh oa on oa h - - - - - ha hn ha a - - - - - - an aa n - - - - - - - na a - - - - - - - -
And then each digraphs are given a value by the following formula:
combined value / (location_of_first_leter + location_of_second_letter)
Like this:
1 2 3 4 5 6 7 8 p i t o h a n a 1 p - 33,33 25 20 16,66 14,28 12,5 11,11 2 i - 25 20 16,66 14,28 12,5 11,11 10 3 t - - 16,66 14,28 12,5 11,11 10 9,09 4 o - - - 12,5 11,11 10 9,09 8,33 5 h - - - - 10 9,09 8,33 7,69 6 a - - - - - 8,33 7,69 7,14 7 n - - - - - - - 6,66 8 a - - - - - - - -
Notice how some digraph appears twice or more? We just simply add the frequency. Do this for all other 8 combos and add them up!
After all is finished. We sort the digraphs with the largest value to the lowast.
Like dis:
["it","ha","pi","an","to","na","tp","ti","th","ah","op","oh","ai","ap","oi"]
["tu","ni","pu","ur","ut","ta","ip","it","in","rt","rp","rn","at","ap","an"]
["se","sa","mi","is","ei","am","ms","im","ss","mm","sm"]
Now... the engine will perform the wordbuilding.
Continues to the next reply.
Oh god. I just realize there's a mistake in how I translate my algorithm to Javascript 
BRB, changing the JS.
BRB, changing the JS.
Okay, where was I again? Ah yes... the digraph list. Right, here is all the digraph generated in the first row:
["it","ha","pi","pt","hn","an","aa","po","io","ii","to","na","ti","ia","tt","ip","ih","ta","ai","at","th","tp","hi","in","ht","ah","ap","pa","tn","ni","oi","nt","hh","hp","pp","ph","ot","ao","oa","pn","nh","np","oh","ho","op","nn","no","on","oo"]
The length of the resulting mix is the average length of the parents rounded to the next integer. But for example, let;s ignore that limitaiton. Now... we take the very first digram "it" and mark it out from the array:
"it": ["ha","pi","pt","hn","an","aa","po","io","ii","to","na","ti","ia","tt","ip","ih","ta","ai","at","th","tp","hi","in","ht","ah","ap","pa","tn","ni","oi","nt","hh","hp","pp","ph","ot","ao","oa","pn","nh","np","oh","ho","op","nn","no","on","oo"]
Next, we find a digrams that start with "t" and mark it out:
"ito": ["ha","pi","pt","hn","an","aa","po","io","ii","na","ti","ia","tt","ip","ih","ta","ai","at","th","tp","hi","in","ht","ah","ap","pa","tn","ni","oi","nt","hh","hp","pp","ph","ot","ao","oa","pn","nh","np","oh","ho","op","nn","no","on","oo"]
Do it until reached the desired length:
"ito": ["ha","pi","pt","hn","an","aa","po","io","ii","na","ti","ia","tt","ip","ih","ta","ai","at","th","tp","hi","in","ht","ah","ap","pa","tn","ni","oi","nt","hh","hp","pp","ph","ot","ao","oa","pn","nh","np","oh","ho","op","nn","no","on","oo"]
"itoi": ["ha","pi","pt","hn","an","aa","po","io","ii","na","ti","ia","tt","ip","ih","ta","ai","at","th","tp","hi","in","ht","ah","ap","pa","tn","ni","nt","hh","hp","pp","ph","ot","ao","oa","pn","nh","np","oh","ho","op","nn","no","on","oo"]
"itoio": ["ha","pi","pt","hn","an","aa","po","ii","na","ti","ia","tt","ip","ih","ta","ai","at","th","tp","hi","in","ht","ah","ap","pa","tn","ni","nt","hh","hp","pp","ph","ot","ao","oa","pn","nh","np","oh","ho","op","nn","no","on","oo"]
"itoiot": ["ha","pi","pt","hn","an","aa","po","ii","na","ti","ia","tt","ip","ih","ta","ai","at","th","tp","hi","in","ht","ah","ap","pa","tn","ni","nt","hh","hp","pp","ph","ao","oa","pn","nh","np","oh","ho","op","nn","no","on","oo"]
And so forth.
What if we haven't reached the wanted length but there's no viable digrams? Well... the engine will take a step back and find for next digram.
The medial and final filter are placed to prevent the engine from making phonotactically illegal strings.
["it","ha","pi","pt","hn","an","aa","po","io","ii","to","na","ti","ia","tt","ip","ih","ta","ai","at","th","tp","hi","in","ht","ah","ap","pa","tn","ni","oi","nt","hh","hp","pp","ph","ot","ao","oa","pn","nh","np","oh","ho","op","nn","no","on","oo"]
The length of the resulting mix is the average length of the parents rounded to the next integer. But for example, let;s ignore that limitaiton. Now... we take the very first digram "it" and mark it out from the array:
"it": ["ha","pi","pt","hn","an","aa","po","io","ii","to","na","ti","ia","tt","ip","ih","ta","ai","at","th","tp","hi","in","ht","ah","ap","pa","tn","ni","oi","nt","hh","hp","pp","ph","ot","ao","oa","pn","nh","np","oh","ho","op","nn","no","on","oo"]
Next, we find a digrams that start with "t" and mark it out:
"ito": ["ha","pi","pt","hn","an","aa","po","io","ii","na","ti","ia","tt","ip","ih","ta","ai","at","th","tp","hi","in","ht","ah","ap","pa","tn","ni","oi","nt","hh","hp","pp","ph","ot","ao","oa","pn","nh","np","oh","ho","op","nn","no","on","oo"]
Do it until reached the desired length:
"ito": ["ha","pi","pt","hn","an","aa","po","io","ii","na","ti","ia","tt","ip","ih","ta","ai","at","th","tp","hi","in","ht","ah","ap","pa","tn","ni","oi","nt","hh","hp","pp","ph","ot","ao","oa","pn","nh","np","oh","ho","op","nn","no","on","oo"]
"itoi": ["ha","pi","pt","hn","an","aa","po","io","ii","na","ti","ia","tt","ip","ih","ta","ai","at","th","tp","hi","in","ht","ah","ap","pa","tn","ni","nt","hh","hp","pp","ph","ot","ao","oa","pn","nh","np","oh","ho","op","nn","no","on","oo"]
"itoio": ["ha","pi","pt","hn","an","aa","po","ii","na","ti","ia","tt","ip","ih","ta","ai","at","th","tp","hi","in","ht","ah","ap","pa","tn","ni","nt","hh","hp","pp","ph","ot","ao","oa","pn","nh","np","oh","ho","op","nn","no","on","oo"]
"itoiot": ["ha","pi","pt","hn","an","aa","po","ii","na","ti","ia","tt","ip","ih","ta","ai","at","th","tp","hi","in","ht","ah","ap","pa","tn","ni","nt","hh","hp","pp","ph","ao","oa","pn","nh","np","oh","ho","op","nn","no","on","oo"]
And so forth.
What if we haven't reached the wanted length but there's no viable digrams? Well... the engine will take a step back and find for next digram.
The medial and final filter are placed to prevent the engine from making phonotactically illegal strings.
After fixing the JS. Here's the result of the mixer. The last column is the one I chose to be a word in Hanam:
CAN1 CAN2 CHOS itoi hana itoy tut put tut mis sei mis sio ioi syo ñuta itas nguta iese muki muki nanni irkop nanni iera iate yera kokiu kiuko kokyu toie ieri toye mono onom mono tsënë titsë cënë ioduo miodu yodwo
Anyway, let's back to the language. FIrst. to make it more cohesive. I should determine some sort of rules, namely phoneme and phonotactics.
For phoneme. The language will be roughly similar to middle Chinese. That means. it will have three series of stops, voiced, voiceless, aspirated.
The only fricative would be /s h/
The only liquid will be r-l which appears in variation simlar to korean, /r/ at the start of a syllable and /l/ at the end.
There will be six vowels /a e i o u @/
So... all in all the language will have the following inventory.
As for phonotactics. The language will allow C[r/y/w]VC at the most complex. With ty, thy, and dy written as c,ch,j.
For phoneme. The language will be roughly similar to middle Chinese. That means. it will have three series of stops, voiced, voiceless, aspirated.
The only fricative would be /s h/
The only liquid will be r-l which appears in variation simlar to korean, /r/ at the start of a syllable and /l/ at the end.
There will be six vowels /a e i o u @/
So... all in all the language will have the following inventory.
p t k ph th kh b d g m n ng s h r/l a e i o ë
As for phonotactics. The language will allow C[r/y/w]VC at the most complex. With ty, thy, and dy written as c,ch,j.
1st: nal
2nd: nen
3rd: gil
Reflexive: sësso
Proximal: ikol
Non-proximal: ces
Interrogative (thing): hawe
Interrogative (person): nuyu
Since all the parent languages distinguish between three state of rice, so does this language!
rice (plant): yey
rice (husked): meye
rice (cooked): papi
Let's generate a word for "eat"... ah it's "meta"
Now... since both korean and japanese is SOV with case marker. I'll make this language as such:
topic: wan / an
object: wol / ol
subject: ga / ak
locative: neyo / eyo
genitive: noy / oy
instrumental: dey
We have our first sentence in this language:
Naran papiwol meta.
Papiwan narak meta.
Puppy one...
2nd: nen
3rd: gil
Reflexive: sësso
Proximal: ikol
Non-proximal: ces
Interrogative (thing): hawe
Interrogative (person): nuyu
Since all the parent languages distinguish between three state of rice, so does this language!
rice (plant): yey
rice (husked): meye
rice (cooked): papi
Let's generate a word for "eat"... ah it's "meta"
Now... since both korean and japanese is SOV with case marker. I'll make this language as such:
topic: wan / an
object: wol / ol
subject: ga / ak
locative: neyo / eyo
genitive: noy / oy
instrumental: dey
We have our first sentence in this language:
Naran papiwol meta.
Papiwan narak meta.
Puppy one...