View unanswered posts | View active topics It is currently Sun Nov 18, 2018 2:08 pm



Reply to topic  [ 5 posts ] 
 ai/nlp/ 
Author Message
Rank 6
Rank 6

Joined: Wed May 17, 2017 7:35 am
Posts: 25
This NLP script looks like it was written by someone who learned English solely from the internet. If it's your second language that's quite an achievement, but if it's your first then it would behoove you to stay awake in class more often. The rules of grammatical syntax have the same logical consistency as a programming language, and if you studied them from that perspective you would find it simple to port them to executable code. Parts of speech need to be clearly defined in order to define "natural" sentence structures.

Aside from the grammar, there are a lot of mistaken entries:

This in "nlp_greet.txt":
Code:
#greetval = <hee>       : hello
"hee" does not mean "hello". Someone may deduce from context that it's your greeting, but at best they'll think you're idiosyncratic, at worst they'll think you're mentally disabled.
This in "nlp_verb.txt":
Code:
#event = <sm>             : bsdm
I don't know what 'bsdm' is, but it's probably not kink related, and definitely doesn't mean BDSM (Bondage Discipline Sado-Masochism)

Code:
#event = <hardcore>       : sex
#event = <hentai>         : sex
I hate to break it to you lifestyle fappers, but hardcore and hentai aren't sex, they're genres of porn. Watching porn isn't sex, though any accompanying autoeroticism would be.

Code:
#doi     = <doing>   : do
The 'ing' forms aren't syntactically verbs, because they are in gerund form they function as nouns, and require the use of an auxiliary verb. This is one reason auxiliary verbs should be distinctly defined from action verbs. 'Do' actually is an auxilary verb, and 'do doing' is grammatically correct but linguistically awkward (not natural language.)

Code:
#verb  = <anal>        : anal
#verb  = <blow>        : blowjob
"Anal" and "Blowjob" aren't verbs, and it isn't natural language to use them as verbs (and "blowjobbing" isn't correct or natural!) "Sodomy" and "Fellate" are verbs, though not used as naturally as many colorful euphemisms and dysphemisms.

This in "nlp_rewrite.txt":
Code:
<sex>       : <fuck>
"Sex" isn't a verb, and it's not equivalent to "fucking". It's a noun, and "having sex" can refer to any sexual activity, while "fucking" means penetration.

Quote:
<u> : <you>
This may be a user convenience, but one letter words are so bound to be error prone that you could save yourself headaches by minimizing the use of them. The only ones you can't get away from are the pronoun "I" and the article "a", and their contexts can be more easily defined.


Thu May 18, 2017 4:50 am
Profile
Site Admin
Site Admin

Joined: Thu Feb 07, 2013 11:16 am
Posts: 1853
Good points,

thanks


Thu May 18, 2017 8:52 am
Profile
Rank 2
Rank 2

Joined: Sun May 14, 2017 10:36 pm
Posts: 5
how is the project going for 4.0 any updates is it stilll alive or canceled?


Thu May 18, 2017 10:45 pm
Profile
Rank 6
Rank 6

Joined: Wed May 17, 2017 7:35 am
Posts: 25
I see a handful of search patterns that use a form of 'be' in the same sentence as 'is'. That's probably the most irregular verb in the language. 'to be' is the infinitive form of 'is', and 'be' is only used as the imperative form, eg "Be quick or be dead." Someone who uses 'be' in the present tense is probably using it ironically, because even foreign speakers don't make that mistake.

This pattern:
Code:
<#actor * is    * been * VERBi OBJ>

isn't natural because 'been' is the past participle of the same verb. "#actor was" would be most natural there., and the only auxiliary verb that could go with 'been' is 'has/had/have' "How have you been?"

I think your patterns in general are way too liberal with the wildcards for filtering natural language. If you start with the assumption that garbage is coming in, then you've already contaminated your output quality. If you process language the way people naturally do, then you could quicken your word searches and formulate more plausible and comprehensive outputs. People are receptive to patterns, and from the first word they begin anticipating the next by word form, whether they consciously know the rules or not. Start searches by the most common parts of speech a sentence can begin with and that will inform you of the next likely parts of speech.

If a speaker insists on using unnatural language it will take longer to find words and patterns, but simulated personalities aren't obligated to rescue the semantic survivors from the wreckage of a speaker's communication, so you could terminate the syntax matching at any time according to relationship values and the listener's perception of the speaker. Just translate the semantic values of words already identified to determine whether the listener would reply with a curt dismissal or an enraged rant.


Sun May 28, 2017 4:45 pm
Profile
Rank 16
Rank 16
User avatar

Joined: Wed Aug 07, 2013 4:51 pm
Posts: 306
IMO the issues with NLP are:

-unusable with VR
-unusable if you only got one hand free (*cough*)
-not self explaining ("what can I type?")

The fact that the patterns have been implemented badly is probably not even important in this case.


Tue May 30, 2017 10:56 am
Profile
Display posts from previous:  Sort by  
Reply to topic   [ 5 posts ] 

Who is online

Users browsing this forum: No registered users and 0 guests


You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot post attachments in this forum

Search for:
Jump to:  
Powered by phpBB® Forum Software © phpBB Group.
Designed by X-Moon Productions.