EnParser module of the

First Working AGI Mind in JavaScript

MindForth for robots -- ghost.pl for webservers

1. Diagram of the EnParser Mind-Module

   /^^^^^^^^^\ EnParser Determines Parts Of Speech /^^^^^^^^^\
  /   EYE     \  MINDCORE               _____     /   EAR     \
 /             \ CONCEPTS              /New- \   /             \
|   _______     |   | | |      _____  (Concept)-|-------------\ |
|  /old    \    |   | | |     /Old- \  \_____/  |  Audition   | |
| / image   \---|-----+ |    (Concept)------|---|----------\  | |
| \ recog   /   |   | | |     \_____/-------|---|-------\  |  | |
|  \_______/    |  a| | |          |________V   |  c    |  |  | |
|               |  b|C| |         / EnParser \  |   a   |  |  | |
|   visual      |  s|O|f|         \__________/  |    t  |  |  | |
|               |  t|N|i|              |noun?   |     s-/  |  | |
|   memory      |  r|C|b|              |verb?   |          |  | |
|               |  a|E|e|              |adj.?   |  e       |  | |
|   channel     |  c|P|r|              |adverb? |   a      |  | |
|               |  t|T|s|              |prep.?  |    t-----/  | |
|   _______     |   | | |              |conj.?  |             | |
|  /new    \    |   |_|_|        ______V____    |  f          | |
| / percept \   |  /     \      /           \   |   i         | |
| \ engram  /---|--\ Psi /-----( InStantiate )  |    s        | |
|  \_______/    |   \___/       \___________/   |     h-------/ |

2. Purpose

EnParser serves the purpose of not only identifying a part of speech such as a noun, preposition or verb, but also of comprehending the part of speech in context by helping to assign associative tags among concepts in the Psy conceptual array. Thus EnParser and its Russian counterpart RuParser serve the purpose of Natural Language Understanding (NLU).

3. Algorithm of the EnParser AI Mind-Module

Since the main feature of the concept-based AI Minds is their demonstration of solving the AI-hard problem of natural language understanding (NLU), the EnParser module for parsing English is the main instrument of achieving the NLU goal.

The ghost.pl AI Mind in Strawberry Perl Five is becoming a conversational agent that may be installed to run in either the background or the foreground on any host computer.

The JavaScript AI Mind requires no download of either source code or programming language. Simply clicking on the link brings the tutorial AI Mind into your MSIE browser, where you may use the AI to teach students or AI coders.

If you want to issue verbal instructions to your autonomous humanoid robot, MindForth as the robot brain enables you and the robot to engage in a discussion of what work the robot should be doing, and how the work should be done, and how the robot should engage in back-and-forth communication using English for you to clarify youe instructions and for the robot to report to you its completion of tasks.

3.A. Words included in the MindBoot sequence lighten the load of EnParser.

Populating the MindBoot sequence with English words, the AI mind maintainer tries to include the most frequent English nouns and verbs, all the conjunctions, all the prepositions, and all the pronouns. Except for disambiguation, the embedded words do not need further parsing as to their parts of speech.

3.B. The first AI Minds could only parse SVO sentences into nouns and verbs.

The subject-verb-object (SVO) format of admissible inputs made it simple for the earliest AI Minds to classify English words as nouns or verbs. The parsing module could skip over the known English articles and an important adverb like the word "NOT" being used to negate an English sentence.

3.C. The ghost112.pl AI became able to parse prepositional phrases.

Since the MindBoot sequence innately contains all the English prepositions identified as such, the AI Mind easily detects the input of a known English preposition and sets the $prepcon flag to a positive number one and loads the $tpp time-of-preposition flag with the ending time-point at which the input preposition is being instantiated as a node in conceptual memory. Then $tpp is used to zero in on the associative-tag flag-panel of the preposition, filling in the next noun or pronoun as the $seq of the preposition. Then the $tvb time-of-verb flag is used to insert the concept-number of the preposition as a $seq of the verb.

3.D. The ghost114.pl AI began to parse the indirect objects of verbs.

To parse and comprehend an indirect object, EnParser tentatively fills the time-of-indirect-object $tio flag with the input-time of the first noun being input after a verb. The $tio flag is set only once by requiring that it be at zero for it to be set. Simultaneously, the time-of-direct-object flag $tdo is filled with the same value as the $tio flag for an indirect object, because it is not yet known whether one noun or two nouns are being entered subsequent to the input of a transitive verb. If and when a second noun comes in, the value for the time-of-direct-object $tdo flag, originally filled with the same value as the indirect-object $tio flag, is replaced or overwritten with the new time of the second post-verb noun in the input stream.

If only one noun comes in after the verb, the identifier of the direct object is set only once. If two post-verb nouns come in, the first noun becomes the indirect object and the second noun becomes the direct object.

4. Code of EnParser() from ghost296.pl AI source code in Perl

# 2017-09-03: Requirements of English Parser module:
# 2017-09-13: EnParser shall at first assume that a new word is a noun.
sub EnParser() {  # http://ai.neocities.org/EnParser.html 
  $act = 48;  # 2016apr27: an arbitrary activation for InStantiate()
  $bias = 5;  # 2015jun04: Expect a noun until overruled. 
  if ($fyi > 2) {  # 2016feb08: if mode is Diagnostic 
    # 2018-07-01: DIagnostic or tutorial message.
  }  # 2016feb07: end of test for Diagnostic or Tutorial mode.
  if ($pos == 5) { $bias = 8 }  # 2016feb10: after noun, expect verb.
  if ($pos == 7) { $bias = 8 }  # 2016feb10: after pronoun, expect verb.
  if ($pos == 8) { $bias = 5 }  # 2016feb10: after verb, expect noun
  if ($prepcon == 0) {  # 2017-09-13: if not handling a preposition...
    if ($pos == 5 || $pos == 7) {  # 2016mar21: expanding then-clause
      $tsj = ($t - 1);     # 2017-09-13: subject?
      if ($verbcon == 1) {  # 2017-09-13: if a verb has come in...
        if ($tio == 0) {$tdo = $tult}  # 2017-09-13: set once or twice
        if ($tio == 0) {$tio = $tult}  # 2017-09-13: set only once
        if ($tio > 0)  {  # 2017-09-13: if $tio previously set...
          $tdo = $tult;  # 2017-09-13: second noun sets time of dir.obj.
        }  # 2017-09-13: end of test to make 2nd noun the direct object.
        $tkb = $tdo;  # 2017-09-20: TEST
        my @k=split(',',$psy[$tvb]);  # 2017-09-13: expose flag-panel of verb
        $pre = $k[1];  # 2017-09-13: verb psi will be $pre of direct object
        @k=split(',',$psy[$tio]);  # 2017-09-13: expose flag-panel of indir.obj.
        $iob = $k[1];  # 2017-09-13: excerpt indirect-object concept for k[11]
        . "3,$k[8],$k[9],$k[10],$k[11],$k[12],$k[13],$k[14]";   # 2017-09-13: concatenate
        # 2017-09-13: Above lines insert k7 dba=3 for dative-case indirect object. 
        @k=split(',',$psy[$tdo]);  # 2017-09-13: expose flag-panel of dir.obj.
        $k[10] = $pre; # 2017-09-25: TEST
        . "4,$k[8],$k[9],$k[10],$k[11],0,$k[13],$k[14]"; # 2017-09-20: no $tkb for dir.obj.
        # 2017-09-13: Above lines insert verb-$psi as k10 $pre of direct object. 
        @k=split(',',$psy[$tvb]);  # 2017-09-13: expose flag-panel of main verb.
        $tdo = $t - 1;  # 2017-09-13: insert time-of-direct-object for nounlock; 
        $tkb = $tdo;  # 2017-09-13: TEST
        # 2017-09-13: Next two lines create psy-array row for a verb. 
        . "$k[7],$k[8],$k[9],$k[10],$iob,$psi,$tkb,$k[14]";      # 2017-09-13
      }  # 2017-09-13: end of test for a condition following a verb.
    }  # 2017-09-13: end of test for noun not object of a preposition.
  }  # 2017-09-13: end of test for a non-prepositional condition.
  if ($pos == 6) { $prepcon = 1; $tpp = $tult }  # 2017-09-13: prepare for noun.
  if ($pos == 5 || $pos == 7) {  # 2017-09-13: 
    if ($prepcon == 1) {  # 2017-09-13: 
      my @k=split(',',$psy[$tpp]);  # 2017-09-13: expose flag-panel of preposition;
      $pre = $k[1];  # 2017-09-13: Let $pre briefly be the preposition.
      $prep = $k[1];  # 2018-11-08: identify prep. to be the $seq of the verb. 
      $k[13] = $tult;  # 2018-11-08: establish $tkb between preposition and its object.
      . "$k[7],$k[8],$k[9],$k[10],$k[11],$psi,$k[13],$k[14]";  # 2017-06-08
      @k=split(',',$psy[$tult]);  # 2017-09-13: expose flag-panel of obj of prep.
      . "4,$k[8],$k[9],$pre,$k[11],0,$k[13],$k[14]";            # 2017-09-13
      @k=split(',',$psy[$tvb]);  # 2018-11-08: expose flag-panel verb. 
      . "$k[7],$k[8],$k[9],$k[10],$k[11],$prep,$k[13],$k[14]";  # 2018-11-08
      $prep = 0;  # 2018-11-08: Reset to prevent carry-over. 
      $prepcon = 0;  # 2017-09-13: Reset to prevent carry-over. 
    }  # 2017-09-13: end of test for a positive $prepcon.
  }  # 2017-09-13: end of test for a noun or pronoun.
  if ($pos == 8 && $psi != 818) {  # 2018-12-10: if verb other than "DO"
    $tvb = ($t - 1);  # 2017-09-13: hold onto time-of-verb for flag-insertions.
    $verbcon = 1;  # 2017-09-13: verb-condition is "on" for ind. & dir. objects.
    my @k=split(',',$psy[$tsj]);  # 2017-09-13: expose flag-panel of subject noun
    $subjpre = $k[1];  # 2017-09-13: Hold onto $subjpre for the pos=8 verb
   . "1,$k[8],$k[9],$k[10],$k[11],$psi,$tvb,$k[14]";        # 2017-09-13
    @k=split(',',$psy[$tult]);  # 2017-09-13: expose flag-panel of verb.
    . "$k[7],$k[8],$k[9],$subjpre,$k[11],0,$k[13],$k[14]";    # 2017-09-13
    $subjpre = 0;  # 2017-09-13: Reset for safety. 
  }  # 2017-09-13: end of test for a pos=8 verb.
  InStantiate(); # 2017-09-13: for creating @psi concept-nodes
}  # 2017-09-13: EnParser() returns to OldConcept() or NewConcept().

5. Variables for First Working Artificial General Intelligence

$act -- quasi-neuronal activation-level

$bias -- EnParser(); NewConcept(): expected part-of-speech POS

$fyi -- (for your information) display mode, which the user selects by pressing the Tab key and toggling through the modes: 1 = Normal; 2 = Transcript; 3 = Tutorial mode; 4 = Diagnostic mode

$pos -- (part of speech) 1=adj 2=adv 3=conj 4=interj 5=noun 6=prep 7=pron 8=verb

$pre -- pre(vious) associated @psy concept, such as the subject of a verb, or such as the verb of a direct object.

$prep -- a preposition used in the EnPrep mind-module for English prepositions.

$prepcon -- prepositional condition-flag for parsing.

$subjpre -- subject-$pre to be held for verb in parsing.

$tdo -- time-of-direct-object for a parser module.

$tio -- time-of-indirect-object for parser module.

$tkb -- time-in-knowledge-base of an idea. The EnParser() module stores time-of-verb $tvb in the k[13] $tkb slot of the subject of a sentence being stored at the time-of-subject $tsj time-point in the Psy conceptual array. The EnParser module stores $tkb as equal to $tdo (time of the direct object) in the Psy array row of the verb at time-of-verb $tvb. Thus for subjects $tkb means "verb" and for verbs $tkb means "direct object". No $tkb is stored for a direct object, because the associative tags go from subject to verb to object (SVO) but no further.

$tsj -- time-of-subject for parsing.

$tult -- t penultimate, or time-minus-one.

$tvb -- time-of-verb for parsing. The EnParser module sets "t - 1" as the tvb if the word being parsed is a verb by part-of-speech. InStantiate uses the time-of-verb tvb to fill the tkb-slot in a verb-panel with the time-location of the direct object of the verb. EnParser uses the time-of-verb tvb to find the numeric verb-concept that will be the pre of a direct object being parsed. EnParser also uses the time-of-verb tvb to store the time-of-direct-object tdo as the tkb in the flag-panel of the verb. Therefore the tvb-flag needs to be reset to zero not during but after AudInput, when program-flow has returned to the Sensorium module.

$verbcon -- verb-condition for seeking (in)direct objects. $verbcon is set to a positive unitary one in the EnParser module when a verb comes in during input and remains at one while any indirect or direct object comes in. After an input sentence, $verbcon is reset to zero.

6. Troubleshooting and Robotic Psychosurgery

6.1.a. Symptom: (Something goes wrong.)
6.1.b. Solution: (AI Mind Maintainer devises solution.)

To debug the function of assigning indirect and direct objects, the AI Mind Maintainer enters a typical sentence such as "I give the boy a robot" that contains both an indirect object and a direct object, and presses Escape to halt the AI after the input. Then the Maintainer examines the display of the conceptual array to see if $iob and $seq have been properly assigned. The AI coder may also insert diagnostic "print" messages into InStantiate() and EnArticle() so as to observe the process of assigning a tag during the input and comprehension of a sentence.

7. Future Development

- For future development. EnPronoun could be used to replace a noun with a relative pronoun and to introduce a subordinate clause, such as;

"I know a man who loves music."

EnParser must deal with the NLU of such a sentence, using the incoming relative pronoun as a trigger to treat the ensuing verb as referring back to the relative pronoun.

8. Resources

  • http://duriansoftware.com/joe/Constructing-human-grade-parsers.html

  • http://www.scholarpedia.org/article/Word_sense_disambiguation

    9. AiTree of First Working AGI Mind-Modules

    Return to top; or to
    javascript subreddit for AI Mind discussion;
    Forth subreddit for MindForth discussion;
    Perl subreddit for ghost.pl AI Mind Maintainers.
    Joint Artificial Intelligence Center (JAIC) -- the AI Manhattan Project.