PCFG and Adaptive FSM

I'm having some success with a first approach to machine reading. The algorithm is to convert a PCFG parse into a sequence and map this to a sequence of predicate assembly instructions using an adaptive FSM. The adaptive FSM algorithm basically maps sequences from one domain to another and utilizes patterns in the mappings of subsequences to optimize memory usage. This optimization relates to generalization in which patterns can be induced from training examples. I'll post further results as they occur; basically, the algorithm is an online parser (word-by-word) to convert sentences to predicates though the training data is structural (sentence-by-sentence).

Artificial Intelligence

After WikiCorpus and some NL algorithms for machine reading/writing, I'm thinking of moving towards language understanding (NLU) post-processing for sentences like:

“Once the characteristic numbers of most notions are determined, the human race will have a new kind of tool, a tool that will increase the power of the mind much more than optical lenses helped our eyes, a tool that will be as far superior to microscopes or telescopes as reason is to vision.”

— Leibniz

This sentence compares a described tool to microscopes and telescopes, reason to vision, and compares those two comparisons. So, after an initial semantic parsing, a rule system may permute the structured knowledge into a format more conducive to machine reasoning and NLU.

However, this is moving from NL towards artificial intelligence or, at least, machines that can demonstrate intelligent behavior.

user> Birds are to air as fish are to what?
machine> Water.
user> Why?
machine> Birds move through the air and fish move through water.

Or,

user> Telescopes are to vision as what is to reasoning?
machine> A new kind of tool that the human race will have after the characteristic numbers of most notions are determined, according to Leibniz.

This sort of NL interaction appears to require machine reading, kb search, paraphrase generation, kb search and machine writing. A goal of mine after processing text into a knowledge representation is an NL interface (requiring machine reading/writing) to demonstrate NLU. The algorithm for the above examples is tentatively all-paths discovery, path permuting and parallel search of the kb.

Pseudocode:

// n1 is to n2 as n3 is to what?
// n1 is to n2 as what is to n4?
Nodes[] IsToAsIsTo(n1,n2,n3,n4)
{
 Paths[] P = FindAllPaths(n1,n2);
 Paths[] PP = Rephrases(P);
 Nodes[] N = null;
 if( n4 == null ) N = Search(n3,PP);
 else if( n3 == null ) N = Search(PP,n4);
 return N;
}


Wikipedia Example 3

The term Artificial Intelligence (AI) was first used by John McCarthy who used it to mean "the science and engineering of making intelligent machines". It can also refer to intelligence as exhibited by an artificial (man-made, non-natural, manufactured) entity. The terms strong and weak AI can be used to narrow the definition for classifying such systems. AI is studied in overlapping fields of computer science, psychology, philosophy, neuroscience and engineering, dealing with intelligent behavior, learning and adaptation and usually developed using customized machines or computers.

Research in AI is concerned with producing machines to automate tasks requiring intelligent behavior. Examples include control, planning and scheduling, the ability to answer diagnostic and consumer questions, handwriting, natural language, speech and facial recognition. As such, the study of AI has also become an engineering discipline, focused on providing solutions to real life problems, knowledge mining, software applications, strategy games like computer chess and other video games. One of the biggest difficulties with AI is that of comprehension. Many devices have been created that can do amazing things, but critics of AI claim that no actual comprehension by the AI machine has taken place. [1]

{The term Artificial Intelligence|A} [was first used by|B] {John McCarthy|C} <who|D> [used|E] {it|F} [to mean|G] <"|H> {the science and engineering|I} [of|J] {making intelligent machines|K} <"|L>. {It|M} [can also refer to|N] {intelligence|O} as [exhibited by|P] {an artificial (man-made, non-natural, manufactured) entity|Q}. {The terms strong and weak AI|R} [can be used to narrow|S] {the definition|T} [for|U] [classifying|V] {such systems|W}. {AI|X} is [studied|Y] [in|Z] {overlapping fields|AA} [of|AB] {computer science|AC}, {psychology|AD}, {philosophy|AE}, {neuroscience and engineering|AF}, [dealing with|AG] {intelligent behavior|AH}, {learning and adaptation|AI} and [usually developed using|AJ] {customized machines|AK} [or|AL] {computers|AM}.

{Research in AI|AN} [is concerned with|AO] [producing|AP] {machines|AQ} [to automate|AR] {tasks|AS} [requiring|AT] {intelligent behavior|AU}. {Examples|AV} [include|AW] {control|AX}, [planning and scheduling|AY], [the ability to answer|AZ] {diagnostic and consumer questions|BA}, {handwriting|BB}, {natural language|BC}, {speech and facial recognition|BD}. <As such|BE>, {the study|BF} [of|BG] {AI|BH} [has also become|BI] {an engineering discipline|BJ}, [focused on providing|BK] {solutions|BL} [to|BM] {real life problems|BN}, {knowledge mining|BO}, {software applications|BP}, {strategy games|BQ} [like|BR] {computer chess and other video games|BS}. {One|BT} [of|BU] {the biggest difficulties|BV} [with|BW] {AI|BX} [is|BY] <that|BZ> <of|CA> {comprehension|CB}. {Many devices|CC} [have been created|CD] that [can do|CE] {amazing things|CF}, <but|CG> {critics|CH} [of|CI] {AI|CJ} [claim|CK] that <no|CL> {actual comprehension|CM} [by|CN] {the AI machine|CO} [has|CP] {taken place|CQ}.

B(A,C)
G(E(C,F),J(I,K))
N(M,P(O,Q))
S(R,U(T,V(_,W)))
Y+Z(X,AB(AA,AC&AD&AE&AF&AG&AH&AI))
AJ(X,AL(AK,AM))
AO(AN,AR(AP(_,AQ),AT(AS,AU)))
BE(AW(AV,AX&AY(_,_)&AZ(_,BA)&BB&BC&BD),BI(BG(BF,BH),BJ))
BE(AW(AV,AX&AY(_,_)&AZ(_,BA)&BB&BC&BD),BK(BJ,BM(BL,BN&BO&BP&BR(BQ,BS))))
BY(BW(BU(BT,BV),BX),CB)
CG(CE(CD(CC,_),CF),CK(CI(CH,CJ),!CP(CN(CM,CO),CQ)))

I'm looking into the linguistic reification of occurrence in the representation format (... CP CQ). A logic-based rule system may aid in this area.

Also interesting in this text is the passive voice, “Many devices have been created that can do amazing things.” The underscore indicates that the creator is unknown to the machine reading algorithm: CA(BZ,_) as CA corrolates to ~Created. I'm using the notation P(_,_) to represent gerunds.

One strategy in developing machine reading algorithms is to look at children's books and texts. These are already categorized by reading level, so any algorithm that appears to somehow build on itself as the reading level is increased would, philosophically, have additional merit, in my opinion.

[1] Artificial Intelligence, Wikipedia

Foundations of Machine Reading/Writing

[Leibniz][took up][the question][in][his baccalaureate thesis], and [argued][in][the true scholastic style][for][a principle of individuation] which [would preserve][the independence of universals] [with respect to][ephemeral sensations], and [yet][embodied][universal ideas][in][the eternal natures of individuals].

Let us label each bracketed chunk with a letter, A through S.

with the [yet] (O), we would have

D(B(A,C),E)
{

G(F+I(A,J),H)
K(J,M(L,N))

}
O
{

P(A,R(Q,S))

}

Theoretically resembling:

D(B(A,C),E)
G(F+I(A,J),H)
K(J,M(L,N))
P(A,R(Q,S))
O(G(F+I(A,J),H),P(A,R(Q,S)))
O(K(J,M(L,N),P(A,R(Q,S)))

but let us look at the following four and call the processing of O a metaoperation.

D(B(A,C),E)
G(F+I(A,J),H)
K(J,M(L,N))
P(A,R(Q,S))

So, in this paradigm of machine reading, the goal is to turn a sequence of chunks into a structure with necessary internal states to manage the noun chunks during the process. Transitions may occur via chunked and non-chunked tokens.

How would we describe the process of building these four predicates?

Start a new predicate with A as the left argument.
Name that predicate B.
Place C in the right argument of that predicate.
Move that predicate into the left position of a new predicate.
Name that predicate D.
Place E in the right argument of that predicate.
Start a new predicate with A as the left argument.
Name that predicate F.
Move that predicate into the left position of a new predicate.
Name that predicate G.
Place H in the right argument of that predicate.
Now back to the last predicate that we just moved into the left position of this predicate.
Add I to that predicate's label.
Place J in the right argument of that predicate.
Start a new predicate with J as the left argument.
Name that predicate K.
The next statement is set in the right argument of that predicate.
Start a new predicate with L as the left argument.
Name that predicate M.
Place N in the right argument of that predicate.
Start a new predicate with A as the left argument.
Name that predicate P.
The next statement is set in the right argument of that predicate.
Start a new predicate with Q as the left argument.
Name that predicate R.
Place S in the right argument of that predicate.

Patterns:

(×4)
Start a new predicate with <X> as the left argument.
Name that predicate <X+1>.

(×3)
Start a new predicate with <X> as the left argument.
Name that predicate <X+1>.
Place <X+2> in the right argument of that predicate.

(×2)
Name that predicate <X>.
The next statement is set in the right argument of that predicate.
Start a new predicate with <X+1> as the left argument.
Name that predicate <X+2>.
Place <X+3> in the right argument of that predicate.

“A” is the subject of the sentence, "Start a new predicate with A as the left argument" (×3)

In two of the occasions that "Start a new predicate with A as the left argument" occurs (Start a new predicate with <SUBJ> as the left argument), it is preceded by "Place <X> in the right argument of that predicate." As the sentence ends with that instruction, it is possible that in a sequence of sentences, all three occurances would be.

I theorize that in a processed document, a continuous sequence of these instructions (across sentence boundaries) would have robust and complex patterns indicative of natural writing style. Furthermore, I theorize that this methodology will be able to explain why people read active tense faster than passive and how these are processed differently in this paradigm.

Machine Reading, Paraphrases

A rule system may allow semantic subtrees to be mapped to one another. For example, the paraphrase:

Tommy put the book on the shelf to be helpful.
InOrderTo(Did(tommy,PutOn(book,shelf)),Is(tommy,helpful))

The following rule Is(X,helpful) → Help(X,_) might resemble that if something is helpful then that something help(s/ed) some thing or things. Formulating rules in this manner might transcend, in part, hermeneutic circles that occur when relating lexical elements to one another.

Hermeneutic circle refers to the semantic interconnectedness of a set of entities— for example, the definitions in a dictionary refer to other words in that dictionary. Philosophers from Schleiermacher and Dilthey through Heidegger and Gadamer considered this phenomenon. Wittgenstein said regarding this that light dawns gradually over the whole.

A rule system lexicon might further assist the semantic equivalence of paraphrases allowing semantic substructures to map to one another based on the lexical hermeneutic circle.

Like most things in AI, machine reading is more easily described than programmed. Any system that can equate noun-order paraphrases to the same set of predicates (or permutable equivalents) would be a milestone in my opinion.

Speech Technology, Semantic Web

I'm optimistic about the combination of the following technology I've been reading about:

1) Speech to text
2) Machine reading (NLP)
3) Knowledgebase / web
4) Machine writing (NLG)
5) Text to speech

Maybe someday people will be able to talk to their computers to add to and access collective encyclopedic knowledge.

Ontology and File Compression

Thinking on this post-SRL/pre-NLG ontology made me realize there was no easy way to compare different models.

If we look at rule systems, ontology and taxonomy as interoperating towards efficiently storing knowledge, then there might be a metric. That is, if system A compresses the same knowledge set better than system B and is more computationally efficient (in decompression/utilization), then we can say that system A is superior to system B (on that set) without resorting to aesthetics or philosophy. We have SUMO, the CYC upper ontology, ISO 15926, and others designed around real-world data and it's difficult to rank them.

The metaphor of file compression to knowledgebases might allow competition between differing methods. As systems are envisioned that mechanically generate rules, ontological structure or taxonomy (optimizing generators that create a system for a given knowledgebase), these metrics may be of use in comparing the resulting generated systems. Personally, I think it would be interesting to have algorithms that compress knowledgebases like tar, zip and 7zip do to files. Unfortunately, this approach is storage and speed-based and doesn't consider interface considerations-- for example, sets of things that are categorized for navigation.

Here's a link to a paper describing a relationship between AI (my field of research) and file compression. Apparently, there's a prize for compressing Wikipedia.