LACS building

Introduction

Context

This page is related to the construction of the Assothink passive jelly. It does not deal with the active jelly.

This page describes a theoretical approach, not the practical software processes used to build the Assothink concept universe.

This page is about :

• LACS (language-anchored concept space)
  
• LAC (language-anchored concept)
  
• Assothink concept universe building

Concepts and percepts

The passive jelly includes concepts, percepts and variants. Variants will not be discussed here.

As explained elsewhere, concepts

• are intrinsically unnamed
• are universal (most of them)
• are primary object (in the Assothink model)
• exist mainly thru the links they have with other concepts

On the other side, percepts

• are language words
• are used to represent concepts
• are not universal (because language anchored)
• are secondary (coming as perceptions of concepts)

A human being uses his brain to

• manipulate concepts(link, excite, focus on...)
  
• communicate through percepts(speak, read, write...)
  

Concept categories

Concepts are organized in categories.

They are many ways to structure concepts into categories.

The Assothink model handles 8 main categories

But in this page, only the 4 most common categories are considered:

• nouns (N) (things)
• verbs (V) (actions)
• adjectives (P) (qualifiers for things)
• adverbs (D) (qualifiers for actions)

LACS

LACS definition

A LACS is a Language-Anchored Concept Space.

A LACS is a (big) set of LAC.

It is a set of connected words and concepts.

Example of LACS (reviewed with more details below) include:

• Wordnet
• Wikipedia (and brother DBpedia)
  
• Wiktionary
• Freebase

And of course Assothink uses its own LACS, the Assothink LACS.

MoLACS and MuLACS

A MoLACS is a Mono-Language-ACS.

A MuLACS is a Multi-Language-ACS.

LACS compared

LACS may be described with various criteria

• categories handled
• MoLACS or MuLACS
• concept-centric or percept-centric
• size

The following table summarizes the properties of the most known LACS, compared to the Assothink LACS.

	Categs	Mu or Mo	Size	Centric	Remarks
Wordnet	NPVD	MoLACS  english	average	Concept centric	The brilliant precursor Concepts = synsets Weak multi-lang attempts Weak extensibility see wordnet.princeton.edu
Wikipedia	N...	MuLACS	big	Word-centric mostly	Brilliant and rich Weak organization Good growth process see www.wikipedia.org and www.dbpedia.org
Wiktionary	NPVD	MuLACS	big	Word centric	Briliant and rich Weak organization Good growth process see www.wiktionary.org
Freebase	N(pvd)xxx	MuLACS	huge	Hybrid Heterogenous	Anarchic linking Remarkably exhaustive Nice extensibility Uncontrolled growth see www.freebase.com
Assothink	NPVD	MuLACS	small	Concept centric	The best is coming! Only 10K... 30K concepts see www.assothink.com

The Assothink LACS is certainly not the biggest but it is the most demanding in terms on coherence and strength. This is necessary given the global goals of the Assothink project.

The Assothink LACS would be the first concept-centric full-NPVD MuLACS, so it is a pioneer.

LAC

LAC definition

The LAC is a Language-Anchored Concept.

A LACS is a (big) set of LAC.

LAC importance

LAC are organized differently in all LACS.

But the matching (convergence) of a LAC in LACS A with another LAC in LACS B is a critical process.

This matching process is realized easily and frequently by human beings. It is maybe a typical and major performance of the human brain. Actually something similar is done whenever a word si perceived by a human. "La tour prend le fou".

And this matching process is also the central part of the Assothink LACS building.

LAC content

A LAC unit is defined by

• the LACS it is part of
  
• a LAC identifier (an abtract, non-interpretable key)
  
• a set of LA (Language Anchor), one per language (1 in a MoLACS, many in a MuLACS)
  

And any LA contains whatever possibly links a concept to a given language:

• a 'main' word
  
• synonym words
  
• definition(s)
  
• example(s)
  
• optional hyper concept(s) (in concept-centric LACS)
  
• optional hyper word(s) (in percept-centric LACS)
  
• optional anto concept(s) (in concept-centric LACS)
• optional anto word(s) (in percept-centric LACS)
• optional context concept(s) (in concept-centric LACS)
• optional context word(s) (in percept-centric LACS)
  
• etc...

The LA contents are very different in all known LACS.

Building the Assothink LACS

The Assothink LACS is not built per se.

The main building tasks are integration and matching.

Integration

It is the process, from a given LACS, to interface, analyse, decode, classify, select, filter data from other LACS. This process is heavy in terms of managed data (the full wikipedia dumps are huge, and freebase is even much bigger).

An valuable example of integration is the integration of Wordnet by Freebase. It is not perfect, but it provides an excellent start point for nouns (but not for any other category of concepts).

Matching

It is the process of matching, from a given LACS, concepts present in other LACS to create a richer or more homegenous set of concepts.

The Wordnet matching in Freebase aims at exhaustivity.

The global matching in Assothink aims at homogeneity.

Fuzzy Logic

A matching result is necessary binary: 2 LAC from 2 distinct LACS are declared to describe the same concept - or not.

But the results gained in a matching process are widely imperfect because

• the source LACS use globally different approach, and different concept granularity. This implies that 1 LAC in LACS A matches (covers, includes) many LAC in LACS  B.
• the matching process produces errors

Thus the binary results should not be gained thru a binary logic, but rather thru a fuzzy logic. This implies likelihood measures, scoring systems, acceptance thresholds, etc...

Constraints for the Assothink LACS

The Assothink LACS aims to be a square cross-referencing universe.

It is also concept-centric full-NPVD MuLACS

This implies

1. NPVD category coverage
2. concept-centric hyper, anto, context... linkings
3. bijective cross-references with other LACS (whenever they possibly exist in other LACS)
4. multi-language anchors for all concepts

The Assothink LACS integrates selected parts of all LACS listed above.

Practical building of the Assothink LACS

Practically, the integration process used to built the noun Assothink LACS uses mainly Wordnet, Wikipedia, Freebase (and language thesauri). Freebase delivers good matching data.

But the integration process used to built the verb/adverb/adjective Assothink LACS uses mainly wordnet and wiktionary.

The process details and the software description are not covered in this page. It is a wide, complex and evolving subject.