How Does Virtual Lie Detector Processing Work?
When your documents, files or recordings are received, they are first CODED by a professional Systems Engineer for input into VLD Inc.'s proprietary Deep Cyan™* Computational Linguistics Supercomputer. All elements of the data are then TRANSLATED into matrix and array processor OBJECTS, TOKENS, AUDIO PINS and INDEXES. These objects are then FORMATTED with an inital pass that makes them able to be PROCESSED by our proprietary, high level NLP - pattern recognition language-- XTMH™ (for eXtreme Testimony Matrix Hyperlanguage). XTMH has many proprietary and unique features, including the ability to assign timing, quantity and mathematical (rhythm) values to NLP objects, as well as array vectors, audio pins, statement animation, and geometry. Without giving away the "secret sauce," XTMH is the world's only SLAP program that translates document objects into animated agents.
The resulting FILE CONVERSION is then INTERPRETED (in XTMH) by the high level TestiPro™ software animation program. TestiPro (for Testimony Processor) is a "SLAP" program-- in the family of Supercomputer Linguistics Analysis Programs), and is as robust and visual as a modern game processor. Matrix and array processors use the "MICE" (Misunderstandings - Inconsistencies - Contradictions - Equivocation (Fabrication)) TESTS to initially flag deception, correspondence, correlation, parallelism, conjugation, independent and dependent sentence structures, term, letter and wordcount discrepancies, timing and rhythm discrepancies, and hundreds of other math, logic, KBS, Keyword and NLP program PARAMETERS as the animated vector agents interact (the statements in your recording, email, document, testimony, etc). These parameters are then passed to ANOVA engines** to generate initial PROBABILITIES of correspondence and contradiction between agents, and thus statements. Exhibits generated include our proprietary Computational Audio Pins, animations and scalar truth exhibits, and all have been found admissible under a variety of tests, including Commonwealth v. Serge, Frye and many others.
As with any high level Supercomputing program using KBS and AI, Testipro requires human iteration and intervention (course correction) at each step to evaluate and adjust minimax trees, truth likelihoods, audio pins, animations, machine weightings, etc. At each of these steps, more insight is gained by and for the Forensic Linguist. Since many of our clients are FL's, even the initial raw relationship trees, audio pins, and sentence diagrams and animations can be immediately helpful in identifying high likelihood truth and fallacy structures, and especially correspondence, contradiction, coaching, coercion and fabrication.
After all passes are complete, a team evaluates the OUTPUT and prepares a confidential REPORT for the client, using our proprietary TestiPro Veracity Scale. If the client is a Forensic Linguist or Attorney, they then INTERPRET the report results in context of their overall case, and use it, or its elements, in finalizing their own reports, exhibits, evidence, briefs, appeals, motions or expert testimony.
A Note For Techies: The Specialized Future of NLP
NLP is Natural Language Processing-- add an "S," and you get Statistical Natural Language Processing. NLP, Computational Linguistics, and many related fields within AI and KBS used to involve processing "Corpora" and corpus structures-- often large bodies of words and symbols-- with languages like PERL, Python, Prolog, etc. Rule based processors use these traditional languages, whereas newer memory based processing uses TiMBL and other custom programs.
Many specialized fields have evolved from the early efforts, from keyword searches and analysis to voice recognition, statistical list processing, analysis of sentences, building grammars, translation programs, ancient language analysis, and much more. Tools also included Natural Language Toolkits (NLTKs), WordNet, VerbNet, FrameNet, etc. Today, many of those early efforts are now spawning specialized applications like XTMH and others, field by field.
Statistical NLP takes on different meanings within Forensic Linguistics, because FL's have to deal not only with connecting statistical likelihoods between cognition, words and logic, but also between FL research and case precedents. For example, if a witness statement is back end loaded and the probability of deception goes from 8% to 90% via structure and content, the program not only has to correlate the deceptive structures with each other, but also with the relevant FL research, and the case precedents.
There is a bright future for NLP in the area of statement animation-- using game programming techniques, mathematics, statistics, information theory etc. to create intelligent agents within the corpora, then modeling their statistically likely and unlikely interactions. Within FL, these interactions then have to be additionally compared to the KBS of FL research parameters, then flagged with case precedents from the additional corpus of legal case precedents within FL, Lexis Nexis, etc.
A level deeper, veracity testing via computational linguistics is creating a new field that is only in its infancy: Temporo Spacio Linguistics, or TSL. TSL, after the fact, relates language and symbols to their interaction with time and space via statement animation, conflict resolution, collision detection, computational audio pinning, etc. Before the fact, TSL explores the ontology and epistemology of how language creates reality, including the continuum of event horizons that reaches from human evolution to words and symbols that not only describe reality, but create it.
Keywords: NLP, Natural Language Processing, Statistical Natural Language Processing, Texts, Words, Python, PERL, Prolog, ANLP, LAMP (Linux, Apache, MySQL and Perl/PHP/Python), Computational Linguistics, SLAP, SNLP, WordNet, VerbNet, FrameNet, XTMH, Nutch, LingPipe, Lucerne, Gate, Strings, Text processing, Raw Text, Normalizing, Audio Pins, Computational Audio Pins, Audio Pinning, Segmenting, List Processing, Style, Semantics, Taggers, N-Gram, Decision Trees, Chunks, Chunking, Chunkers, Classification, Mapping, Intelligent Agents, Recursion, TSL, Temporo Spacio Linguistics, KBS, Knowledge Based Systems, Extraction, XML, Propositional Logic, Order Logic, Feature structures, Feature based grammars, Parsing, Dependency Grammars, Syntax, Array processor, Matrix, Vector, Matrices, Collision detection, Statement animation, Context sensitive grammar, Morphology, Syntax, Lexicon, Part-of-speech taggers, Grammar induction, Text mining, Semantic relatedness, Dialog systems.
*Deep Cyan (DC) is a 1.5 PFLOP (Petas) custom shared network contender for Cray Jaguar's reigning position among custom dedicated SC's today (Deep Cyan processes with a direct bw of 300 Terabytes-- 300,000 Billion Bytes) core dedicated processing, plus core surrounding shared network. Originally based on molecular modeling and satellite image array processing, DC has evolved to massive parallel language stream evaluation, and now full AI driven Computational Linguistic and statement animation analysis. Our goal is 19 PF by 2012, in a combination central/ distributed processing network. Google's 500,000 servers are approaching 100 PF by 2011, which gives some indication of how complex language processing, with even simple search objectives, can become. Although XTMH is currently resident on a custom VLSI chip, it also will be made generally available in late 2011 as a software program for Computational Linguistics researchers. How much processing power is 300 Terabytes of core processing bandwidth? Each split second pass, you could process three times the entire printed collection of the Library of Congress, or 300,000 copies of Brittanica. Note that this is processing bandwidth, not just storage, as individual 300 Tera storage devices are now common at all SC facilities, and dedicated drives will be common in that size by 2011.
**Anova engines used as variance predictors, to flag statistically unlikely sentence interactions, are a generalization within models rather than between models, and collision detection in statement animation includes many other modelling criteria such as the Akaike Information Criterion, Bayesian IC, Variational Bayes, Laplace Approximation (Steepest descent / saddle integrals), etc. The keyword is FLAG. "Evidence" gleaned by statistical runs and supercomputers will be uniformly rejected by the courts if presented directly, whereas FL Exhibits, especially when animated, are routinely accepted when translated back into their original written or recorded forms, then presented as admissible animations (eg. Commonwealth v. Serge), expert testimony, or prosecution/ defense illustrative points of view, exhibits, testimony or analysis.
NOTE: As you read this website, when you look at the words in bulk, they seem to be dark green. As you scan down a specific sentence, or stare at a word, that word or sentence can appear to turn black, bolder or darker. This of course is impossible, as the site can't possibly know what you are looking at. How does this happen? Depending on your vision, you also may notice other subtle differences! Words, perception and cognition are inextricably linked. These links include tone, body language and facial expression as much as the language itself.
|
 |
| Click Picture to Order Report |
|
|
Your IP address is displayed with our compliments in the little box below. Is this via an SSI-enabled HTML file, ASP, PHP, PERL, Python, or another Server Side Script?
|
|