Utemeljevanje simbolov in pragmatika v umetni inteligenci

magistrsko delo

Niko Šetar (Author), Bojan Borstner (Mentor)

Abstract

Problem utemeljevanja simbolov se prvič pojavi leta 1990, njegova osnovna formulacija pa temelji na starejšem miselnem eksperimentu, imenovanem kitajska soba. Slednji govori o tem, da lahko neki inteligentni akter obravnava nepoznane simbole na podlagi prepoznave vzorcev, brez da bi razumel pomen simbolov, ki jih obravnava, ali pomen rezultatov njegovih operacij s simboli. Za razumevanje tega, kar počne, bi moral biti zmožen te simbole utemeljiti, se pravi povezati abstraktni simbol z referentom, ki ga ta simbol predstavlja. Prvotni poskusi reševanja tega problema so se osredotočali predvsem na tako imenovane reprezentacionalistične pristope, ki poskušajo utemeljiti simbole v predstavah, ki jih ti simboli v akterju sprožijo na podlagi zaznav in izkušenj akterja. Kmalu se je pokazalo, da reprezentacija ni nujna za utemeljevanje, medtem ko je smiselno obdržati zaznave in izkušnje kot osnovo za utemeljevanje simbolov s konkretnimi, zaznavno dostopnimi referenti. Dodatni, univerzalno sprejet pogoj je, da mora biti umetni akter, ki želi doseči tovrstno utemeljevanje utelešen, saj mu to omogoča interakcijo z okoljem. V letu 2005 avtorja Taddeo in Floridi izdata prelomen članek, v katerem izpostavita, da mora uspešen pristop k utemeljevanju zadostiti pogoju, ki zahteva, da umetni akter ob začetku postopka utemeljevanja simbolov ne sme vsebovati nikakršnih semantičnih virov, lahko pa vsebuje orodja, ki mu pomagajo pri razvoju semantike skozi utemeljevanje. Kasneje Bringsjord (2014) predstavi pogoj, da mora vsak teoretični poskus utemeljevanja simbolov voditi v akterjevo zmožnost razumevanja in sodelovanja v naravni komunikaciji, kar predpostavlja razumevanje pragmatike in ostalih jezikovnih sredstev višjega reda. Ob sprejetju teh pogojev lahko poskusimo orisati koherenten teoretični mehanizem za vzpostavitev utemeljevanja simbolov v umetnih akterjih, ki se sklicuje tudi na razvoj razumevanja in utemeljevanja maternega jezika pri otrocih, ter predvideva večfazno utemeljevanje, se pravi utemeljevanje kompleksnejših simbolov v že utemeljenih, enostavnejših simbolih. V končnem delu magistrskega dela poskušamo pokazati, da lahko takšen mehanizem utemeljevanja simbolov privede do razumevanja izbranih pragmatičnih sredstev: govornih dejanj, metafor in metonimij. V ta namen pregledamo sodobno teorijo govornih dejanj in človeško razumevanje metafor in metonimij. Pokažemo tudi, da so nižja sredstva, s pomočjo katerih lahko umetni akter razume pragmatično izražanje, bodisi slovnična ali sintaktična, torej so lahko vnaprej vsebovana v umetnem akterju, ali pa semantično monosemijska, kar pomeni, da jih lahko akter utemelji enoznačno brez semantičnih dvoumnosti.

Keywords

magistrska dela;utemeljevanje simbolov;umetna inteligenca;pragmatika;umetno učenje;utelešena teorija;

Data

Language:	Slovenian
Year of publishing:	2020
Typology:	2.09 - Master's Thesis
Organization:	UM FF - Faculty of Arts
Publisher:	[N. Šetar]
UDC:	165:004.8(043.2)
COBISS:	28450563
Views:	385
Downloads:	46
Average score:	0 (0 votes)
Metadata:

Other data

Secondary language:	English
Secondary title:	Symbol grounding and pragmatics in artificial intelligence
Secondary abstract:	Symbol grounding problem first appears in 1990. Its primary formulation is based on an older thought experiment called Chinese Room. The letter speaks o fan intelligent agent, who can manipulate unknown symbol based on recognition of patterns in symbol strings, without understanding the meaning of the symbols it is manipulating, or meaning of the output of its manipulations. To understand what it is doing, it has to be able to ground the symbols, that is to say connect an abstract symbol to the referent a symbol represents. Initial attempts at solving this problem focused mainly on representationalist approaches, which strive to ground symbols in representations those symbols evoke in regard with agent's perceptions and experience. Soon it became commonly accepted that representation is not necessary for grounding, while it is sensible to maintain perception and experience as a basis in order to ground symbols with concrete, empirically accessible referents. An additional, universally recognised condition is also that an artificial agent should be embodied to achieve proper grounding, as embodiment enables its interaction with the environment. In 2005, authors Taddeo and Floridi publish a ground-breaking article in which they emphasise that a successful approach to symbol grounding must satisfy the condition that an artificial agent should not contain any semantic resources when the symbol grounding process is initiated, but may contain tools that help it develop its own semantics through the grounding process. Later, Bringsjord (2014) introduces a condition that states all theoretical attempts at symbol grounding should entail the agent's ability to understand and cooperate in natural human communication, which requires the understanding of pragmatics an other higher-order linguistic means. Having accepted these conditions, we may attempt to outline a coherent theoretical mechanism for implementing symbol grounding in arificial agents, which additionally refers to understanding and grounding native (first) language in children, and relies on multiple-phase grounding, i. e. grounding of complex symbols in already grounded simple symbols. In final part of this thesis we attempt to show that this mechanism of symbol grounding may lead to an agent's understanding of certain pragmatic means, such as speech acts, metaphors and metonymy. To do so, we review contemporary speech act theory and human understanding of metaphor and metonymy. We also demonstrate that lower-order means, by which an artificial agent can understand pragmatic expression, are either grammatical or syntactic and thus may be innately contained in an arificial agent, or semantically monosemic, which means that an agent may ground them without semantic ambiguities.
Secondary keywords:	master thesis;symbol grounding;artificial intelligence;pragmatics;machine learning;embodied theory;
Type (COBISS):	Master's thesis/paper
Thesis comment:	Univ. v Mariboru, Filozofska fak., Oddelek za filozofijo
Pages:	IX, 56 str.
ID:	11973594