Bir robot koluna kumanda eden doğal dil anlama sistemi

Abstract

Doğal Dil işleme (DDİ), yapay us, dilbilim, felsefe- ve psikoloji alanlarındaki çalışmaların birleştirilmesinden oluşan bir araştırma alanıdır. DDÎ, doğal dillerin işleyişlerinin bilgisayarla yorumlanmasını ve anlaşılmasını sağlayan sistemlerin oluşturulmasını amaçlar. DDİ uygulamaları içinde doğal dil "anlamaya" yönelik programları geliştirmek önemli yer tutmaktadır. Bu programlar, ses, biçimbirim, sözdizim, anlam, uygulama ve dünya bilgilerine gerek duyarlar. Yapılan bu tez çalışmasında, ses bilgisi ve uygulama bilgisi dışındaki bilgi düzeylerinin incelenmesi, üretilmesi ve yorumlanması üzerine yöntemler geliştirilmiştir, Bu tez çalışmasında, giriş olarak komut, tanım, sorulan kabul eden ve öğrenebilen bir yazılım sistemi tanıtılmaktadır. Sistem, mevcut endüstriyel bir robot kolu üzerinde denenmiştir, ancak geliştirilen sistem hareketli robotlar için de uygulanabilir düzeydedir. Robot, bir varlığı bir noktadan bir başka noktaya taşıyabilmekte, varlıkları ayırt edebilmekte, kullanıcının gerçek dünya modeline ilişkin sorularına yanıt verebilmekte, tanımlamaları gerçek dünya modeline ekleyebilmekte ve kullanıcının "öğrettiği" sözcüklerle sözlüğünü genişletebilmektedir. Gerçeklenen sistem; sezgisel yaklaşım, biçimbirimsel çözümleyici, sözdizimsel ayrıştırıcı, anlamsal yorumlayıcı, gerçek dünya modeli, bilgisayarlı sözlük ve iç gösterimi robot komutlarına dönüştüren bir birimi içermektedir. Sözdizimsel ayrıştırıcının ürettiği işaretlenmiş sözcük öbekleri, durum gramerlerinden yararlanılarak anlamsal yorumlayıcıya aktarılmıştır. Ağırlıklı olarak robota yönelik olarak tanıtılan yöntemler ve algoritmalar DDİ konusunda yapılan çalışmanın bir uygulamasını oluşturmaktadır. Bu bağlamda, gerçeklenen sistemin çeşitli birimleri daha geniş kapsamlı DDİ uygulamalarının bileşeni olacak niteliktedir. Gerçeklenen özgün yazılım C programlama dilinde geliştirilmiştir. Robota özgü programlama dili de yazılımın içine yerleştirilmiştir.

Natural Language Processing (NLP) is a research discipline at the juncture of artificial intelligence, linguistics, philosophy, and psychology that aims to build systems capable of understanding and interpreting the computational mechanisms of natura! languages. Two primary functions which NLP systems are presentiy serving are the following:. Providing lexical and syntactic analysis tools such as style and spelling checkers; concordance, index, and table of contents generators; and textual analysis programs for purposes such as verification of authorship. in general these applications require liftle ör no semantic knowledge.. Providing the communication interface for existing systems such as inforaıation retrieval and database management systems. These applications require semantic information to interpret what the user means by his/her request and possibly even what the requested information means. Developing programs that "understand" a natural language is öne of the most challengmg tasks in artificial intelligence. Rich and Knight describe "understanding" as follows: "To understand something is to transform it from öne representation into another, where this second representation has been chosen to correspond to a set of available actions that could be performed and where the mapping has been designed so that for each event, an appropriate action will be performed....the success ör failure of an "understanding" program can rarely be measured in an absolute sense but must instead be measured with respect to a particular task to be performed." A language understanding program must have considerable knowledge about the structure of the language including what the words are and how they combine into phrases and sentences. it must also know the meanings of the words and how they contribute to the meaning of a sentence. Finally, a program must have some general world knowledge. The component forms of knowledge needed for an understanding of natural language are classified according to the following levels.. Phonological. This is knowledge which relates sounds to words. ix . Morphological. This is lexical knowledge which relates to word constructions from basic units called morphemes. A morpheme is the smallest unit of the raeaning.. Syntactic. This knowledge relates to how words are put together ör structured to form grammatically correct sentences in the language.. Semantic. This knowledge is concerned with the meanings of words and phrases and how they combine to form sentence meanings.. Pragmatic. This is high-level knowledge which relates to the use of sentences in different contexts and how the context affects the meaning of the sentences.. World. World knowledge relates to the language a user must have in order to understand and carry on a conversation. it must include an understanding of the person's beliefs and goals. The system implemented in this study makes use of ali the knowledge forms described above except for the phonological and pragmatic knowledge. The underlying task of the system designed and implemented in this thesis is to control a robot arm in real time. The natura! language front end of the system provides a dialog betvveen the user and the system. The system takes commands, questions ör declarations as input from the user and performs the appropriate tasks, such as moving a box from öne place to another ör defining the position of a point in three dimensional space. The natural language interface is utilized to solve referential ambiguities, to introduce new entities to the system, to extend the lexicon, to answer the user's questions about entities, and to inform the user about the results of the semanüc interpretation. This system consists of several components such as a heuristic approach, morphological analyzer, a syntactic parser, a semandc interpreter, and a robot driver. The components mentioned above make use of a lexicon and construct the real world representation. Both the lexicon and the world knowledge can be extended with interactive use of the system. This structure enables the system to acquire new words and to improve its world representation. The system implemented in this thesis consists of seven components. The functions of these components are explained below.. Heuristic Approach. The heuristic approach consists of seven general rules such as the Vowel Harmony Rule, Consonant Harmony Rule ör Syllabification Test which most of the Turkish words obey to. This component is used in conjunction with the morphological analyzer and the lexicon for eliminating typographical errors. If the heuristic approach encounters a word which does not obey to these general rules it warns the user and so prevents further investigation of the whole input sentence. x . Morphological Analyzer. The morphological analyzer is based on the agglutinative nature of the Turkish language.- For a language of this nature, the concept of a word is very important because a word can contain an amount of seraantic information equivalent to a coraplete sentence in another language. Each suffix in the language has a certain function and modifies the semantic information in the stem preceding it¬ in order to construct a morphological analyzer, the morphophonemics and the morphotactics of Turkish have been investigated. According to this investigation, the morphological analyzer is implemented as a root-driven, from left to right proceeding öne which consists of two parsers for nominal and verbal roots. Both of the parsers are constructed as a finite state machine and produce ali possible parses for a given word. The morphological analyzer does not handle most of the derivational suffıxes and limits the number of transitions between the two parsers. The morphological analyzer consults to the lexicon to fınd out the root of a word. After deciding whether the word being analyzed is nominal ör verbal the appropriate parser is utilized to discover the suffixes affbced to the root of the word.. Lexicon. Each item in the lexicon consists of the root of a word and the necessary flags indicating the grammatical category, type, and consonant change type of a word. The user can extend the lexicon by providing the root of a word and the values of the flags. This property brings üexibility and learning capability to the system. The set of three components mentioned above can be used for speUing checking, as well.. Syntactic Parser. The syntactic parser is a finite state machine realization of a grammar for a relative small subset of Turkish. it uses the tagged output of the morphological analyzer and separates the phrases from each other with phrase markers. The output of the syntactic parser is made available to the semantic interpreter with use of case grammars. Case grammars were fırst introduced by Filmore in 1968. He proposed that there exists a unique relation between a noun phrase and a verb that indicates the "deep case" of the phrase. A case relates to the semantic role that a noun phrase plays with respect to verbs and adjectives. Case grammars use the functional relationships between noun phrases and verbs to reveal the deeper case of a sentence. These grammars use the fact that verbal elements provide the main source of structure in a sentence since they describe the subject and objects. xi In Turkish, most of the deep cases can be identified by the cases of noun phrases or by postpositions. The deep cases are specified for each verb as 'obligatory' or 'optional' or 'not allowed'. The syntactic parser supports the free word order of Turkish. The only restriction is the assumption that the verb is at the end of the sentence. In some cases, the agent in a sentence might be optional, as well. If the agent is not contained in the input sentence except for sentences denoting declarations, the system assumes that the robot has been mentioned. This property provides the system a method for dealing with covert subjects. Semantic ^Interpreter. Semantic interpretation constructs relationships among phrases and the main verb by assigning deep cases to the phrases. This component of the system refers to the world representation to solve referential ambiguities or asks the user to supply the missing information such as the position of an object. The semantic interpretation requires the classification of verbs as explained above. According to the domain of the system, verbs related to the actions of the robot are taken and classified in this manner. The semantic interpreter also finds out the meaning of postpositions which define a position such as "üst" and "üzer" ("top"), "alt" ("under"), "sağ" ("right"), "sol" ("left"), "ön" ("front"), "arka" ("behind"). These postpositions are implemented as embedded functions. When the semantic interpreter encounters one of them, the appropriate action is triggered. The problem regarding all postpositions of this type is what we exactly "understand" from the "top of the table". The simplest approach is to define the center of the top surface of the table as the "top of the table". This approach is not satisfactory if the robot has to move two or more objects to the top of the table. In this case, all the objects would be placed at the same point without considering the free space on the table. Two solutions to this problem are proposed: The first solution is to produce randomly selected positions on the table. The second solution is to produce a new position with a predefined policy such as beginning from the center or from the left side of the top surface of the table. In both cases, previously allocated positions must be taken into account. A further function of this component is to inspect subject- verb agreement and the sequential use of postpositions. A phrase such as "masanın üstünün üstü" ("top of the top of the table") is considered as meaningless. This component also produces robot commands by considering both the status of the robot and the action to be performed. These commands are sent to the robot driver for further processing. Real World Representation. The real world representation includes three types of entities: actors, objects, and positions. Actors are capable of performing actions on objects but not on positions. The main difference between an object and a position is that a position has no dimensions. This representation of entities xu simplifies the assignment of deep structures to the phrases: The phrase denoting a location must be a position, the entity to be moved must be an object. Robot Driver. This final component transforms robot commands into electrical signals and controls the motions of the robot.