• No results found

Overview of Language Technology

2.2 ICALL Next Generation CALL

2.2.1 Overview of Language Technology

Language is one of the oldest information medium and communication system humans ever developed and used. Since early hominins started to share inten- tionality millions of years ago, languages have evolved into different language families, and it was estimated that the number of languages in the world today was up to 7000 (Hauser et al. 2002). In most of the time in human history, languages were only “spoken”. The first written language appeared after it was spoken for more than ten thousand years (Zimerle 2010). Written languages also evolved with time, for example, in less than 3500 years, Chinese written

language has evolved from oracle bone script into today’s simplified14. Besides,

population shift and national amalgamation also played important roles in lan- guage formation. With text, human knowledge was able to be maintained, even

when the language in which the text was written, was no longer spoken. Different technologies, including algorithms, computer programs and electric devices have been developed to analyze speech and text, or even to produce and modify them efficiently. Over the years, these technologies evolved and became more specialized in solving corresponding issues, as listed in Table 2.1 and 2.2.

Examples of Speech Technologies Speech Recog-

nition Transform speech into human readable text. This tech-nology is realized differently between tasks. Recognition of lists of commands or small sets of sentences can be done by pattern matching; while in order to recognize whatever is spoken (dictation), large amount of corpus and training is required.

Speaker Identi-

fication Identify the speaker from a set of known speakers, orverify if the speaker is known to the system. A typical use case is in the authentication part of the security process. Also known as voice recognition.

Speech Synthe-

sis The process to produce human speech, from either read-able text, or other representation of words and sen- tences, for example phonetic transcription. Speech syn- thesis is widely used in today’s computer software and electrical devices, as both stand-alone product and ac- cessory functionalities.

Speech Coding Compress speech data while maintaining important fea-

tures such as emotion, intonation and identity of speak- ers. Speech coding provides fundamental support for other speech technologies and boosts their efficiency. Table 2.1: Common Speech Technologies

However, language technologies are not simple summarization of speech and text technologies. On one side, there was never a clear boundary between the roles that spoken and written languages played. Traditionally, people use speech to communicate and text to record. But text has also been used in communication for thousands of years between people in distance that can’t be reached via speech, e.g. mails or note tied onto pigeons’ legs, whereas speech were able to be recorded and served as maintenance and transmission of knowledge too. As a result, technologies earlier applied to text are now found in speech, such as audio indexing and retrieval (Makhoul et al. 2000), and vice versa. On the other side, more technologies need to be involved when processing language. For example, We attach facial expression to convey our feelings as we speak. Writing styles also reflect writers’ emotion. To analyze such features, technologies like

2.2 ICALL - Next Generation CALL 24 Examples of Text Technologies

Text Catego-

rization Analyze texts and categorize them. Filtering is a specialcase of text categorization when only two categories are involved.

Text Summa-

rization Generate a much shorter version of given texts thatsummarize the main ideas. Different summarizations are generated when requirements like length and writ- ing style vary.

Text Indexing

and Retrieval Store texts in indexed database for efficient retrieval.This technique is widely used for full text search among all search engines. Performance is enhanced when com- bining with text categorization and text summarization. Information

Extraction Extract relevant information pieces, such as topics,named entities and relations between entities, from given texts.

Text Mining Analyze extracted information pieces from coherent

sources, in order to formulate conclusion or discover new information.

Question An-

swering Automatically answer questions in natural language.Answers are created by querying knowledge bases, which could be either a structured database or an unstructured collection of texts.

Translation

Technologies Translate texts between languages automatically (ma-chine translation) or assist humans while translating (computer-aided translation). Depending on the close- ness between source and target language, different meth- ods can be applied.

Table 2.2: Common Text Technologies

pattern recognition need to be applied. For disabled people, languages are not only spoken and written, but also signed and brailled. Assistive technologies are required to handle the communication and knowledge access of deaf and blind people. Furthermore, modern multimedia technologies have mixed language with more elements like pictures and videos, and wrapped them together with speech and text as a whole, thus speech and text technologies overlapping with each other are employed together with other technologies to handle multimodal communication and multimedia documents (Uszkoreit 2002).

Different methods and algorithms are employed to realize language technolo- gies, especially non-discrete mathematical methods, e.g. statistical techniques

and neural networks. Moreover, linguistic knowledge and formalisms are also utilized, among them are dictionaries, morphological and syntactic grammars, and rules for semantic interpretation. For training statistical models and testing purposes, large amount of corpus are also collected or produced, in both speech and text form.

It’s not surprising that many applications and tools with human language knowl- edge already existed and changed our lives. We use automatic correction, com- pletion and grammar checking tools when we type. We also subscribe channels in news applications to read only what we are interested in. Modern operating systems are equipped with speech input to support the operations. And digi- tal intelligent personal assists, which are pre-installed on many mobile phones, could answer to the users’ query and provide information searched from the Internet, or react to the users’ commands and generate proper feedback. But language technologies can do better. With more advanced natural language un- derstanding technology and multimodal technologies including facial expression and gesture simulation, computers are expected to become truly communicative partners, from which we can access global knowledge using natural interaction. Apart from this, language technology applications also help humans communi- cate with each other, particularly between people with different mother tongues. Although automatic translation of unrestricted texts still has a long way to reach acceptable accuracy, prepared translation is enough to deal with daily communi- cations when traveling aboard. For example, mobile applications YoChina (Xu et al. 2014) has lists of useful phrases and sentences, in which the syntactics are fixed but the information are empty (e.g. I’m _ years old). By translating, the information, that the users provide, are applied to the target language templates stored in the database. In this way, translations are guaranteed correct in both syntactic and content (Uszkoreit 2002).

In a word, language technologies have changed our lives and will keep changing them, since the ultimate goal of language technologies is barrier free communi- cation among all human beings and machines.