4.2.4.2 Phonology and Sound-to-Text
Description
This section examines emerging and developing technologies in language processing, including:
Machine Translation
Machine translation is still a developing technology and most functional systems are based on two language pairs. A
system that takes internationalization into account might have a generalized parser that can be easily modified for other language
pairs. One approach is to translate into an intermediate structure sometimes called a deep structure, and then a
translate-in program and a translate-out program can be written for each natural language. The system would be internationalized
and not dependent on a language pair.
Speech-to-Text
Most speech recognition systems today operate on a lexical system. They recognize an utterance, but cannot parse
connected speech. Some of the more modern ones can recognize connected speech and thus deal with grammatical structure. For a
connected speech product, a general parsing engine with plug-ins for supported languages constitutes a good, internationalized
design.
Text-to-Speech
Some text involves multiple scripts and languages, and so needs access to multiple dictionaries at the same time. For
phonetic output, super-segmental conditioning such as intonation might need to be included.
Pen-to-Text
As in text-to-speech, some text contains several scripts and languages and requires multiple dictionaries at once. For
some scripts, stroke direction and order are significant and can aid in determining the character written. Other scripts focus on
completed character shape; therefore, a method for determining character completion must be established.
Command Line Interface
Command line interfaces can read pen or speech based data in and return text data.
Character Interface
A character interface, like command line interfaces, take speech or pen input and produce character
output.
Graphical Interface
For graphical interfaces, speech or pen data can be layered onto graphical objects, adding a layer of complexity to the
usual graphical interface handling.
Application Protocols
Protocols can include pieces of speech as part of the protocol stream.
Storage and Interchange
Storage and interchange formats can encapsulate speech or pen data.
Application Programming Interfaces (APIs)
APIs can take speech data as parameters to calls and return speech data.
Requirements for Compliance
In general, providers must supply functions that can accommodate the lexical and syntactic manipulation needs of the
consumer. Consumers must use provider functions and manage the word and sentence structure, so as to accommodate data in any of the
provider locales. This means that the consumer must supply the provider functions with the required information for correctly
processing the data.
Command Line Interface
Providers must supply speech recognition functions for inputting and returning lexical data to the command line in any
language they support.
Consumers must use provider supplied speech or pen recognition functions, making sure to accommodate the speech or
script structures of other languages.
Character Interface
Providers must supply speech or pen input method functions for inputting and returning data to the character interface
in any character set they support.
Consumers must use provider supplied speech or pen method functions, making sure to accommodate multi-byte characters
for input and output, as well as single byte.
Graphical Interface
Providers must supply speech or pen recognition functions for managing character data with various elements of the
graphical user interface (GUI), such as buttons, drop-down lists, and title bars. These functions must accommodate all supported
character sets.
Consumers must use provider functions for creating the GUI, supplying the required language data.
Application Protocols
Providers must construct the protocol so as to accommodate the necessary speech or pen data in some specified
format.
Consumers must implement the protocol with all related character information, including charset, language, and
locale.
Storage and Interchange
Providers must allow for storage of any speech or pen data, supplying formats that contain relevant information for
proper retrieval.
Consumers must include all relevant information in the storage and interchange formats so speech or pen data in any
language can be properly retrieved.
Application Programming Interfaces
Providers must supply interfaces that accommodate any speech or pen data, where relevant.
Consumers must supply relevant speech or pen data descriptions to the API functions to properly process speech or pen
data.
|
|
Print-friendly Version
