Machine Translation MT

1
Machine Translation MT Computer-Assisted
Translation CAT
2
Machine Translation

• Introduction to MT systems
• Generations of MT systems
• Different types of MT systems
• Construction of MT systems
• Knowledge representation
• Knowledge processing
• MT engines
• New directions of MT systems
• Evaluation of MT CAT systems

3
Machine Translation Introduction

• Machine translation ( MT) is a long-term
  scientific dream of enormous social, political
  and commercial importance.
• It was one of the earliest applications suggested
  for digital computers, but turning this dream
  into reality has turned out to be a much harder.
• Despite different problems and difficulties, some
  degree of Machine translation is now a daily
  reality and it is likely that in the future, the
  bulk of routine technical and business
  translation will be done with some kind of
  machine translation tools.

4
Machine Translation History

• The history of MT research has gone through a
  number of phases in which certain frameworks have
  dominated.
• First generation From the late 1960s the
  syntactic orientation was dominant with syntactic
  transfer approaches.
• Second generation In the 1980s the AI
  orientation was popular and more attention was
  paid to semantics.
• Third generation from 1990s the corpus-based
  model with example-based methodologies is the
  focus of much translation activity. (e.g. old
  versions of Electronic Dictionaries)
• Forth generation from 2000s research on spoken
  translation has developed into a major focus of
  MT activity. (e.g. latest versions of Electronic
  Dictionaries )
• Last ten years research on Computer-Assisted
  Translation CAT has developed into a major focus
  of translation activity.

5
How To Construct an MT system
6
Knowledge Representation

• Different kinds of knowledge are generally needed
  for Machine translation and must be represented
  in such a way it can be processed automatically
  by MTs
• Knowledge of the source language
• Knowledge of the target language
• Knowledge of the various correspondences between
  source language and target language (at least
  knowledge of how individual words can be
  translated)
• Knowledge of the culture, social conventions,
  etc.
• Etc.
• Several kinds of linguistic knowledge are usually
  distinguished
• Phonological knowledge
• Morphological knowledge
• Syntactic knowledge
• Semantic knowledge

7
Knowledge Representation Dictionary

• The central and largest component of an MT system
  is Dictionary.
• The size and quality of the dictionary limits the
  scope and coverage of a system and the quality of
  translation that can be expected.
• Electronic dictionaries of MT must at least
  represent the information we can find in paper
  dictionaries in an appropriate fashion.

8
Knowledge Representation Paper
Dictionaries
9
Knowledge Representation Electronic Dictionaries

• Entries in MT monolingual-dictionary will be
  equivalent to collection of attributes and
  values, like the following
• Lex button, catn, ntypecommon, numbersing,
  humanno, concreteyes.
• Lexbutton, catv, vtypemain, finite, person,
  number,.
• Entries can be implemented as records in a
  database.
• Entries in MT bilingual-dictionary are generally
  represented by translation rules, like the
  following
• Button ?? ?? / ???? / ???/ ??? ... ???
• This allows the replacement of certain source
  language oriented information with corresponding
  target language information.

10
Knowledge Representation Morphology

• Morphology is concerned with the internal
  structure of words and how words can be formed.
• MT CAT systems must add a morphological
  components that can recognize different word
  formation processes
• Inflection a word is derived from another word
  form by maintaining the same part of speech or
  category walk ? walks
• Describe regular inflections by general rules,
  like
• Lex walks, catv, finite, person3rd,numbersing
  , tensepres) ?? Vs
• Describe irregular inflections by explicit rules,
  like
• Lexbe, catv, finite, person3rd,numbersing,
  tensepres) ?? is
• Derivation a word of a different category is
  derived from another word or word stem by
  application of a process involving stems and
  affixes grammar, ?grammatical, arrive ?arrival
• regular derivational processes can be described
  by rules
• Irregular derivations can be solved simply by
  listing all derived words
• Compounding a new word or unit is formed by
  combination of two or more words

11
Knowledge Representation Syntax and Grammars

• Syntax is concerned with how sentences can be
  made up out of words.
• To describe syntax, a grammar (set of rules) is
  generally used in MT CAT.
• For the first kind of information, programmers
  and developers with consultations of linguists
  have to represent the concerned divisions of the
  sentence into their constituent parts and the
  categorization of these parts as nominal, verbal,
  and so on.
• Consider that in English a sentence consists of
  noun phrase followed by an auxiliary verb
  followed by a verb phrase. Noun phrase consists
  of etc. We can represent these knowledge by
  the following grammar
• S ? NP (AUX) VP
• NP ? (DET) (ADJ) N PP
• VP ? V (NP) PP
• PP ? P NP
• N ? user printer
• V ? clean
• AUX ? should
• DET ? the a
• P ? with
• a user should clean the printer is a sentence
  in the above grammar

12
Knowledge Representation Meaning

• Knowledge about the meaning of sentences are an
  important part of the translation process and
  allow MT CAT systems to produce better results.
• Three useful kinds of knowledge relating to the
  meaning can be distinguished
• Semantic knowledge meaning of words and
  sentences independently of the context they
  appear in.
• Pragmatic knowledge meaning of expressions in
  situations
• Real world or common sense knowledge
• It is useful to represent these kind of knowledge
  in MT CAT systems in order to increase their
  performance. Accomplishing this goal proved to be
  the most difficult task in the developing the MT
  CAT systems.

13
Knowledge Processing

• We give now an idea of how knowledge can be
  manipulated automatically by MT systems
• This can be done in two stages parsing and
  generation
• Parsing is the process of taking an input string
  of expressions and producing representations
  appropriate to the translation
• Generation is the process of taking an
  appropriate representation and producing the
  corresponding sentence
• A graphical representation will be used for
  parsing and generation processes. However, the
  internal representations are lists (very useful
  data structures).

14
Knowledge Processing Parsing

• The task of a parser is to take a formal grammar
  and a sentence and
• Check if it is indeed grammatical
• Show how the words are combined into phrases
• Different parsing methods exist and are
  subdivided into two categories Top-Down parsing
  method and Bottom-Up parsing method.
• Examples of parsing using grammars defined in the
  previous sections and sentence the user should
  clean the printer are given bellow.

15
Parsing Bottom-Up algorithm
16
Parsing Top-Down algorithm
17
Latest Engines in MT

• Speech Recognition MT trying to apply to MT
  techniques which have been highly successful in
  Automatic Speech Recognition.
• Computer-assisted Translation the idea is to
  collect a bilingual corpus of translation pairs
  and then use a best match algorithm to find the
  closest example to the source phrase in question.
  Ex Trados, Worsfast etc.

18
What is a CAT Tool?

• CAT stands for "Computer Aided Translation Tool".
  The terms "Translation Memory" and "TM" are
  sometimes used to refer to the same type of tool.
  A CAT tool is a computer program that helps a
  translator to work efficiently. This is
  achieved through three main functions
• A CAT tool breaks texts into segments (sentences
  or sentence fragments) and presents the segments
  in a convenient way, to make translating easier
  and faster. In some tools, for example Tardos ,
  each segment is presented in a special box, and
  the translation can be entered in another box
  right below the source text.

19

• The translation of each segment is saved together
  with the source text. Source text and translation
  will always be treated and presented as a
  translation units (TU). You can return to a
  segment at any time to check the translation.
  There are special functions which help to
  navigate through the text and to find segments
  which need to be translated or revised (quality
  control).
• The main function of a CAT tool is to save the
  translation units in a database, called
  translation memory , so that they can be re-used
  for any other text, or even in the same text.
  Through special "search" features. The search
  functions of CAT tools can also find segments
  which do not match 100. This saves time and
  effort and helps the translator to use consistent
  terminology.

20
Evaluation of MT CAT Systems

• The evaluation of MT CAT systems is a complex
  task. This is not only because many different
  factors are involved, but because measuring
  translation performance is itself difficult.
• Clarity a traditionally way of assessing the
  quality of translation is to assign scores to
  output sentences.
• Accuracy It is important to check whether the
  meaning of the source is preserved in the
  translation.
• Error Analysis tries to establish how seriously
  errors affect the translation output.
• Test Suite running the system on a large corpus
  of test texts will reveal different possible
  problems.

21
How to start using Trados?

• Steps to follow for creating, opening and
  exporting a translation memory, and further basic
  features of the software.
• You have to take into account that these steps
  correspond to SDL Trados 2006, so some menus can
  be different in other Trados versions.

22
To create a translation memory

• 1. Go to Windows / Start /All programas/ SDL
  Internacional SDL Trados 2006 / Translators
  Workbench. The software will start running and
  will request the user name.
• 2. Go to File / New.
• 3. A window will show where you have to choose
  the source and target language by clicking on
  Add. Then, click on Create.
• 4. A window will display where you have to enter
  the name for the TM and browse where to save it.
• NoteNext time you open Translators Workbench,
  the last memory used will be opened by default.

23
(No Transcript)
24
(No Transcript)
25
To open a translation memory

• There are two ways of opening a translation
  memory
• 1. You can double click on the icon of the TM you
  wish to open,
• or open Trados Translators Workbench.
• 2. Go to File / Open
• 3. A window will be displayed, were you have to
  look for the TM you want to open, and once found,
  click on it.

26

• The Trados TM will provide an existing equivalent
  sentence in the TL if it matches 100.
• The Trados TM will provide suggested words or
  phrases in different colors if the equivalent
  sentence does not match 100.
• Easily select from the possible suggestions
  offered by the MT.
• Confirm these suggestions offered by MT or
  simply type your own words or phrases.
• click Ctrl Enter to confirm and move on to
  another sentence.
• Once finished translating the whole text, click
  File . Save as .. rename the file . Saving is
  accomplished .

27
Some pieces of advice

• Dont press Enter when you are inside a
  translation unit since you can break it.
• Using the commands from the keyboard speeds up
  the job.
• If you have any problems go to Help/Help Topics
  in Translators Workbench.

28
Creating a Multiterm Termbase to Use in SDL
Trados Studio

• TWO IMPORTANT NOTES BEFORE YOU GET STARTED 1.
  Multiterm is a separate program, it's not part of
  Trados or Studio. It needs to be downloaded and
  installed separately, and it appears as a
  standalone program in your SDL folder in your All
  Programs list in Windows. If you don't see it
  there, make sure to go to your SDL account and
  download and install the program from the My
  Downloads page.2. Termbases cannot be created
  in Trados or Studio. The "Create New Termbase"
  you see in the SDL Trados main page or the
  "Terminology Management" button in the Studio
  home page are merely links that will take you to
  Multiterm, if it's installed in your computer.

29
Creating a simple Multiterm termbase

• Multiterm can be as simple or as complex as you
  want it to be. In this example, the simplest kind
  of termbase will created
• source term
  target term.
• No other index fields will be included.1. Open
  SDL Multiterm Desktop, Go to File, then select
  Create Termbase then Save your termbase in the
  dialog box that opens

30

• Click Next on Step 1 to 5 of the Termbase Wizard
  .choose your languages..Click Finish,
• In this case the termbase has been created but
  it's empty.

31

• To manually add terms, click on the Terms tab on
  the bottom left of your screen, click F3 or click
  on the Add New Entry icon right under the Edit
  menu. You will see the Entry screen, as shown
  below.

32

• Double click on the little box next to the pencil
  icon and enter the term for each entry.

33

• Press F12 to save the changes.
• The term is now part of your termbase and
  therefore will be available when you use the
  termbase in Studio.This concludes the basics of
  termbase creation.