Topics H: Innovative Application
Aladin: An Alignment Management Tool for Translators
Mohamed N. Lokbani
lokbani@iro.umontreal.ca
RALI/DIRO
Université de Montréal
C.P. 6128, succursale Centre-ville
Montréal (Québec)
Canada, H3C 3J7
This application summary describes the Aladin software: a tool that enables translators to manipulate bi-text. Aladin automatically calculates the correspondences between segments of two texts that are mutual translations, allows the user to verify and eventually correct and export these correspondences to a translation memory. Aladin runs under the Windows (95/98/NT) or Unix operating system.
Aladin is divided in two main modules: the alignment kernel, and the
graphical user interface.
The Alignment Kernel
Text alignment is the process of finding the links between the corresponding segments of the two texts, independently of the text units (sentence, paragraphs etc.) on which the segmentation is based. The alignment kernel has been already described in [Simard et al, 1992].
The kernel takes as input two separate text files that are mutual translations, segments both of them separately to find the sentence boundaries, and automatically calculates the alignments between the corresponding segments.
The segmentation process, first, normalizes the contents of the two text files to ensure that they adhere to the typographic conventions of the language in which they are written. The segmentation program then identifies a set of tokens: words, punctuation, numerical and symbolic expressions, and markup ; and finally recreates from these tokens, the sentences, the paragraph markups, and the main divisions of the original text, all of which will be reorganized in a set of segments.
Each pair of linked segments is called a couple, to which a score is
associated. This score represents the degree of confidence that its two
segments match. Using these scores and a dynamic programming technique,
we obtain an alignment which will explicitly make the translational correspondence
between the segments of the two texts. For example, the alignment tells
which sentence from the target text is a translation of which sentence
from the source text.
Graphical User Interface
The graphical interface allows the user to view the results of the alignment, to make some modifications or corrections to the alignment, and to save or export the alignment to a translation memory database.
When the user launches the software, he/she obtains a window vertically divided in to two parts. The user may choose to open a new session, in which case the source text will be in the left part of the window, and the translated text will be in the right part of the window. The alignment kernel, described above, is called automatically, and provides the alignment results to the graphical interface, whose task it is to display the links between each couple prominently.
Using the keyboard or the mouse and the sliders, the user may then navigate within the bi-text. The user can use the search function to look up a particular couple, or can modify the pair of segments in case the kernel provided an incorrect result. When the bi-text is approved, the user can quit and save the session, or export the alignment to a translation memory database.
The user can also re-open a previous session. In this case, the graphical
interface will not call the kernel to automatically compute an alignment
because the result has already been saved from a previous session. The
graphical interface will however allow the user to make new modifications
to the previous alignment, already saved in a previous session.
Aladin is available in both French and English, the user can choose
the preferred working language. The GUI was developed using TCL/TK programming
language, so that Aladin can run under both the Unix and Windows (95/98/NT)
operating systems.
[Simard
et al, 1992] Michel Simard, George Foster and Pierre Isabelle (1992).
Using Cognates to Align Sentences in Bilingual Corpora. In Proceedings
of the 4th Conference on Theoretical and Methodological Issues in Machine
Translation (TMI), pp. 67-82, Montréal, Canada.