Page 140 - CyberSecurityEssentialsEnglish

These measures reveal aspects of understanding, knowledge, and reasoning, but they may
not represent your organization’s regulations, your audience’s dialect, your field’s
terminology, or the way dates and numbers are presented. Use them to understand the

general direction, then build a representative local test.

TABLE 3-14: A MATRIX FOR TESTING ARABIC LOCALLY

Dimension Test cases What is reviewed

Instructions, summarization, Soundness of meaning, clarity of style, and
Standard Arabic
comparison, and professional writing. the absence of literal translation.

Understanding the intent, and not
Real samples from the target dialect
Dialects converting the meaning into another
with a clear context.
dialect or distorted standard Arabic.

Consistency of terminology, and no
Sentences that mix Arabic and English
Code switching reversal of character order or invented
with field abbreviations.
expansions.

An approved glossary with correct and Translation, spelling, and consistency
Terminology
incorrect examples. across singular, plural, and abbreviation.

No distortion, no unwanted
Names and Names of people, institutions, cities,
transliteration, and no merging of two
entities and products.
entities.

Arabic and Hindi numerals,
Numbers and Value, direction, separator, unit, and no
percentages, currencies, and both
dates swapping of day and month.
calendars.

Position of numbers, order of columns,
Layout and Numbered lists, tables, and English
and copying into Word or the target
tables terms inside Arabic.
system.

Documents and A scanned PDF, small type, and a OCR, speaker separation, and linking the
audio recording with dialect and noise. result to a page or a timestamp.

Designing a Fair Sample
Do not test Arabic with five sentences translated from English. Gather real cases from your
own work, balance easy against difficult, standard against dialect, and clean text against a

realistic document. Ask native speakers to review accuracy and style, and keep the
evaluation of language separate from the correctness of the information.

When comparing tools, use the same material and the same instructions as far as possible,
and record whether the interface altered the encoding or the direction of a table. A model

may be linguistically strong while the application breaks the layout or the export; that is a
system problem, not only a model problem.

139