The ATA Chronicle - May/June 2021 - 26

Figure 5 Unvalidated Terminology Extraction in English by Sketch Engine
available at all. On the bright

" Translation memory

terminological best practices.

Europe, compound nouns

side, plenty of invigorating

net rate schemes have four

The relationships and

longer than three words

linguistic analysis still needs

categories: Repetition:

hierarchy among the terms

should generally be avoided.

to be done and then taught

A segment repeated in

are indicated with bullets.

This is because they impede

to the machines to reach

a document; In-context

greater levels of automation,

match: A segment which

and that will keep specialists

matches a segment

y	 translation memory (4)

working far into the future.

already stored in the

y	 match (n.) (5)

carrying out term extraction

translation memory ,
including its context; the

for the first time, a great way

preceding and following

to get started is by manually

segment s or context keys

eliminating any words from

match; 100% match: A

your primary content that

segment which matches a

you know for sure are not

segment already stored in

special terms. This will help

On small projects, or when

y	 segment (7)

y	 100% match-type of
match
	 repetition (1)-type of
100% match

understanding as readers
have to pause to unpack the
modifications taking place.7
In this case, it will be fairly
clear to specialists that the
five-word compound phrase
in the text actually consists
of at least two compound
nouns: " translation memory "

	 in-context match

+ " net rate scheme. " Then

(1)-type of 100%

again, it could also consist

match; prevent

of three nouns: " translation

you develop " stop lists, "

the translation memory;
Fuzzy match : A segment

variation (i.e., " in

memory " + " net rate " +

which you can use to teach

which partially matches a

context match " and

" scheme, " where " scheme "

your term extraction tool

segment already stored in

" in-context match " )

is modified by both the terms

to automatically filter these

y	 context keys match

" translation memory " and

words out from your results. A

the translation memory; .
Fuzzy match es can range

resource like this is obviously

from 99% down to zero.

one with great potential for

Fuzzy match es below

robust growth over time.

a 75% match rate are

Here's an example of

considered new words and

a selection of text within

are newly translated. "

the domain of localization

 " Net Rate Schemes "

and the sub-domain of
CAT technology that has
undergone this process.

by Memsource
This selection of text

(1)-type of " in-

" net rate. " This identification

context match "

is not something that's easy

y	 " fuzzy match " or " 75%
match " -synonyms
	 new word (1)-per
Memsource, type of
fuzzy match
The list above contains
compound nouns, a major

for a machine to get right.
Someone who isn't qualified
might also misinterpret
this grouping or translate it
inconsistently throughout a
document, which will result
in costly rework in later
stages of production.
Determining whether terms

Repetitions are highlighted

produces the following

using various colors.

list of terms. All terms in

work. According to the

refer to unique concepts

Synonyms appear in orange

the list are presented in

Simplified Technical

or are synonyms for other

(one pair), and potential

singular form and lowercase

English specifications of

terms is another challenge

terms that were discarded

(unless the term is a proper

the AeroSpace and Defense

within terminology work,

are crossed out in red.

noun), in keeping with

Industries Association of

especially because each

26

The ATA Chronicle | May/June 2021

challenge in terminology

www.atanet.org
http://www.atanet.org
The ATA Chronicle - May/June 2021

Table of Contents for the Digital Edition of The ATA Chronicle - May/June 2021
