A large language model (LLM) is a type of artificial intelligence (AI) algorithm that uses deep learning techniques and massively large data sets to understand, summarize, generate and predict new content.
Archives: Glossary Terms
-
Machine in the Loop (MITL)
Machine in the loop is an approach to human/machine translation. Under Machine in the Loop, a human expert (i.e. a translator) makes use of computer-based tools to support them in the translation process. This approach contrasts “human in the loop” (HITL), which only has subordinate human involvement. HITL also does not state whether the human involved is necessary an expert.
-
MT@EC
MT@EC was the European Commission’s statistical MT system that operated until 2017, when it was superseded by eTranslation.
-
MTF
For someone working in securities trading it is a multilateral trading facility, which is a self-regulated trading venue. If you are working MTF in the office, you are either in Mondays, Tuesdays and Fridays (as I am!) or Mondays, Thursdays and Fridays.
-
mynorca
A mynorca expands a word using a word for each letter to reflect the meaning of that word. So for example, a mynorca based on the word “bike” could be “best invention known ever”. Otherwise, Delta (the airline) could have a mynorca “doesn’t ever leave the airport”.
-
Normzeile
An established charging unit for translation in the German-speaking world. A Normzeile (standardised line) consists of 55 characters including spaces. Typically, it is also specified whether the actual amount is to be measured on the basis of the source or target text.
-
notional daily output
A benchmark for productivity in terms of words translated. It is particularly useful for estimating project duration for multi-day translations. Words can prove easier to count than lines, pages or characters.
-
Outcome-based translation
Translations produced to ensure a particular time-related outcome – e.g. the timely conclusion of a supervisory procedure. If outsourced, the interest is frequently to ensure that the desired outcome is achieved, and therefore commanding premium rates.
Typical examples include:
- ensuring the prompt opening of court proceedings,
- ensuring that supervisory procedures are not thwarted by a statute of limitations,
- providing information as a certified translation to expedite cross-border supervisory proceedings,
- providing official assistance to other authorities.
-
paired monitors
SDL Studio is a lot easier to use with more monitor area at your disposal. In the office I now have paired monitors – i.e. two screens that are identical in size and setup to one another, allowing me to extend my screen area horizontally.
-
passive avoidance
Methods for avoiding using a passive construction in the target translation – i.e. making an sentence active.
For example:
The penalty kick was saved the goalkeeper.
An active alternative would be:
The goalkeeper saved the penalty kick.
-
pivot language
A language used in machine translation as an intermediate step in the translation from the source language (SL) to the target language (TL) in the event that there is no bilingual engine directly between SL and TL. For example there might not be a bilingual Maltese to Finnish model, as so English is used as the pivot language. I.e. MT >>> EN >>> FI.
-
POETS Day
A British English acronym that is short for “push off early, tomorrow’s Saturday” referring to the practice, particularly where a lengthy commute is involved, of leaving the office early on a Friday, usually also associated with frontloading the working week to enable this.
-
Post-edited Machine Translation (PEMT)
Post-edited machine translation (PEMT) involves human editing of machine translation.
-
Postleitzahl
Austria uses a four digit numerical postcode system, (ie from 1000 to 9999). The first digit relates to the province (1 = Wien, 2/3 = Niederösterreich, 4 = Oberösterreich, 5 = Salzburg, 6 = Tirol/Vorarlberg, 7 = Burgenland, 8 = Steiermark, 9 = Kärnten).
-
QA
Quality assurance (QA) tools in Trados can be used for example to check that the terminology used corresponds to that in a Termbase.
-
Raw Machine Translation Output
The output from machine translation that has not been post-editted.
-
Regular Expressions
Regular expressions for translators are useful search tools for finding (and on occasion replacing) complex strings of characters. They can be used for ensuring consistent formatting, isolating cells of a certain format, and also for converting parts of TUs into non-translatable tags.
-
Retrieval-augmented generation
Retrieval-augmented generation (RAG) is a technique that LLMs use to retrieve and incorporate new information from external data sources. The LLMs firstly refer to a specified set of documents and then respond to queries by the user, with these documents acting as supplementary information to the training data that has been used to train the LLM. In the early days of LLMs this was an approach that was frequently used to add either domain-specific or updated information that the training data didn’t contain (remember those messages that ChatGPT used to deliver about its knowledge cut-off date?). RAG can be used to allow internal company data to be used, or for using authoritative sources for generating responses.
-
semantic bleaching
Semantic bleaching is the loss of semantic content. More specifically, with reference to grammaticalization, bleaching refers to the loss of all (or most) lexical content of an entity while only its grammatical content is retained. For example in Inuit there are many different terms for snow – each with embedded semantic detail (in terms of the texture, density and type of snow), which would be semantically bleached if merely translated/rendered as “snow”.
-
Single Supervisory Mechanism (SSM)
The Single Supervisory Mechanism (SSM), housed at the European Central Bank, has been operational since November 2014 and is the mechanism under which Europe’s largest banks are directly supervised by the European Central Bank – through Joint Supervisory Teams (JSTs) staffed by the European Central Bank and national competent authorities (NCAs).
See the SSM Website for further information.
-
Single-Person Language Services Unit (SPLSU)
A Single-Person Language Services Unit (SPLSU) is an in-house language services unit consisting of a single dedicated employee.
-
soft logon
Logging on to my work laptop “incognito” i.e. showing as offline, the night before I officially return to the office. I do this to complete all the updates before my first day back in the office.
-
stochastic parrot
Coined by the linguist Emily Bender, a stochastic parrot is a way “for haphazardly stitching together sequences of linguistic forms (…) according to probabilistic information about how they combine, but without any reference to meaning.”
-
termbase
A terminology database – maintained in my case in MultiTerm.