See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/300470210

Errors, omissions and infelicities in broadcast interpreting

Chapter · January 2011

DOI: 10.1075/btl.94.15gil

CITATIONS READS

32 1,912

1 author:

Daniel Gile
Université de la Sorbonne Nouvelle Paris 3
152 PUBLICATIONS   3,029 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Listener perception of media interpreting quality View project

All content following this page was uploaded by Daniel Gile on 25 October 2017.

The user has requested enhancement of the downloaded file.

John Benjamins Publishing Company

This is a contribution from Methods and Strategies of Process Research. Integrative

approaches in Translation Studies.
Edited by Cecilia Alvstad, Adelina Hild and Elisabet Tiselius.
© 2011. John Benjamins Publishing Company
This electronic file may not be altered in any way.
The author(s) of this article is/are permitted to use this PDF file to generate printed copies to
be used by way of offprints, for their personal use only.
Permission is granted by the publishers to post this file on a closed server which is accessible
to members (students and staff) only of the author’s/s’ institute, it is not permitted to post
this PDF on the open internet.
For any other use of this material prior written permission should be obtained from the
publishers or through the Copyright Clearance Center (for USA: www.copyright.com).
Please contact rights@benjamins.nl or consult our website: www.benjamins.com
Tables of Contents, abstracts and guidelines are available at www.benjamins.com
 Errors, omissions and infelicities
in broadcast interpreting
Preliminary findings from a case study

Daniel Gile
ESIT, Université Paris 3 Sorbonne Nouvelle, France

Recordings and transcripts of simultaneous interpretation outputs of President

Obama’s inaugural speech as broadcast by TV stations in French, German and
Japanese were scrutinized and analyzed for errors, omissions and infelicities
in an investigation of cognitive saturation and language-pair specific difficul-
ties. Inter alia, it was found that many misinterpreted or omitted source-speech
micro-units were simple and non-technical and that there were more errors
and omissions in the Japanese renderings than in either the German or French
renderings. Findings also suggest that there may be different interpreting styles
in terms of preference given to language correctness versus information com-
pleteness. All these are consistent with the tightrope hypothesis, according to
which interpreters tend to work close to cognitive saturation, which also makes
language-specific and language-pair specific idiosyncrasies relevant parameters
in the interpreting process. Findings also suggest that simple methodology and
naturalistic studies can make valuable contributions in Interpreting Studies.

Keywords: errors, omissions, infelicities, tightrope hypothesis, language-pair

specificity, naturalistic research, methodology

1. Introduction – Experimental and sophisticated versus

naturalistic and simple

Experimental research is hailed as an advanced form of investigation that makes it

possible to target specific phenomena and the effects of specific variables by control-
ling the relevant local environment. In research into translation, the most widespread
experimental paradigms over the past two decades have been think-aloud protocols
(TAPs) and computer logging, with gaze analyzers becoming increasingly popular (see
for instance Göpferich 2008). In her work on explicitation in translation, Englund

© 2011. John Benjamins Publishing Company

All rights reserved
202 Daniel Gile

Dimitrova (2005) chose to use triangulation, focusing on TAPs and computer logs. In
Chapter 3 of her book, she discusses the validity of TAPs – as does Toury (1995: 234–
238) – and highlights doubts about this paradigm in the research community. TAPs
cannot be implemented in interpreting, where other experimental online and quasi-
online methods have been used, including measurements of blood pressure, heart rate,
skin conductance, pupil diameter, EEG patterns, brain activation, cortisol levels in the
blood to name just a few (Tommola and Hyöna 1990; Klonowicz 1994; Kurz 1995;
European Parliament 2006; Ahrens, Kalderon, Krick and Reith 2010). All these meth-
ods raise issues of ecological validity because of the artificial nature of the environ-
ment in which phenomena are observed and/or because of their intrusiveness, hence
the importance of naturalistic data for cross-checking.
Another aspect of recent empirical research into both translation and interpret-
ing is the increasing use of sophisticated methods (including complex experimen-
tal designs) and technology (especially for brain imaging and other physiological
measurements). These require considerable know-how, and some require expensive
equipment as well. And yet, Translation Studies is mostly conducted within either
translator and interpreter training programs or modern language departments as
opposed to research centers with significant resources, and many investigators are
translation or interpreting instructors with little or no training in research methods
or graduate students working with them. It seems to me (as was already the case in
the 1980s – see Gile 1990), that for such colleagues, simple methods using naturalistic
data can be a good introduction to empirical research and an opportunity to hone
their skills before attempting more sophisticated studies. What is perhaps equally im-
portant is that such simple methods are still likely to yield an abundance of solid,
meaningful findings that can then be tested and fine-tuned with other designs, thus
contributing significantly to TS.
With regard to research into interpreting, until recently, few authentic source and
target speeches were available for public scrutiny, and those available were gener-
ally confined to single interpretations into single languages. Nowadays, important
speeches, press conferences and press briefings are regularly broadcast by television
stations with simultaneous interpretation, each station with its own interpreters,
and both the original delivery and several interpretations can often be found on the
internet, including the large database of European Parliament speeches interpreted
into all the EU languages (see http://www.europarl.europa.eu/activities/plenary/cre.
do?language=EN). This provides opportunities for comparisons. One pioneering
comparative study was Strolz’s doctoral dissertation on three German target speeches
(Strolz 1992) – see also Pöchhacker (1997) and Neuberger (2010). This paper presents
the initial findings from an ongoing study. It is an attempt to explore the potential of
the analysis of easily available naturalistic data to detect trends and test hypotheses
with little technology and simple methods that do not necessarily follow standard
formal procedures.

© 2011. John Benjamins Publishing Company

All rights reserved
 Errors, omissions and infelicities in broadcast interpreting 203

2. The material and the situation

The material is taken from live French, German and Japanese interpreting renditions
of US President Barack Obama’s inaugural speech delivered on January 20, 2009, on
French, German, Swiss, Austrian and Japanese television stations (France 2, LCI, TF1
and Public Sénat for France; Arte, SF1 and ZDF for German renditions; and NHK,
CNBC, TBS, Fuji and CNN Japan for Japan).
The text of the original speech is available on many internet sites (e.g. http://www.
nytimes.com/2009/01/20/us/politics/20text-obama.html), and video recordings can
also be found on the internet (e.g. http://www.youtube.com/watch?v=VjnygQ02aW4).
The speech was delivered at a speed of about 130 words per minute on average, though
when taking into account several pauses of several seconds, the actual rate during in-
dividual speaking chunks could be somewhat higher (2413 words in about 18 minutes
and 30 seconds). AIIC recommends, somewhat vaguely, an ideal presentation rate
of about 3 minutes per 40-line page of speech (http://www.aiic.net/ViewPage.cfm/
article14.htm), and more generally in the literature, a speech delivery rate of about
120 wpm is considered “normal” – for instance, in Shlesinger (2000), it was selected
as the lower rate for a comparison with a faster rate of 140 wpm. This would make
Obama’s inaugural speech somewhat faster than average.
Broadcast interpreting is a professional specialization for interpreters in some
countries, in particular in Japan, where it has often been analyzed in the literature
(see for example BS Hōsōtsūyakugurūpu 1998). In European countries, it tends to be
done by regular conference interpreters. While in some cases, specific norms apply to
broadcast interpreting as opposed to conference interpreting, in the case of Obama’s
speech, besides a high stress level (see Kurz 2002, 2003), I will assume that the work
and the interpreters’ priorities were similar to those applicable in other assignments
where speeches of important personalities are interpreted either on- or off-site. Some
parts of the speech were difficult to translate because of information density and style,
metaphors, biblical/historical allusions and citations, but many other parts pose no
particular difficulty in terms of language or encyclopedic knowledge.
The sample includes five veteran interpreters working into French, five working
into German and six working into Japanese. The analysis in this paper looks essen-
tially at the performance of

– three experienced interpreters (who had graduated from highly selective gradu-
ate training programs and/or had extensive experience interpreting speeches in
English by scientists, intellectuals and Heads of State with more than 15 years
of experience) working into their French A with a text (the performance of one
younger interpreter working without a text will only be mentioned in passing);
– two interpreters working into German with a text (the third interpreter worked
without a text, while the transcripts of the last two interpreters were only received
a week before the manuscript of this paper was sent to the editors and could
not be included); on the basis of both the audio files (kindly provided by Franz

© 2011. John Benjamins Publishing Company

All rights reserved
204 Daniel Gile

Pöchhacker), including their voice, prosody and overall performance (few hesi-
tations, false starts, pronunciation problems, etc.), and the circumstances (an
important speech interpreted in countries where professional conference inter-
preters work regularly on similar occasions), I will assume that they were also ex-
perienced professional interpreters working into their A-language (this and other
parameters should be checked further later in the project);
– five interpreters working into Japanese with a text (working conditions were dif-
ferent for the sixth team, which interpreted the speech after one day of prepara-
tion); according to Chikako Tsuruta, a veteran broadcast interpreter and professor
at Tokyo University of Foreign Studies, who kindly sent me the Japanese tran-
scripts, all five Japanese interpreters were professionals as well, with Japanese A
(native language) and English B (active foreign language) and more than 15 years
of experience.

The following important limitations arising from the context apply and have been
taken into account when selecting the methods for processing the data:

1. My A-language is French; while it seems plausible that I can reliably spot blatant
deviations from linguistic acceptability (grammatical errors, gross mispronuncia-
tions, glaring infelicities) in the French versions, no such claim can be made as
regards the Japanese and German versions.
2. For Japanese, only the transcripts were available, and I could not check them
against the recordings.
3. Working conditions were not identical in all cases. Three of the French interpret-
ers and two of the interpreters working into German had the text 20 to 40 min-
utes ahead of the actual (live) address. In Japan, all the interpreters are reported
to have had the text 20 to 40 minutes before the speech, and worked in teams of
two. Chikako Tsuruta reports that one team divided up the speech between the
two interpreters during preparation and believes that the same was done in the
other teams, as is standard practice among broadcast interpreters in Japan (per-
sonal communication). With such an arrangement, in relative terms, preparation
time for the Japanese interpreters would be at least double that available for the
other interpreters (same time but half the text for the interpreter translating the
first part of the speech, more time for the second half of the text for the second
interpreter).

3. Errors, omissions and infelicities, Effort Models

and the Tightrope Hypothesis

The conceptual framework selected for this study can be described briefly as follows.

© 2011. John Benjamins Publishing Company

All rights reserved
 Errors, omissions and infelicities in broadcast interpreting 205

3.1 The Effort Models and the Tightrope Hypothesis

The “Effort Model” for simultaneous interpreting (see an up-to-date discussion in

Gile 2009) is a conceptual framework in which cognitive operations involved in si-
multaneous interpreting are pooled into three core “Efforts”: the Listening and Analy-
sis Effort L, the Production Effort P (including self-monitoring) and the short-term
Memory Effort M. A Coordination Effort C is also required for processing capacity
management. Simultaneous Interpreting is represented as follows (the “plus” sign is to
be interpreted in a wide sense, not as an arithmetic sum of numerical values):

(1) SI = L + M + P + C

It is assumed that each of the Efforts requires some processing capacity, and that in
spite of possible “cooperation” between them, the simultaneous operation of any two
Efforts or of the three core Efforts entails an additional cognitive cost above the cog-
nitive cost of any single Effort. Most importantly, it is assumed that total processing
capacity requirements during simultaneous interpreting are close to total available
capacity (the “Tightrope Hypothesis” – see Gile 1999b, 2009).
In simultaneous interpreting with text (formula 2), which corresponds to the type
of interpreting investigated in this study, the additional Reading Effort R is assumed
to cooperate with the Listening Effort (and reduce its processing capacity require-
ments), but it also competes with it insofar as speakers often deviate from their text,
and that interpreters need to keep their attention on both the incoming speech sounds
and the text and compare the two so as to advance at the right speed and detect pos-
sible discrepancies. Moreover, in simultaneous interpreting with text, the constant
visual presence of source language words and structures before the interpreter’s eyes is
assumed to induce higher risks of language interference, and target language produc-
tion is assumed to require further processing capacity to keep the target speech free
of such interference.

(2) SI with text = L + R + M + P + C

For smooth interpreting, at any time, total available processing capacity in the inter-
preter’s brain needs to be larger than or equal to total processing capacity require-
ments; in addition, processing capacity allotted to each Effort at any time needs to be
sufficient to cover the needs associated with the task it is engaged in.
When one or more of these conditions is/are not met, performance in one or
more Effort(s) deteriorates, which may give rise to errors, omissions and/or infelici-
ties. EOIs can therefore reflect not only insufficient knowledge of the relevant lan-
guages and/or insufficient extralinguistic knowledge, but also saturation due to high
processing capacity requirements as well as processing capacity management errors.
While the particular pooling of cognitive operations into “Efforts” chosen for the
Model is only a convention, the Tightrope Hypothesis is a “theory” that requires em-
pirical support. Such support can come from several phenomena, such as:

© 2011. John Benjamins Publishing Company

All rights reserved
206 Daniel Gile

1. Many mistranslations and omissions unlikely to result from insufficient source

speech comprehension;
2. An unusually high frequency of infelicities;
3. When comparing several interpreters’ renderings of the same source speech, a
high spread of EOIs (different units of the source speech are affected in each in-
terpreter’s output) – such evidence was found in an experimental study with ten
interpreters (Gile 1999b), and again in a replication by Matysiak (2001);
4. When scrutinizing the same interpreter’s first and second rendition of the same
source speech, the presence of EOIs in the second run that were not present in
the first suggested that the lack of linguistic and extralinguistic knowledge to un-
derstand and reformulate the relevant segment adequately could not be incrimi-
nated; such evidence was also found in Gile (1999b).

3.2 Causal relationships in EOIs

Cognitive saturation can affect any and all of the Efforts, either directly (for instance,
when insufficient availability for listening leads to incomprehension of an incoming
speech segment), or indirectly, through carry-over effects (see Gile 2009), and it is
difficult to attribute individual EOIs to specific causes. However:

1. A strong norm taught in conference interpreting schools is grammatical correct-

ness in the interpreter’s output. Professionals tend to prefer omissions to ungram-
matical or unfinished sentences. When they feel cognitive load becomes too heavy
for the completion of all tasks required to produce a satisfactory output, they
may decide to omit part of the information and keep their speech grammatically
correct; omissions are often (but not necessarily) the result of tactics, whereas
grammatical errors are the result of loss of control. A large number of grammati-
cal errors in an interpreter’s output suggests a loss of control more strongly than
omissions in an output that suffers from no grammatical errors.
2. Long omissions (long clauses, full sentences and above) are often associated with
tactical choices, especially in simultaneous interpreting with text, when the in-
terpreter is lagging behind and may decide that the loss associated with such an
omission is less damaging than the risk of lagging even further behind and miss-
ing more important information.
3. While some blurring of consonants and vowels is a “natural” secondary effect of
high articulation speed, blatant pronunciation errors when articulation speed is
moderate can also be interpreted as reflecting loss of control.
4. Meaning errors can result from insufficient background knowledge or linguis-
tic knowledge, or from signal distortions (the speaker’s strong unfamiliar accent,
background noise), from cognitive saturation affecting the Listening Effort, or,
more interestingly, from a processing capacity deficit in the Production Effort. In
non-technical everyday language (as opposed to rare words or specialized terms),

© 2011. John Benjamins Publishing Company

All rights reserved
 Errors, omissions and infelicities in broadcast interpreting 207

false cognates in a skilled interpreter’s output into an A language (native lan-

guage) are likely to result from such cognitive failures.

Clearly, distinctions can be subtle and difficult to document reliably in a given re-
cording or transcript of an interpreter’s output. They are given here essentially as an
indication of how veteran conference interpreters and interpreter trainers tend to in-
terpret the relevant phenomena in the field and in the classroom. Over many years of
experience in examination committees that discuss the performance of interpreting
students and interpreters point by point, I have found that such interpretations are
highly convergent.

4. Research questions

This study seeks tentative answers to the following questions:

1. In this corpus, are there enough EOIs not likely to result from the interpreters’
insufficient linguistic or extralinguistic knowledge to suggest that at least in the
interpretation of similar speeches under similar circumstances, they are a regular
feature of interpreting performance?
2. Are EOI patterns similar for all interpreters or do they differ by individual? By
target language? Is there evidence supporting the idea that interpreting in some
language combinations is more difficult than in others?

In addition, this study seeks to assess the power of analyzing the interpreting output
at micro-unit level (the procedure is explained later in this paper) to identify patterns
and trends.

5. Materials and method

5.1 Speeches and comparisons

– The French versions found on the Internet were transcribed (by myself), and both
the recordings and the transcripts were used for analysis. In view of the objectives
of this study, as well as my linguistic limitations and the fact that I would depend
on others for other transcripts, my transcription was limited to plain text and
intonation-based punctuation (commas and periods), with no marks for prosody
and only triple-dot marks for what I perceived as hesitation pauses.
– German recordings were compared “online” to a transcript of Obama’s speech in
order to spot blatant errors and omissions. A comparison of the transcript of the
original speech with German transcripts from Neuberger 2010 was done later,
when they became available thanks to Franz Pöchhacker. The errors and omis-
sions I identified in both comparisons were the same (see below).

© 2011. John Benjamins Publishing Company

All rights reserved
208 Daniel Gile

– Only transcripts were used for the analysis of the Japanese speeches, as I did not
have access to the recordings.

5.2 Sensitivity level

One important strategic decision was where to set the sensitivity level. High sensi-
tivity detection tools pick up “weak” signals, but often at the cost of a higher risk of
mistaking “noise” for “signal,” that is, erroneously identifying irrelevant phenomena
as significant. Evidence from previous studies suggests that there is high inter-indi-
vidual variability in sensitivity to and norms of linguistic acceptability and fidelity
(Gile 1985, 1995, 1999a). In view of the nature of the questions being investigated
here, it seemed reasonable to prioritize reliability and choose a low sensitivity level.
I therefore only listed what seemed to me clear meaning errors, language errors (in-
cluding blatant infelicities) and omissions of significant bits of information. I did not
at this stage count all infelicities or omissions in cases where I thought that a particu-
lar component of the source text was possibly missing because the interpreter deemed
it insignificant, for instance when saying “we” for “America” or omitting “Today” and
just using the present tense.
Setting the sensitivity threshold at such a low level means that part of the relevant
information may not be picked up in the analysis. It may be taken up at a later stage
of the project. On the other hand, the probability that EOIs identified are “true EOIs”
is enhanced. A few tests were conducted to check this: in a research methods class-
room, a one-page extract of the transcript of one French interpreter’s rendering was
distributed to nine students who were asked to identify language errors and infelici-
ties (maladresses de langue in French). There was some variation in the reports of the
four native French speakers in the class, but all four included all my EOIs. In addition,
over several weeks, I showed colleague interpreters different pages of transcripts of
both the original and target texts in French and German and asked them to identify
meaning errors and omissions (fautes de sens and omissions), or asked them about
individual segments and their translations. All their EOIs invariably included mine.
For Japanese, I asked (graduate) students of Japanese-French-English interpreting to
identify errors and omissions in Japanese renderings on the basis of a table with a
transcript of the original speech (see Table 2 for an illustration of the micro-units
principle) and of transcripts of the Japanese target-language versions and checked
them against mine. Three reports were returned. Again, all the EO’s I identified were
also identified by the students (no infelicities were involved in this analysis).
This procedure may appear casual, and indeed it would not be sufficient if an
attempt was made to identify all EOIs for finer analyses of interpreting quality. As
explained above, in view of the objectives of this study, a low sensitivity level was
selected and the checking of my own EOI diagnoses was performed essentially as
additional insurance against what could have been idiosyncratic norms. By asking
a relatively large number of people (essentially colleagues with whom I was working

© 2011. John Benjamins Publishing Company

All rights reserved
 Errors, omissions and infelicities in broadcast interpreting 209

in the booth) their opinion on a small number of cases each time, I managed to have
more confirmations than would have been possible with a planned procedure under
strictly controlled conditions.

6. Findings

6.1 In the French sample

Table 1 shows the relative frequency of four categories of errors and infelicities. Lan-
guage errors are grammatical errors and blatant infelicities; meaning errors are target
speech segments where it seemed clear that the interpreter had misunderstood the
idea expressed by a word or group of words (see examples below). In Table 1, every
omission is counted once whether it affects a single word, a clause or a whole sentence
(another analysis is offered later with metrics of the amount of information affected).
In the recordings of the four French interpreters, the distribution of EOIs was
found as shown in Table 1.
Here are a few examples of EOIs in the French outputs:
Language errors (and blatant infelicities)
1. When translating “Homes have been lost, jobs shed”: Des foyers ont été perdus,
des emplois disparu (a verb is missing between emplois and disparu). The faulty
French grammatical structure parallels the English structure, which suggests a
lack of processing capacity for Production.
2. When translating “our schools fail too many”: Trop des enfants échouent à l’école
(des enfants should be d’enfants).
3. When translating “this is the price and the promise”: Tel est le prix et la promesse.
This is a grammatical agreement problem, and the sentence should read: Tels sont
le prix et la promesse…. The same grammatical lack of agreement is found in the
source speech, which suggests a lack of processing capacity for monitoring and
correction in Production as in Example 1.

Table 1.  Distribution of EOIs in the French-A sample

Type of EOI / interpreter F1 F2 F3 F4 F4’
Language error  7 14  5 22   32
Meaning error  6 11  5 26   37
Omission 12 34 48 34   49
Phonological distortion  2  5  3  7   10
Total 27 64 61 89 128

Note: Only the first 13 minutes of the speech are available for F4. The numbers in F4’ are a rough linear
extrapolation of what they could be if he had interpreted the whole speech. They are obviously speculative
and only indicated to give readers an idea of the magnitude of differences between the performance of the
4 interpreters.

© 2011. John Benjamins Publishing Company

All rights reserved
210 Daniel Gile

4. When translating “the father of our nation ordered these words to be read to the
people”: A annoncé que les mots suivants soient lus au peuple. A annoncé means
“announced,” and does not “agree” with the subjunctive in soient lus. The sentence
would make sense and be grammatically correct if a annoncé were to be replaced
with a ordonné (ordered). The interpreter may have thought of a ordonné while
pronouncing a annoncé. Again, this seems to indicate insufficient processing ca-
pacity for monitoring and correction in Production, as in Examples 1 and 3.

Omissions
1. “far-reaching” (… network)
2. “enduring” (… spirit)
3. “hard” (in “plowed the hard earth”)
4. “of standing pat” (in “the time of standing pat”)
5. “vital” (“… trust between a people and the government”)
6. “new” (“… threats”)
7. “and understanding” (in “greater cooperation and understanding between na-
tions”)
8. “hard-earned” (“… peace in Afghanistan”)

Meaning errors
1. Pour eux (“for them,” whereas the speaker says “for us”). An insufficient knowl-
edge of English can reasonably be ruled out as the source of this error.
2. Nous produirons des merveilles technologiques (“we will produce technological
wonders,” for “and will wield technology’s wonders” [“… to raise healthcare’s
quality and lower its cost”]. In this case, it is unlikely, but not impossible that this
highly experienced interpreter had problems understanding the use of “wield.”
3. PNB (“GNP” for “GDP”).

Salient phonological error

1. Responsibilité (for responsabilité). The incorrect vowel corresponds to the English
vowel in “responsibility,” suggesting processing capacity saturation and the in-
ability to inhibit interference from the source language.
2. Le foyage (for le voyage). There is no obvious explanation for the phonological
error here.

On the basis of these figures, the following comments can be made:

Roughly, in a speech of less than 19 minutes, there were somewhat more than 1
EOI per minute in F1, and more than three EOIs/min in F2 and F3. In other words,
blatant errors, omissions and infelicities were not exceptional occurrences in a speech
made at a relatively high delivery speed, but with the interpreters having had some
time to study the text beforehand. (F4’s EOIs are even more numerous, but since he
was far less experienced than the other interpreters and had no text, his performance
is not taken into account in this comment).

© 2011. John Benjamins Publishing Company

All rights reserved
 Errors, omissions and infelicities in broadcast interpreting 211

Table 2.  An illustrative extract from the general table identifying errors and omissions
(marked 1) on micro-units in Obama’s speech by the four French interpreters
Source speech segment / interpreter F1 F2 F3 F4
Rather, it has been the risk-takers,
the doers, the makers of things,
some celebrated
but more often
men and women
obscure in their labor,
who have carried us 1
up the long 1 1
rugged 1 1
path 1
towards prosperity 1
and freedom. 1 1
For us 1
they packed up
their few 1
worldly possessions 1
and travelled across oceans
in search of a new life.
For us
they toiled in sweatshops 1
and settled the West, 1 1
endured the lash of the whip
and plowed
the hard 1 1
earth.

F1, F2 and F3, who had graduated from highly selective graduate training pro-
grams and/or had extensive experience interpreting speeches in English by scientists,
intellectuals and Heads of State, may reasonably be assumed to have excellent com-
prehension of English. They may conceivably not be too familiar with some words
and idioms in Obama’s speech such as “when the levees break,” may have failed to
identify the reference to Paul’s Letter to the Corinthians in “the time has come to
set aside childish things” or wondered about the meaning of “country” in “the city
and the country,” but the vast majority of errors and omissions in their performance
affected simple words and grammatical structures (see examples in Table 2 and in
Section 6.3).
EOI performances vary considerably from one interpreter to the next. At this low
level of sensitivity, their number goes from 27 (for F1) to a maximum estimated at
four times that number (for F4), with F2 and F3 in-between. EOI “styles” also seem

© 2011. John Benjamins Publishing Company

All rights reserved
212 Daniel Gile

to vary from one interpreter to the next. F2 makes many language and meaning er-
rors (about 40% of the total number of EOs, vs. 53% of omissions), while F3 makes
far fewer errors (about 16% of the total number of EOs, vs. 79% of omissions). This
suggests that F3’s behavior is more tactical in that he prefers to omit segments rather
than risk making language and meaning errors, while F2 tries to translate everything
and makes more numerous errors. This impression is strengthened by the fact that
infelicities are numerous in F2 and rarer in F3 – but not being at this point in a posi-
tion to show reliable evidence of that difference, I will keep this comment anecdotal.
Note that in F4, language errors, meaning errors and omissions are more evenly dis-
tributed, with a high number of language errors, which are probably symptomatic of
a loss of control.
Having compared several renderings with the original and noted that errors and
omissions often affected single words and in particular adjectives, I then divided the
source speech into micro-units at clause and sub-clause level for the purpose of more
detailed analysis. Table 2 is an extract from the overall table drawn up and illustrates
the data. For every micro-unit, the presence (marked 1) of an error or omission (infe-
licities and language errors were not taken into account) was checked for each inter-
preter’s rendering. By adding the 1s in each column, the total number of micro-units
affected could be calculated. For instance, in the extract in Table 2, 3 micro-units
were affected in F1, 2 units were affected in F2, 9 units were affected in F3 and 3 units
were affected in F4. This procedure allows one to calculate the relative informational
weight of each omission or error (in F3, there were only 3 instances of omissions, but
one affected 6 micro-units and one affected 2 micro-units). Moreover, the tables indi-
cate how many interpreters stumbled on each micro-unit, thus helping probe deeper
into the phenomenon (see Section 6.3).
When this table for the first five minutes of the speech, up to and including “For
us, they fought and died, in places like Concord and Gettysburg, Normandy and Khe
Sahn,” is divided into 212 micro-units, the results for the four French interpreters
(Table 3) show a different pattern from that found in Table 1.
Interestingly, if the total number of EOIs were to be taken as a quality indicator,
Table 1 suggests a ranking with F1 as best, followed in turn by F3, F2 and F4; if the total
number of micro-units of the source speech affected by EOs is the criterion, F1 is still
the best for the first five minutes of Obama’s speech, followed by F2, F4 (the one recent
graduate who had no text) and F3 (now fourth instead of second, essentially because
his omissions are long and tend to affect several micro-units at a time instead of one).

Table 3.  A comparison between the total number of EOs and the number of affected
micro-units in the first five minutes of Obama’s inaugural speech – the case of the French
interpreters
Interpreter F1 F2 F3 F4
EOs in the whole speech for F1, F2, F3, in the first 13 minutes for F4 18 45 53 60
Number of affected micro-units in the first 5 minutes  5 10 26 18

© 2011. John Benjamins Publishing Company

All rights reserved
 Errors, omissions and infelicities in broadcast interpreting 213

Also note (Table 1) that F3 makes the smallest number of language errors, and his
recording shows a small number of phonological distortions, while F2 makes more
and F4 many more. To listeners, F3 might well have been the most pleasant inter-
preter to listen to after F1, as may be checked in a later phase of the project.

6.2 Inter-language comparisons

The issue of language-specificity in interpreting is controversial. Some theorists claim

that if interpreters have the required skills, language-specific and language-pair spe-
cific factors such as differing syntactic structures should have little or no impact on
interpreting difficulty. Setton (1999) reports that in his trilingual corpus with English,
German and Chinese, he found no obvious effects of syntactic differences in interpret-
ing between different language pairs. Another view is that because of high processing
capacity requirements in interpreting, such differences are significant (see a more de-
tailed discussion of the issue in Gile 2009).
Syntactically, lexically and culturally, French could be considered closest to
English, followed by German and then by Japanese (see Wilss 1978; Kurz 1983; Ito-
Bergerot 2006; Kondo 2008). Two aspects of EOIs will be considered here: the over-
all number of micro-units affected by EOs in each target-language sample, and the
number of “long omissions” (affecting several successive micro-units of the source
speech).
Table 4 shows the number of micro-units in the first 5 minutes of source speech
affected in each rendition. F1 to F3 are French interpreters and their renditions, G1
and G2 are German-speaking interpreters and their renditions, J1 to J5 are Japanese-
speaking interpreters and their renditions:
The mean number of affected micro-units in these first 5 minutes of the speech
is lowest in French (with high inter-individual variability), and almost doubles on
average in the German version. Scores are also highly variable in the Japanese sample,
with J1 and J5 within the range of values for the French sample, but J2, J3 and J4 far
above them. In view of the assumptions, mentioned earlier, that the Japanese inter-
preters were all professionals with at least 15 years’ experience and that they had more
time to prepare for their part of the speech than the French and German interpreters,
these data seem to support the idea that language-pair-specific differences can indeed
have an impact on the difficulty of interpreting.

Table 4.  The number of micro-units of the first five minutes of Obama’s speech affected
by EOs in French, German and Japanese renditions
Interpreter F1 F2 F3 G1 G2 J1 J2 J3 J4 J5
Number of micro-units affected by EOs  5 10 26 25 26 13 73 50 57 18
Mean value by language 13.7 25.5 42.2

© 2011. John Benjamins Publishing Company

All rights reserved
214 Daniel Gile

Table 5.  Number of long omissions by interpreter and by language

Interpreter F1 F2 F3 G1 G2 J1 J2 J3 J4 J5
Number of chunks ≥ 3 micro-units 0 1 4 0 3 0 8 3 8 0
Mean by language 1.7 1.5 3.8
Number of chunks ≥ 4 micro-units 0 0 3 0 2 0 4 3 4 0
Mean by language 1.0 1.0 2.2
Number of chunks ≥ 5 micro-units 0 0 2 0 0 0 4 2 3 0
Mean by language 0.7 0 1.8

Analysis of EOs by the number of long omissions (“chunks” of 3, 4 and 5 con-

secutive micro-units respectively) generates a clear “hierarchy” between French and
German on one hand and Japanese on the other, but not between French and German
(Table 5). Since the samples are very small and because of other uncontrolled variabil-
ity factors, these can only be considered tentative indications, especially in view of the
high inter-individual variability in each target-language sample, but they do suggest
that simultaneous interpreting from English into Japanese may be cognitively more
taxing than work into German or French.

6.3 Frequent EOs

Finally, in connection with the question of the relative “interpreting difficulty” of

speech segments affected by EOIs: in the first five minutes of the speech, 132 micro-
units were omitted or mistranslated by the 10 experienced interpreters who worked
with the text into their native language. As can be seen from Table 6, more than a
third of them are found in no more than one interpreter’s rendition, about two thirds
(66%) in no more than two interpreters, and 91% in no more than four interpreters,
which indicates a large spread of the EOs over the speech and replicates one finding
of Gile (1999b).
As to the micro-units most affected by EOs, they are:

1. “fail too many” (in “our schools fail too many”), in eight interpreters out of ten
2. “long” (in “who have carried us up the long, rugged path”), in seven interpreters
3. “far-reaching” (in “a far-reaching network”), in six interpreters
4. “the ways we use” (in “the ways we use energy”), in six interpreters
5. “rugged” (in “who have carried us up the long, rugged path”), in six interpreters
6. “hard” (in “plowed the hard earth”), in six interpreters.

As can be seen, none of these micro-units are linguistically difficult or require special
thematic knowledge, though understanding exactly what is meant by some of them
may require some analysis, and finding an appropriate equivalent in the target lan-
guage may also cost effort.

© 2011. John Benjamins Publishing Company

All rights reserved
 Errors, omissions and infelicities in broadcast interpreting 215

Table 6.  Number of interpreters who omitted or mistranslated each micro-unit

in the first 5 minutes of the speech
Number of EOs found in … 1 2 3 4 5 6 7 8 >8
interpreter(s) out of 10 int. ints. ints. ints. ints. ints. ints. ints. ints.
Number of EOs 50 37 21 12 7 4 1 1 0
Percentage of EOs 38% 28% 16% 9% 5% 3% .8% .8% 0%

As indicated in Section 3.1, all these findings are consistent with the idea that
processing capacity management difficulties and/or cognitive saturation, as opposed
to poor linguistic and/extralinguistic knowledge, cause many EOs.

7. Methodological comments and conclusion

The research procedure followed here leaves some uncertainty: Ideally, all relevant
details about each interpreter and about the specific conditions under which s/he pre-
pared for the assignment and interpreted the speech, including the physical environ-
ment, available help from a colleague on site, type of headset used, type of monitor
available to view President Obama’s speech, quality of the sound, etc., would have
been known and ascertained as comparable. Ideally, all transcripts would have been
performed under and checked against a set of unique transcription rules by a team of
native speakers of French, German and Japanese.
Also, as mentioned earlier, the low sensitivity level chosen for EOI detection here
can only detect part of the picture, and subtle differences in the behavior of both indi-
vidual interpreters and language groups may well have gone unnoticed.
Nevertheless, in view of the research questions addressed, I believe that above
the noise left by uncertainty, the data that emerge allow the following tentative con-
clusions:

1. The large number of EOIs that were not likely to be due to insufficient linguistic
or extralinguistic knowledge is consistent with the hypothesis that cognitive satu-
ration is involved.
2. Different interpreting “styles” with respect to priority given to completeness of
information versus high linguistic quality of the output may be characteristic of
specific interpreters.
3. Differences in the numbers of EOs and long omissions suggest that it was more
difficult to interpret the English speech into Japanese than into French or German
(differences between French and German renderings need to be investigated fur-
ther). This lends some empirical support to the idea that under the heavy cogni-
tive pressure of interpreting, linguistic idiosyncrasies and language-pair specific
idiosyncrasies do matter.

© 2011. John Benjamins Publishing Company

All rights reserved
216 Daniel Gile

Going beyond this specific study towards research policy:

1. For practical reasons, it would be difficult to organize an experiment in which the

output of French, German-speaking and Japanese interpreters could be examined
as was done in this study. If only for this reason, the simple, naturalistic method
chosen here, which can easily be applied to other languages and provide interest-
ing data that would be difficult to collect otherwise, deserves attention.
2. The type of investigation described here can be replicated easily by individual re-
searchers with virtually no technical and financial means other than a computer
with access to the internet. The large amount of material available on the internet
in many languages, as well as individual recordings of TV interpretations, open
up possibilities for many case-study type replications that can be more powerful
in answering questions than a small number of experiments.
3. Finally, performing this type of study for graduation theses can be appealing to
interpreting students who tend to shy away from research, because it involves de-
tailed scrutiny of genuine interpretations that they can use for learning purposes
and never takes them into highly abstract theoretical considerations. My own in-
terpreting students were happy to look at the transcripts and help, though at ESIT
they are not required to conduct research.

References

Ahrens, B., Kalderon, E., Krick, C. M. and Reith, W. 2010. “fMRI for exploring simultaneous
interpreting.” In Why Translation Studies Matters, D. Gile, G. Hansen and N. K. Pokorn
(eds), 237–248. Amsterdam/Philadelphia: John Benjamins.
AIIC. Guideline for Speakers. http://www.aiic.net/ViewPage.cfm/article14.htm. (Accessed June
12, 2010).
BS Hōsōtsūyakugurūpu (BS 放送通訳グループ). 1998. Hōsōtsūyaku no sekai (放送通訳の世
界) (The World of Broadcast Interpreting). Tokyo: Ark.
Englund Dimitrova, B. 2005. Expertise and Explicitation in the Translation Process. Amsterdam/
Philadelphia: John Benjamins.
European Parliament. 2006. Report on the 3rd Remote Interpretation Test, Brussels, 22.11–
1.12.2005. Executive Summary.
Gile, D. 1985. “La sensibilité aux écarts de langue et la sélection d’informateurs dans l’analyse
d’erreurs : une expérience.” The Incorporated Linguist 24 (1): 29–32.
———. 1990. “Research proposals for interpreters.” In Aspects of Applied and Experimental Re-
search on Conference Interpretation, L. Gran and C. Taylor (eds), 226–236. Udine: Cam-
panotto Editore.
———. 1995. “Fidelity assessment in consecutive interpretation: an experiment.” Target 7 (1):
151–164.
———. 1999a. “Variation in the perception of fidelity in simultaneous interpreting.” Hermes
22: 153–172.

© 2011. John Benjamins Publishing Company

All rights reserved
 Errors, omissions and infelicities in broadcast interpreting 217

———. 1999b. “Testing the effort model’s tightrope hypothesis in simultaneous interpreting – a
contribution.” Hermes 23: 51–79.
———. 2009. Basic Concepts and Models for Interpreter and Translator Training. Revised Edi-
tion. Amsterdam/Philadelphia: John Benjamins.
Göpferich, S. 2008. Translationsprozessforschung. Stand – Methoden – Perspektiven. Tübingen:
Gunter Narr Verlag.
Ito-Bergerot, H. 2006. “Le processus cognitif de la compréhension en interprétation consécu-
tive : acquisition des compétences chez les étudiants de la section japonaise.” (Doctoral
dissertation, ESIT, Université de la Sorbonne Nouvelle, France.)
Klonowicz, T. 1994. “Putting one’s heart into simultaneous interpretation.” In Bridging the Gap.
Empirical Research in Simultaneous Interpretation, S. Lambert and B. Moser-­Mercer (eds),
213–224. Amsterdam/Philadelphia: John Benjamins.
Kondo, M. 2008. “Multiple layers of meaning – towards a deepening of the “sense” theory of
interpreting.” In Towards Quality Interpretation in the 21st Century. Proceedings of the 6th
National Conference and International Forum on Interpreting, E. Wang and D. Z. Wang
(eds), 35–40. Beijing: Foreign Language Teaching and Research Press.
Kurz, I. 1983. “‘Der von uns…’: Schwierigkeiten des Simultandolmetschens Deutsch-Englisch.”
In Festschrift zum 40 Jährigen Bestehen des Instituts für Übersetzer- und Dolmetscherausbil-
dung der Universität Wien, 91–98. Tulln: Ott Verlag.
———. 1995. “Watching the brain at work – an exploratory study of EEG changes during simul-
taneous interpreting (SI).” The Interpreter’s Newsletter 6: 3–16.
———. 2002. “Stressfulness of live TV interpreting vs. conference interpreting.” In Transla-
tion: New Ideas for a New Century. Proceedings of the XXVI FIT Congress, Vancouver, BC,
Canada, August 7–10, 2002. 202–206.
———. 2003. “Live TV interpreting. A high-wire act?” In Avances en la investigación sobre inter-
pretación, Á. Collados Aís and J. A. Sabio Pinilla (eds), 159–171. Granada: Comares.
Matysiak, A. 2001. “Controlled processing in simultaneous interpretation: A study based on
Daniel Gile’s Effort Models.” (Master’s thesis, University of Poznan, Poland.)
Neuberger, B. 2010. “Mediendolmetschen und Rhetorik – eine korpusbasierte Analyse.” (Mas-
ter’s thesis, University of Vienna, Austria.)
Pöchhacker, F. 1997. “Clinton speaks German: A case study of live broadcast simultaneous in-
terpreting.” In Translation as Intercultural Communication – Selected Papers from the EST
Congress – Prague 1995, M. Snell-Hornby, Z. Jettmarova and K. Kaindl (eds), 207–216.
Amsterdam/Philadelphia: John Benjamins.
Setton, R. 1999. Simultaneous Interpretation: A Cognitive-Pragmatic Analysis. Amsterdam/Phil-
adelphia: John Benjamins.
Shlesinger, M. 2000. “Strategic allocation of working memory and other attentional resources
in simultaneous interpreting.” (Doctoral dissertation, Bar Ilan University, Israel.)
Strolz, B. 1992. “Theorie und Praxis des Simultandolmetschens. Argumente für einen kontex-
tuellen top-down Ansatz der Verarbeitung und Produktion von Sprache.” (Doctoral dis-
sertation, Vienna University, Austria.)
Tommola, J. and Hyönä, J. 1990. “Mental load in listening, speech shadowing and simultaneous
interpreting: A pupillometric study.” In Foreign Language Comprehension and Production,
J. Tommola (ed.), 179–188. AFinLA Yearbook, Turku Publication of the Finnish Associa-
tion for Applied Linguistics 48.

© 2011. John Benjamins Publishing Company

All rights reserved
218 Daniel Gile

Toury, G. 1995. Descriptive Translation Studies and Beyond. Amsterdam/Philadelphia: John

Benjamins.
Wilss, W. 1978. “Syntactic anticipation in German-English simultaneous interpreting.” In Lan-
guage Interpretation and Communication, D. Gerver and H. Wallace Sinaiko (eds), 343–
352. New York/London: Plenum Press.

© 2011. John Benjamins Publishing Company

All rights reserved
View publication stats