Ponkala Maaret Juuli Anneli QUALITY OF SUBTITLING IN ONLINE STREAMING SUBSCRIPTION SERVICES

Transcription

1 UNIVERSITY OF EASTERN FINLAND PHILOSOPHICAL FACULTY SCHOOL OF HUMANITIES English Language and Translation Ponkala Maaret Juuli Anneli QUALITY OF SUBTITLING IN ONLINE STREAMING SUBSCRIPTION SERVICES A comparative study on subtitling quality between Netflix Original episodes and previously aired episodes of Gilmore Girls on Netflix MA Thesis June 2018

2 ITÄ-SUOMEN YLIOPISTO UNIVERSITY OF EASTERN FINLAND Tiedekunta Faculty Osasto School Philosophical Faculty School of Humanities Tekijät Author Ponkala, Maaret Juuli Anneli Työn nimi Title Quality of subtitling in online streaming subscription services: A comparative study on subtitling quality between Netflix Original episodes and previously aired episodes of Gilmore Girls on Netflix Pääaine Main subject Työn laji Level Päivämäärä Sivumäärä Date Number of pages Pro gradu -tutkielma X English language and translation Sivuainetutkielma June Kandidaatin tutkielma Aineopintojen tutkielma Tiivistelmä Abstract This MA Thesis discusses the subtitling quality on online streaming services. The quality is examined through translation errors discovered from the subtitling. The aim of this study is to evaluate if the subtitling quality is at an acceptable level by examining the quantity and severity of the translation errors. In addition, this thesis will discuss how the subtitling production processes have been studied to affect the subtitling quality, and what roles different actors, such as networks, subtitling production companies, translators and viewers, play in the process, and how they can possibly affect the quality. Streaming subscription services have increased their popularity in the past few years, strengthening their position on the market as providers of entertainment. As the popularity increases, multiple streaming subscription services have been developed. However, the most popular service remains Netflix. One of the ways Netflix has attempted to retain its status with the increasing number of streaming services, has been the introduction of Netflix Originals, i.e. movies and shows that are mainly produced by Netflix and viewable for its subscribers. In addition to the service s popularity, Netflix provides an example of a company that has listened to the critique it has received and aimed to improve its services accordingly. For that reason, this thesis will evaluate the quality of subtitling by examining both the subtitling of Netflix Originals and previously aired non-original content. The aim is to examine if the producer of the content is in relation to the quality of subtitling, and further examine the network s role in the subtitling production process, as it could be assumed that Netflix s vast investment on Original content would extend to the quality of subtitling in order to ensure the service s global success. For the scope of this study, Netflix provides an interesting example on the evaluation of subtitling quality by producing new seasons for some of the non-original content. An example of this is the American television show Gilmore Girls, which was originally produced from 2000 to 2007, and was continued as a four episode Netflix Original miniseries in Therefore, episodes from both the Original and non-original seasons have been selected as the material of this study. In this study, the subtitling quality is examined through translation errors discovered from the material. The discovered translation errors are examined in the three-part analysis of this thesis. The first part will examine the quantity and category of the errors, the second part will examine the quantity and severity of the errors through error analysis to examine the subtitling s overall quality, and the third part will examine, if the possible differences and statistical significance can be used to discover, if the added investment on Original content is extended to subtitling quality. From the analysis, it was discovered that none of the episodes would have passed O Brien s (2012) threshold test, and the quality of subtitling was not in relation to the producer. Therefore, it can be argued that the improvements Netflix has made in order to improve the subtitling quality do not extend to the problems linked with subtitling production processes. Avainsanat Keywords AV translation, subtitling, subtitling quality, translation errors, Netflix, streaming service, Netflix Originals, error analysis

3 ITÄ-SUOMEN YLIOPISTO UNIVERSITY OF EASTERN FINLAND Tiedekunta Faculty Filosofinen tiedekunta Osasto School Humanistinen osasto Tekijät Author Ponkala, Maaret Juuli Anneli Työn nimi Title Quality of subtitling in online streaming subscription services: A comparative study on subtitling quality between Netflix Original episodes and previously aired episodes of Gilmore Girls on Netflix Pääaine Main subject Työn laji Level Päivämäärä Date Sivumäärä Number of pages Englannin kieli ja kääntäminen Pro gradu -tutkielma Sivuainetutkielma Kandidaatin tutkielma Aineopintojen tutkielma X Kesäkuu Tiivistelmä Abstract Tämä pro gradu -tutkielma käsittelee suoratoistopalveluiden tekstitysten laatua. Laatua käsitellään tekstityksestä havaittujen käännösvirheiden kautta, ja tutkimuksen tavoitteena on selvittää, onko suoratoistopalveluiden laatu hyväksyttävällä tasolla käännösvirheiden määrää ja vakavuusastetta tarkastelemalla. Lisäksi tarkastellaan, miten tekstitysten tuotantoprosessien on tutkittu vaikuttavan laatuun, ja minkälainen rooli prosessin eri osatekijöillä, kuten suoratoistopalvelulla, käännösyhtiöllä, kääntäjällä ja katsojalla, mahdollisesti on tekstityksen laatuun. Suoratoistopalvelut ovat viime vuosien aikana kasvattaneet suosiotaan entisestään, saavuttaen yhä suuremman jalansijan viihteentarjoajina. Suosion kasvaessa palveluita on kehitetty useita, mutta suosituimpana suoratoistopalveluntarjoajana jatkaa Netflix. Palveluntarjoajien lisääntyessä yksi tavoista, joilla Netflix on pyrkinyt säilyttämään asemansa on ollut Netflix Originals -sisältö, toisin sanoen sarjat ja elokuvat jotka ovat yleensä Netflixin tuottamia ja joiden katselumahdollisuus rajoittuu lähinnä palvelun tilaajille. Suuren suosionsa lisäksi Netflix on esimerkki yhtiöstä, joka on kuunnellut saamaansa kritiikkiä ja pyrkinyt kehittämään palveluaan sen mukaisesti. Sen vuoksi tässä tutkimuksessa laatua tarkastellaan tutkimalla sekä Netflix Originals -sisällön tekstitystä että aiemmin muuta kautta tuotetun sisällön tekstitystä. Tarkoituksena on selvittää, onko sisällön tuottajalla merkitystä tekstitysten laatuun, jonka kautta voidaan pohtia palveluntarjoajan merkitystä tekstitysten tuotantoprosessissa, sillä voitaisiin olettaa, että Netflixin valtava investointi Originals-sisältöön uloittuisi tekstityslaatuun, siten taaten investoinnin kannattavuuden myös maailmanlaajuisen menestyksen kannalta. Tämän tutkimuksen kannalta Netflix tarjoaa mielenkiintoisen näkökulman tekstityslaadun tutkimisella tuottamalla uutta sisältöä muille palveluntarjoajille aiemmin tuotettuun sisältöön. Esimerkkinä tällaisesta sisällöstä on amerikkalainen televisiosarja Gilmoren tytöt, jota tuotettiin alun perin vuosina , ja joka sai jatkoa Netflixin tuottamana neliosaisena minisarjana vuonna Sen vuoksi tähän tutkimukseen on valittu jaksoja sekä sarjan aiemmin esitetyiltä kausilta että uudelta, Netflixin tuottamalta kaudelta. Tutkimuksessa tekstityksen laatua tarkkaillaan käännösvirheiden kautta. Löydettyjä käännösvirheitä tutkitaan kolmeosaisessa analyysissä, jonka ensimmäisessä osassa käännösvirheiden määrä ja kategoria selvitetään, toisessa osassa käännösvirheiden määrää tarkastellaan virheanalyysin kautta laadun selvittämiseksi, ja kolmannessa osassa tutkitaan, voidaanko käännösvirheiden mahdollisten erojen ja tilastollisen merkitsevyyden kautta päätellä, että Netflixin lisäpanostus Originals -sisältöön ulottuisi myös tekstitykseen. Analyysin perusteella selvisi, että valitun materiaalin yksikään jakso ei olisi läpäissyt O Brienin (2012) esittämiä raja-arvoja hyväksyttävälle tekstityslaadulle, eikä tekstityksen laatu ollut yhteydessä sisällön tuotantotapaan. Tämän vuoksi voidaan päätellä, että parannukset, joita Netflix on pyrkinyt tekemään käännöslaadun parantamiseksi eivät poista tekstitysprosessiin liittyviä ongelmia. Avainsanat Keywords av-kääntäminen, tekstitys, tekstityslaatu, käännösvirheet, Netflix, suoratoistopalvelu, Netflix Originals, virheanalyysi

4 1. Introduction Theoretical framework Audiovisual Translation Subtitling Finnish subtitling conventions compared to Netflix s subtitling conventions Production process of audiovisual translation and its relation to quality Viewers quality expectations and reception Translation errors Translation error categorization O Brien s quality evaluation model The quality of subtitling in AV translators and laymen s perspective Research material Netflix Gilmore Girls Research material in this study and its validity Research methods Translation error assessment Quality evaluation Quality difference evaluation through statistical significance Analysis Translation error analysis Overt errors Semantic errors Omission errors Cohesion errors Stylistic errors Covert errors... 66

5 Idiomatic errors Connotation errors Continuity errors Convention errors Translation error severity analysis Severity error categories Quality difference analysis through statistical significance Discussion Conclusion References... 91

6 1. Introduction Online streaming continues to increase their popularity, as more and more viewers watch their favorite shows and movies on the services, instead of depending on the supply of entertainment on traditional television channels. In the past few years, streaming service providers have strengthen their position on the market, providing viewers with a vast amount of readily available content, often originally in a foreign language to the viewer. From the service providers, Netflix remains to maintain its status as the world s leading streaming subscription service. Since launching its streaming subscription services in the United States in 2007, both the number of Netflix s subscribers and operating countries have been steadily growing, reaching the number of 100 million subscribers in 190 countries in 2017 (Statista1, Netflix). Although Netflix has a number of competitors, both Finnish and international, it has accomplished to attain the status of the most popular streaming subscription service among Finnish viewers with over 500,000 subscribers (Karkimo 2015, Heyman 2015). As an increasing number of viewers use and subscribe to online streaming services, the quality of subtitling in the services becomes an increasingly pertinent issue. Due to service s status and global spread, this MA Thesis aims to study the subtitling quality of online streaming subscription services by executing a case study on Netflix. Since the competition for streaming services grows with the popularity, one of Netflix s means to maintain its status and gain competitive edge is the launch of Netflix Originals, i.e. shows and movies that are produced for Netflix and mostly available on the subscription service. The investment on content has been estimated to be 6 billion US dollars for the year 2017 (Castillo 2017), rising up to 8 billion US dollars in the year 2018 (Koblin 2017). From the total content budget, Netflix has reported that 25 per cent has been spent towards Original programming, and the portion is expected to grow even larger in 2018 (Koblin 2017). Finland has a long tradition in subtitling foreign content for Finnish viewers. The start of Finnish subtitling began at Suomen Televisio (now known as Yleisradio, from here on referred to as Yle) in the late 50s and late 60s (Luova 2010). Nowadays subtitles are among the most read texts (Mäkisalo 2006), and for years broadcasting companies such as Yle and MTV had their own in-house translators providing translations of high quality, and the subtitling conventions of those networks, especially Yle, have been studied to further examine the construction of a high-quality subtitling (Vertanen 2007b, Lång 2013). However, the practice of broadcasting networks using in-house translators has decreased in recent years, 1

7 as many networks have outsourced their translation services to multinational translation companies (Palosaari 2016). With the changing practices, the decreasing quality of subtitling has not gone unnoticed (see, for example, Palosaari 2016). The differences in quality have often been examined through the structure of the subtitling production processes (Abdallah 2012), and contributed to the increased competition between translation companies resulting in lower compensation for audiovisual translators (from here on, audiovisual translators will be referred to simply as AV translators ) (Lång 2013). The production processes and their effect on subtitling quality will be discussed in subsection Poor quality subtitling can have a notable effect on the viewing experience of a television show or a movie. Noticeable translation errors in subtitling can cause the viewer s attention to be drawn to the subtitling itself, instead of focusing on following the plot (Vertanen 2007b). Furthermore, as mentioned in the blog maintained by an anonymous audiovisual translator (Pseudonym U 2012), idiomatic errors in the sentence structure and other anomalies in subtitling, which the viewer might not even notice, can have a negative effect on the viewing experience. This negative effect can leave the viewer with the impression that the fault is in the content instead of the subtitling (Heikkilä 2014, discussed in subsection 2.3.). Subtitling quality on Netflix has not always been a priority for the service. The Finnish subtitling on Netflix have been criticized on its quality, production practices and the compensation AV translators receive from the work since the Finnish version was launched in 2012 (Jansson 2012). After the critique Netflix received, the service has attempted to upgrade its subtitling quality by acquiring subtitling services from acceptable providers, and controlling the level of the translators professional abilities by implementing the HERMES test in 2017 (Netflix5). However, as the test is only aimed to ascertain the translators abilities, instead of aiming to enhance the entire subtitling production process, it is worthwhile to continuously study Netflix s subtitling to monitor how the quality and production evolves with the added developments. Furthermore, as this study began before the implementation of HERMES, it should provide some insight if the developments have been a necessary addition. In my BA Thesis (Ponkala 2016), I analyzed the quality of subtitling on Netflix using Buffy the Vampire Slayer as an example. In the study, the total number of translation errors was as high as 669 in three episodes. Most of the translation errors discovered from the material fell into the categories that would indicate visibility to the viewer (see, for example, Vehmas-Lehto 2005, Heikkilä 2014 on translation errors viewers are more prone to notice, 2

8 discussed in subsections and 2.3.). Furthermore, Vertanen (2007b) points out that the main function of subtitling is to work as an aid, so that the viewer is able to follow the plot of the show, and not draw unnecessary attention to the subtitling itself. For that reason, both the number and quality of the translation errors discovered from the material indicated that the quality of subtitling was not at the level one would expect from chargeable streaming subscription service. As the quality of subtitling in online streaming services was examined using Netflix as an example in my BA Thesis (Ponkala 2016), and the high number of translation errors was discovered, I wanted to study if the quality would improve when Netflix Originals, shows that were produced by and for Netflix, were concerned. Since Netflix Originals are a major investment for the streaming service, aiming to gain leverage among other streaming services and to ensure that subscribers stay with Netflix, it could be imagined that the investment would reach all aspects of the content, including subtitling. Furthermore, Netflix is an example of a service listening to the critique it receives from its customers and making active attempts to improve on quality, therefore examining if the service has reached an acceptable level of subtitling quality is an interesting subject to study. Netflix provides an interesting example for the purposes of this study by producing new seasons for popular shows that have been previously produced and aired in other networks, therefore the decision to study a show that had both Original and non-original content was an obvious choice. One of the shows featuring both Original and non-original episodes, Gilmore Girls, was selected as the research material of the study. The show had seven seasons that were originally aired between 2000 and 2007 (Internet Movie Database1, from here on IMDb1), and a newer season, produced as a Netflix Original in 2016 (IMDb2). From the show, three episodes were randomly selected from the non-original seasons and two from the Original season produced by Netflix for this study. The subtitling quality of the selected episodes is studied by using the same methods that were used in the BA thesis (Ponkala 2016) to find out the number of translation errors. A new aspect used in this study to further examine the subtitling quality will be provided by studying the number and severity of the translation errors by using O Brien s (2012) translation quality evaluation model to examine if the subtitling quality can be deemed acceptable. Once the quality has been examined, the translation quality of the Netflix Original episodes will be compared to the non- Original episodes to determine through statistical significance whether the subtitling quality on Netflix is higher on the Original episodes, and if the large investment on Original content follows through to an integral aspect for Finnish viewers, subtitling. Furthermore, as the 3

9 subtitled non-original episodes were added to Netflix s content library prior to the release of the Original episodes, the statistical significance test could indicate if the quality is evolving in the right direction. This MA Thesis will proceed as follows. Section 2 will present the theoretical background of this study, discussing the existing conventions of subtitling, how subtitling has been examined before, and what could be assumed from the previous studies and how the studies can be applied in this thesis. Section 3 will then present the research material of this thesis: Netflix and Gilmore Girls, and what needs to be understood from them to be able to discuss the quality and execute this study. Section 4 will present the research methods applied in this study and how the analysis will be conducted. The material will then be analyzed in section 5, divided into three parts of analysis. In the analysis, the first part will examine the number and type of translation errors found from the material, the second part will apply O Brien s (2012) translation quality evaluation model to determine if the translation quality can be deemed acceptable, and finally in the third part, the numerical data on the Original and non-original episodes discovered from the first two parts of the analysis will be compared to each other in order to investigate whether the added investment to Original content on Netflix extends to subtitling. After the analysis, section 6 will compile the results from the three-part analysis and discuss what can be assumed from it. Finally, section 7 will then conclude the research by discussing how the situation can be connected to the previous research on subtitling quality and subtitle production processes, and if the situation could be improved. 4

10 2. Theoretical framework This section will present the theoretical background of this study. The section is divided into three main subsections that will present and explain the main aspects that propel this thesis listed below. Audiovisual translation and especially subtitling, and the conventions that are associated with a good-quality subtitling in Finland. Translation errors and how they affect the viewing experience. Viewers tendencies to evaluate subtitling. Díaz Cintas (2010: 344) explains that audiovisual translation is an umbrella term that covers all areas of translating that combine an audiovisual element with a linguistic element that is in the need of translating. By combining auditory elements, such as music and other non-linguistic sounds, and visual elements with a linguistic element, an audiovisual translation is always in a transparent context, as nothing of the original elements are omitted when subtitles are added to the material, in contrast with the traditional translations (ibid.). In other words, the original content of the show is always visible and audible with the translated content, and therefore the translation can be constantly compared to the original content (ibid.). For that reason, subtitled shows and movies provide an incredibly lucrative source of study and therefore critique for not only academic research but also for the consumers of those shows, as the consumers are constantly able to compare the subtitling to the original utterance within their individual language skills (ibid.). Therefore, it is not surprising that the translation errors found in subtitling are discussed and collected on web sites, and in Finland, even compiled into books (such as Paakkinen 2003 & 2005, discussed in Vertanen 2007a: 322). Finland has a long tradition of subtitling foreign content for television (Luova 2010), and for that reason, the Finnish viewers undoubtedly have certain expectations regarding the quality and of subtitling. When a relative newcomer, such as Netflix, arrives on the market and the subtitling in the service possibly breaks some of the well-established conventions and the quality does not meet the expectations, it inevitably raises a question of whether or not the quality of subtitling is, indeed, acceptable. Therefore it should be examined, if the subtitling quality is the result of Netflix s own subtitling conventions or if the reason could be discovered from the subtitling production processes. Furthermore, as Netflix has made active attempts to improve its quality of subtitling, examining an aspect outside of 5

11 the service s attempts, production processes, could provide some further explanation on the remaining issues in subtitling quality. Even though a translator is usually considered the sole culprit of a translation error, in reality, they are only a cog in the wheel of subtitle production (Lång 2013: 52). In addition to the translator, production companies and networks have a significant role in the production, yet the role of the viewer may often stay overlooked, as it can arguably seem less obvious. However, as a subtitled program can be considered a product, where an AV translator is one of the providers, the final customer of the product is the viewer (James 2001: 152, discussed in Tuominen 2013). Therefore, the customers comments on the quality of a product should not stay overlooked, but be taken into consideration in order to improve the product and its production. Because the consumer of the show, or a layman from the point of view of Translation Studies, is able to constantly compare the original content with its translation (Díaz Cintas 2010: 344), it is certainly understandable that the translation is then mainly publicly evaluated by the number and quality of translation errors the consumer would be able to notice in the subtitling of a show. However, streaming subscription services, such as Netflix, are still businesses offering services in exchange for a monthly payment. Whether the money is made through selling commercials or direct payment from customers, the networks can still be simplified as businesses (channel, streaming subscription service) selling a product (movie, show, documentary) for their customers (viewer). An important aspect in evaluating a product s quality is obviously customer satisfaction. For that reason, in this thesis, the quality of subtitling will be examined through translation errors found in the subtitling Netflix provides for its content. Of course, different types of translation errors have a different type of effect on the viewing experience of a show. Some errors can have a more profound effect on following the plot of a show, whereas some errors may not affect the understanding per se, but rather annoy the viewer or draw unnecessary attention to the subtitling (Vehmas-Lehto 2005: 53). However, each type of translation error can have an effect, regardless of their significance to the viewer s ability to follow the plot of the show or the error s level of annoyance (ibid.). If the translation error continuously appears in the subtitling, it raises a question whether the translation quality can be considered acceptable. The terms used in this thesis to refer to various aspects of subtitles are subtitling, caption, and translation. In this thesis, the word subtitling is used as a large-scale term, when the subtitling of a movie or show is discussed in general. The term caption, in turn, is used 6

12 when the subtitling is discussed in detail, referring to a section of translated text that is visible on screen at one time. This is done to separate the issues, so that they can be discussed more comprehensively without the risk of accidental transposing. In the analysis part of this study, the subtitling will then be discussed by the relation of translation errors to individual translations, where an individual translation is considered to include only the translated caption of a single character s utterance visible on screen at one time. A more detailed explanation of how the translations are calculated in this thesis is provided in subsection 4.1. where it is introduced as one mean in calculating erroneous translations. Having now described the basis of this study, the following subsections will present a more detailed theoretical background for the claims made above. Subsection 2.1. will explain the theoretical background of audiovisual translation and subtitling and their special characteristics compared to the traditional translation methods. Subsection 2.2. will then further explain the background of subtitling in Finland and the established conventions and quality expectations behind them, as well as Netflix s own guidelines and how the subtitling quality of Netflix has been discussed in media since the subscription service first launched its services in Finland. The following subsection will then discuss the distinct roles AV translators, laymen, and broadcasting networks play in the production process of subtitled content. As subtitling and its conventions in Finland and on Netflix have then been established, subsection 2.3. will then explain how laymen and AV translators discuss translation quality and translation errors, using Maija Heikkilä s MA thesis (2014) as an example. Subsection 2.4. will then explain what constitutes as a translation error and the translation error categories used in this study to further analyze subtitling quality and the effect the errors can have on the Finnish consumer s viewing experience, using Gyde Hansen s (2010) and Inkeri Vehmas-Lehto s (2005) studies on translation errors and explaining how the categorization used in this study is comprised on the basis of the articles and how they are applied to subtitling Audiovisual Translation As mentioned above, audiovisual translation is an umbrella term that covers all areas of translating that combine an audiovisual element with a linguistic element that is in the need of translating, such as opera translating, dubbing, voice-over and subtitling. An audiovisual translation always appears in a multimodal context (Remael 2010), in other words in a context that has more than one mode in a single medium (Merriam-Webster). As the name 7

13 would suggest, the multimodal context of an audiovisual translation comes through two modes; auditory and visual. The auditory mode in an audiovisual translation includes not only the spoken words of the characters, but also any non-linguistic auditory elements, such as music and background noises (Díaz Cintas 2010: 344), as well as verbal fillers or smaller expressions, such as the umms, ahs and huhs. Similarly, the visual mode do not only include the written text on screen, such as signs or text messages, but also gestures and images (ibid.). The appearance of an audiovisual element with the linguistic element, or multimodality, is what separates audiovisual translation from traditional translations, as an audiovisual translation has to take into account the surrounding elements instead of exclusively translating a written text in a source language into a written text in a target language (Remael 2010). Therefore, as Remael (2010) mentions, in studying audiovisual translation, researching only the verbal elements of the content is not sufficient. The context of the linguistic element must be considered, whether it is inaudible, such as gestures or facial expressions, or audible, such as grunts or other non-linguistic sounds, as they can have an inevitable effect on the translation decision. In other words, the visual elements on screen designate, for example, what linguistic elements are represented clearly enough on screen that they can be omitted from the subtitling (Vertanen 2007b). Although still considered as a relatively new field of research in Translation Studies, the beginning of audiovisual translation can, in fact, be dated back over a hundred years. The earliest audiovisual translations can be considered to be the translations of intertitles in silent movies (Remael 2010: 12 17). Remael (ibid.) further explains that as silent movies were eventually superseded by talking movies, i.e. movies with auditory dialogue, the practice of audiovisual translation became a necessity for securing the export of films, especially in the United States. As the United States remains at the top three of the world s leading film industries by country, superseded by only China and India (Statista2), and considered as the country s second largest export (Webb 2009), it could be argued that the rise of audiovisual translation assisted the global success of the industry by securing the global export regardless of language differences. Remael (2010) further mentions that a number of different types of audiovisual translation methods were tried and adapted to secure the export of films, even multiple language versions of one film. With the development of the field of audiovisual translation and the needs of movie industries, two main methods of translating a foreign film and foreign content for export have carried through to this day: dubbing and subtitling (Remael 2010). 8

14 Although both methods pursue the same result, i.e. ensuring that a movie or a show originally in a foreign language to the viewer translates to the viewer s preferred language, the methods differ in execution. Dubbing refers to replacing the original dialogue with a translated recording in the target language that is synchronized with the actor s lip movements, therefore creating an illusion that the actor is actually speaking in the target language (Díaz Cintas & Orero 2010), whereas subtitling is executed by adding a written translation of the dialogue and other necessary linguistic elements on screen while preserving all of the original elements (Díaz Cintas 2010). Although dubbing is commonly used to translate content in a foreign language in many countries, such as Germany and Italy (Díaz Cintas & Orero 2010), in Finland, dubbing is mostly used translating content targeted for children, while the commonly established method of translating foreign language in an audiovisual context is subtitling (Vertanen 2007b: ). Díaz Cintas (2010: 344) rightly argues that the most popular audiovisual translation method used is subtitling, because compared to the other methods, such as dubbing, it is cheap and fast. A similar reason is presented by Vertanen (2007b: ), explaining that the cost-effectiveness of audiovisual translating is preserved in smaller countries and language areas by subtitling, whereas in more populous countries, such as Germany and Italy, the added costs of dubbing content are smaller in relation to the number or potential viewers Subtitling According to the definition by Díaz Cintas (2010: 344), subtitling consists of translating into target language both the utterances of the original dialogue and any other linguistic information that is either visually (written information in letters, text messages etc.) or aurally (songs, voices off) present in a show. In other words, this means that any information in a foreign language that is necessary for the viewers is translated, so that they are able to follow the plot of the show, in order to ensure that the viewer watching the subtitled version is able to have a viewing experience as similar as possible to someone viewing the show in its original language (Vertanen 2007b). This creates certain special characteristics for subtitling as a translation method, as there are restrictions that are an integral requirement to the formation of high-quality subtitling. As mentioned above and by Díaz Cintas (2010), subtitling is a form of audiovisual translation that consists of translating all source language information on screen (dialogue and text) that is necessary for the viewer to follow the show. Therefore, the function of subtitling is to work as an aid for the viewer instead of a separate linguistic unit that could 9

15 be read and understood on its own (Vertanen 2007b: 150), but Vertanen further explains that the information conveyed through subtitling is incomplete on its own, and understandable only when it is displayed on screen at the moment it is needed. The reason behind the allowed incompleteness could be divided into two; the aforementioned multimodal context and the limitations set for subtitling by time and space. The spatial and temporal limitations of subtitling mean that for the viewer to be able to read and understand the subtitling while simultaneously focusing on what is happening on screen, the subtitling needs to follow certain guidelines (Díaz Cintas 2010: 344). Among other things, these guidelines dictate caption lengths and how long those captions of a certain length should be visible on screen based on the speed of the original utterance as well as the assumed reading speed of the viewer (Díaz Cintas 2010: 345, Vertanen 2007b: 152). In other words, this means that not only does the caption need to be on screen the required time for the viewer to read it, the caption must be on screen the same time as the original is uttered and it must be of suitable length when compared to the original utterance or visibility of written linguistic element (Vertanen 2007b: 152). This, of course, causes the need to condense and reduce the information received from the linguistic elements on screen and soundtrack (Díaz Cintas 2010: 344, Vertanen 2007b: 152). Since everything that is said on screen or possibly shown in signs or other visual elements cannot always fit in the caption, the translator must find a way to single out the information that is necessary for the viewer to be able to follow the plot of the show (Vertanen 2007b: 152), or as similarly explained by Díaz Cintas (2010: 345), the subtitling must provide a semantically adequate account of the original dialogue. In other words, the condensed caption should be able to separate the plot pertinent issue from the original utterance and include it in the translation. Díaz Cintas (2010: 345) further explains that for the subtitling to provide that adequate account, each of the captions should be a coherent, logical and syntactical unit in its context. In other words, this means that each of the captions should form a unit that contains all the necessary information given in the dialogue same time on screen and be semantically and syntactically correct (Vertanen 2007b: 154). In this context means that in a single caption, the idea and essence of the original should be presented as a whole, instead of dividing the idea into two captions that are not visible on screen at the same time (ibid.). The reason behind this, as Díaz Cintas (2010: 345) explains, is that the viewer typically does not have the possibility to go back and retrieve the necessary information presented earlier in the subtitling, but rather the show is habitually watched from start to finish without stopping and going back. Therefore the subtitling should be understood 10

16 and placed in context by the viewer at once, so that the viewer is able to understand what is being said on screen and by which character (Vertanen 2007b: 151). In addition to the temporal and spatial limitations, there are certainly limitations regarding the linguistic and grammatical aspect of subtitling. Díaz Cintas (2010: 346) expresses that translating oral dialogue into written form creates a question whether certain aspects of speech, such as accents, can be appropriately rendered into writing. However, Immonen (2005:166) mentions that the subtitling should represent the style of the speaker, i.e. create an impression of how the person speaks. Vertanen (2007b: 153) similarly points out that a caption always represents a character, and if left too neutral, it could be argued that the subtitling loses its representation. However, as Vertanen (ibid.) reminds, a caption should convey the original message simultaneously with the original utterance and be understood instantly, therefore translating a character s line by standing too loyal to the amount of slang or colloquialisms to the original can cause problems with readability. In other words, if the subtitling uses too much slang or colloquialism to convey the character s personal style, the subtitling can become too complicated to quickly understand, and consequently cause the viewer to require more time to comprehend the unfamiliar, unexpected or ambiguous terminology of the subtitling. In result, the subtitling can thereby lose its purpose of working as an aid for the viewer. For that reason, Vertanen (2007b: 153) explains that when creating an illusion of a character s style in the subtitling, instead of using exclusively slang or colloquialisms, a more suitable way to express style would be to use language that merely suggests the presence of style, as to not disturb the readability. In other words, adding a small amount of colloquial language in words that are already easily understood to create an illusion of style should be sufficient. In Finnish, an example of this illusion would be replacing the word minä with more colloquial versions mä or mie to indicate that the character speech is not standard language. In addition to slang and colloquialisms, another way of creating style in both the original dialogue and in subtitling is the use of profanities. Vertanen (2007b: 153) reminds that profanities have a tendency to be more offensive when written down than when expressed orally. An addition to the challenges of translating profanities is brought by the use and origins of different swear words in different cultures, which can cause the translation of the profanity to be excessively emphasized in a confined translation, where time and space are limited and foreign language context is constantly present, such as subtitling (Vertanen ibid.). This certainly means that when a character s lines are full of profanities and swear words, it is not necessary to convey each one of the profanities, yet the presence of them should not be 11

17 completely omitted. However, as Vertanen (ibid.) further expresses, this does not mean that all profanities should be completely excluded from the subtitling, instead used when necessary to convey a character s style and when used, used in a suitable way. To exemplify this, Vertanen (ibid.) suggests that men fighting in the trenches of World War I would probably not use light profanities such as simpskatti, implying that the profanity used in the translation or subtitling should be chosen with the character s style and context in mind. Therefore, the level of offensiveness should be taken into consideration when translating profanities. Vertanen (2007b) further reminds that with subtitling, the viewer has a picture aiding the viewing, and is often able to conclude the vulgarity from the speaker s facial expression and tone without necessarily understanding the original language. Even though there is still no universal consensus regarding subtitling quality and what it should entail, rather the convention are closely related to context and culture, Vertanen (2007b:151) suggest that by following these general guidelines, the subtitling should be as invisible to the viewer as possible, i.e. the viewer is left with an illusion that they understand the original dialogue and language of the show, and preferably the viewer would not even notice that they are actually reading a translation on screen, thus, the subtitling fulfills its purpose of working as an aid for the viewer. In brief, one the most important things when it comes to translating audiovisual content with subtitles is that the viewing experience of a viewer that needs the subtitles is as similar as possible to that of someone watching the same show without the need of subtitling. Despite the restrictions presented above for subtitling, the grammar, linguistic conventions and idiomaticity of subtitling should not be forgotten, or as Immonen (2005: 166) reminds in her article, subtitling should still be grammatically and idiomatically adequate Finnish. As the main aspect of preserving this type of invisibility in subtitling is to not draw unnecessary attention to the subtitling itself, the grammar of the subtitling should certainly follow some guidelines, too. According to Immonen (ibid.), subtitling should follow the linguistic conventions of the target language, established patterns in speaking and writing, as well as grammar. In other words, in addition to being able to preserve the essence of the original dialogue and style while condensing them to suit the limitations set by time and space, the subtitling should still be idiomatic and grammatically correct in the target language. This will be more thoroughly discussed in subsection 2.2. discussing translation errors. In addition to these special characteristics regarding the actual practice of subtitling, the end result of the practice, i.e. a subtitled show, differs in nature from many other forms of translating (Remael 2010: 13). Possibly the most obvious main difference is 12

18 that the subtitling is an addition to the content instead of a replacement, as the original content remains in its original state in the translated version (Díaz Cintas 2010: 346). In other words, subtitling is only added to the content without omitting any of the original content. Díaz Cintas (ibid.) explains that the presence of the original sounds, dialogue and other linguistic elements with the subtitling, leaves subtitling in a particularly vulnerable situation, as it can be evaluated, commented and scrutinized by anyone with the slightest knowledge of the source language. Therefore, as other forms of translation are typically analyzed and criticized in academic research, subtitling can be, and certainly is, under constant scrutiny by the viewer, who can be considered as the final customer for the translation (Chiaro 2008), and a layman from the viewpoint of Translation Studies (as mentioned in Heikkilä 2014, discussed in subsection 2.3.). An example of this are the Finnish books by Jouni Paakkinen (2003 & 2005, mentioned in Vertanen 2007a: 322), where subtitling errors are collected. With the technology of today, the laymen are also able to share their viewing experiences and thoughts on translation quality with others online (Vertanen 2007a: 322, Heikkilä 2014), gaining a larger platform than ever before. Understandably, translation quality is mostly defined and discussed by laymen through the errors made in the subtitling as discussed in Heikkilä (2014). Heikkilä s findings on how laymen s thoughts on translation quality and especially translation errors found in subtitling compared to the thoughts of AV translators will be more thoroughly discussed in subsection Finnish subtitling conventions compared to Netflix s subtitling conventions As discussed in the previous subsection, the requirements of subtitling as a form of audiovisual translation set certain necessities in regards of timing, space and readability. Many of those requirements, such as the need to compress, can be regarded as globally universal in subtitling. However, as Finland has a long tradition of translating foreign content for television (Luova 2010), some of these conventions can be more accurately specified in the Finnish context. Additionally, due to the long traditions and studies showing that subtitling constitutes of a large portion of the texts Finns read daily (Mäkisalo 2006: 254), it could be argued that Finns definitely have certain expectations when it comes to subtitling. For that reason, the observance and reactions of laymen when it comes to the quality of subtitling come as no surprise when the expectations are not met, especially with relative newcomers such as Netflix. Therefore, it is important to study whether the critique is caused by Netflix s own, revised instructions, or if the problem lies deeper in the subtitling production process. To study this, it must first be examined what guidelines Netflix adheres to 13

19 or sets for the subtitling of its content and if they differ from the guidelines and conventions traditionally adhered to when subtitling content for Finnish viewers. If the guidelines differ on important aspects set to ensure the readability of subtitling and their general purpose, it could be assumed that the problem would lie in Netflix. However, if the answer is not found in Netflix s guidelines, further scrutiny should be focused on the overall process of audiovisual translation production to examine, if the problem can be discovered from there. This subsection discusses how the Finnish conventions and traditions are linked to the expectations Finnish viewers might have regarding subtitling quality, and the conventions commonly adhered to in Yle s subtitling and how and if Netflix s own subtitling guidelines differ from them. As mentioned, the practice of translating foreign content has been long used in Finland, starting as early as the end of 1950s (Luova 2010). Finland being officially a bilingual country, the need of subtitling is not only limited to translating foreign content into Finnish, but translating content to both Finnish and Swedish, especially in Yle s channels that are publicly funded. Nowadays, the need for audiovisual translation seems to be continuously growing, as subtitling is not only restricted to translating foreign content to another language, but also taking into account the viewer s other restrictions, with audio description (verbal narration for the visually impaired), and closed captioning and sign language interpretation for the hearing impaired (Yle 2015). The most commonly used practice for audiovisual translation in Finland still remains subtitling (Tuominen 2007: 295, Vertanen 2007b: ). Subtitles also constitute a large portion of both the translations as well as other texts Finns read daily, as discussed by Mäkisalo (2006: 254) and further studied by Salmi (2010: 7), and as the average Finn spends triple the time watching television as they do reading books or newspapers, Lång (2013: 52) further argues that subtitling may well be one of the most commonly read forms of Finnish. Therefore, as Jääskeläinen (2007: 117) argues, the quality of subtitling is not indifferent, and as Vertanen (2007b: 149) further explains that as Finnish children are internationally successful in studies testing reading comprehension, the quality of subtitling should be set to a high standard. Therefore, the conventions of Finnish subtitling as well as the production process are important to study, in order to ensure that instead of deteriorating the Finnish language, subtitling restores it and evolves with it. Due to the long traditions and prevalence of subtitling, it is a well-established practice in Finland. In his article, Vertanen (2007b) introduces Yle s conventions that are commonly adhered to in Finland. The universally applicable main guideline appears to be the 14

20 idea that captions are not meant to form a coherent text that can be understood on its own, but instead, the subtitling is supposed to operate as an aid, so that the viewer is able to follow the plot of the show without paying too much attention to the subtitling. Therefore it is important that the caption appears on screen at exactly the right time, and stays there long enough that the viewer has time to read it and is able to take in its context, i.e. can compare the caption to what is happening on the screen (Vertanen 2007b: ). In other words, the caption should appear on screen at the same time the original line is being uttered. Nevertheless, as explained above and further supported by Immonen (2005: 166), subtitling should still be adequate Finnish that follows the linguistic conventions, established patterns in speaking, writing and grammar. In addition, as Immonen (ibid.) mentions, the subtitling should represent the style of the speaker, i.e. create an impression of how the person speaks. Vertanen (2007b: 150) explains that subtitling should stay true to the original and aim to convey the style and tone as well as possible. In brief, as mentioned before, the most important thing when it comes to translating subtitles is that the viewing experience of a viewer using subtitles is as similar as possible to that of someone watching the same show without them. In addition to these general guidelines, Vertanen s report of Yle s conventions include more specified instructions on what should and should not be included in the subtitling. The more specified instructions collected from Vertanen s (2007b: ) article are listed in Table 1. Netflix already has a vast number of movies and shows in its service, and new content is being constantly added, and thus, the need for subtitling is constant. Therefore, it is understandable that as an enormous client for companies providing audiovisual translations, Netflix has its own set of guidelines to try and regulate the quality of subtitling on its content (Netflix2). The guidelines Netflix provides are divided into two: a general requirements guide (Netflix2) and a style guide specified for Finnish subtitling (Netflix3). The general requirements guide consists of 14 sections of general instructions for universal use and the guide specified for Finnish subtitling consists of 22 sections. Many of the guidelines follow the general rules of subtitling discussed in Vertanen (2007b) and Díaz Cintas (2010), but there are also some that are specific to Netflix. The Table 1 below will present the Netflix subtitling guidelines that are linked to Vertanen s (2007b) guidelines. 15

21 Table 1. Yle s subtitling conventions presented by Vertanen compared to Netflix s subtitling guidelines Yle s guidelines Screen time should be related to the length of a caption. A caption that is full two lines should stay on screen for 4 to 5 seconds. A caption that is the length of a full one line should stay on screen for 2 to 3 seconds. The minimum time a caption should stay on screen is 1 second while the maximum time is 30 seconds. Over 10 seconds is normally too long for a caption to stay on screen. The length of a caption should be in a correct relation to the original line. Short expressions should not be translated as long explanatory captions and longer lines should not be reduced to short expressions in the subtitling. Names are not necessary to be repeated in the subtitling as often as they are mentioned in original speech. American conversations they are constantly repeated in a way that is culturally appropriate in that context, yet in the Finnish the repetition is not common or idiomatic. However, a name should be included in the subtitling when it is first mentioned in the program. Every caption should be understood as a coherent unit, containing a single idea. If a sentence must be divided into two captions, certain guidelines should be followed. If multiple sentences uttered by the same speaker must be divided into multiple captions, each caption should include full sentence. If a sentence must be divided into multiple captions, those captions should not include sections of other sentences. If a sentence is divided into two or more captions, a hyphen is included at the end Netflix s guidelines Screen time is related to the length of a caption. The reading speed for adults is estimated at 17 characters per second The character limitation is 42 on one line. With two lines, the character limit rises to 84 characters in a caption o A full caption of two lines should be on screen for about 5 seconds. The minimum time for a caption to be on screen is 5/6 of a second, the maximum 7 seconds. There is no mention of caption length in relation to the original line. Character names should be left out once they have been clearly established. No mention of the caption being considered a coherent unit. If a sentence is divided into two or more captions, certain instructions should be followed. No mention how the sentence should be divided into two separate captions. However, there are instructions on how to divide a sentence into two lines of a same caption. It could be assumed that these instructions (never separate names, noun phrases, verb from subject, negation etc.) would be applied to separating a sentence 16

22 of the continuing caption to indicate that it will continue in the next caption. If a caption includes two separate speakers, the changing of speakers is indicated with a hyphen and a space before the translation. A caption should never include the end of a speaker s line and the beginning of another speaker s line if those lines are divided into two or more captions. A sentence always ends in a period. Italics should be used in moderation. Italics should be used when the speaker is not on screen. Italics should be used with foreign words. Positioning subtitling. When translating non-verbal elements (signs, text messages etc.) visible on screen, the caption should be centralized. If the element is short, capitalizing it can also be advised. Subtitling should not overlap with anything important happening on screen. between captions. If a sentence is divided into two or more captions, a hyphen is included at the end of the on-going sentence, unless another punctuation, such as a comma, is present. If a caption includes two separate speakers, the changing of speakers is indicated with a hyphen without a space before the translation. No mention on how on-going sentences should be divided between captions. Never omit punctuation. Italics should be used in moderation. Italics should be used when the speaker is not on screen. Italics should be used with foreign words. Positioning subtitling. When translating non-verbal elements (signs, text messages etc.) visible on screen, the caption should be centralized and capitalized, except for long passages (epilogues, dialogues). Captions should be positioned in a way to avoid overlap with onscreen text. As can been seen from the guidelines above, both sets of guidelines mostly focus on the layout of the subtitling, aiming to improve the readability and clarity, thus ensuring that the subtitling is easy and fast to understand. Thereby the subtitling in Netflix would appear to follow and adhere to similar conventions as Yle, attempting to improve the viewing experience and ensure that the subtitling fulfills its purpose of working as an aid for the viewer and not drawing unnecessary attention to the subtitling itself. When Yle s guidelines are compared to Netflix s guidelines it is apparent that they discuss similar aspects of the layout in very similar ways. However, there are some differences in minor technical aspects, such as whether or not there should be a space between a line and a hyphen, when the speaker changes in a single caption, yet overall the guidelines appear very similar to each other. This would suggest that at least in terms of general conventions, Netflix s subtitling would adhere to the Finnish conventions relatively well. 17

23 In addition to the general subtitling guidelines, Netflix also provides some further guidelines in the Finnish context. Many of those guidelines, such as on abbreviating titles, adhere to the Finnish grammatical rules, and in guideline number 22 it is even mentioned that any language-related issue not covered by the guide should be verified from the Institute for the Languages in Finland (Netflix3). Therefore, it could be concluded that Immonen s (2005: 166) call for grammatical correctness and linguistic conventions are required from subtitling in Netflix. How the guidelines set by Netflix are adhered to or enforced in the subtitling production process, however, remains unclear Production process of audiovisual translation and its relation to quality As discussed in the previous subsection, the subtitling conventions of Yle and Netflix do not greatly differ from each other. Comparable results on subtitling guideline differences were discovered by Lång (2013), as he studied the guidelines set by the most prominent subtitling suppliers operating in Finland, such as Yle, SDI Media, Pre-Text and MTV Media. Despite the similarities in the guidelines, the subtitling quality of those suppliers is considered very different: for example, Lång (2013: 51) presents Yle as the supplier that is known for the quality of the subtitling it broadcasts, whereas Netflix s subtitling has been under scrutiny since it launched its services in Finland (Jansson 2012, Pseudonym U 2012). However, Netflix has made continuous efforts to improve the quality of subtitling on its content, by implementing internal guidelines to ensure quality (Netflix2, Netflix3) and tests, such as HERMES, to ensure the translators language skills are at a professional level (Netflix5). Therefore, it is important to discuss other differences between subtitling suppliers to examine which of those differences could possibly affect the quality of subtitling. Abdallah (2012) has studied the production processes of AV translations from the translator s viewpoint. In Abdallah s (ibid.) study, it was concluded that the way a translator is assessed as a part of the process can be linked to the subtitling quality, especially when evaluated through subtitling errors. This subsection discusses the subtitling production processes and how they have been studied in relation to subtitling quality. As mentioned earlier and expressed by Lång (2013: 52), translators are still usually considered as the sole culprit of translations errors found in subtitling. However, producing subtitled programs involves many other actors in the process than solely a translator; the most prominent being considered as the commissioner (broadcasting network), the translation company, and the translator (Abdallah 2007). 18

24 In Finland, the subtitling industry is primarily divided into two sectors: public and private (Abdallah 2007: 272). In the public sector, represented nowadays mostly by Yle, the broadcasting network has their own in-house translators with the addition on freelancers, whereas in the private sector, a broadcasting network outsources their translation needs to a detached translation company. One of the often discussed differences in public and private sectors (see Abdallah 2012, Lång 2013), is the compensation translators receive from their work and expertise. For example, on the public sector Yle has agreed to adhere to the Yhtyneet agreement, which guarantees adequate compensation for AV translators and sets other guidelines to ensure the translators working conditions are reasonable (av-kääntäjät.fi 2012). However, on the private sector, it is argued that as broadcasting networks provide an increasing number of content that requires subtitling, the competition between translation companies for clients and the amount of production increases with it, which in turn is causing the decreasing of compensations freelance translators receive of their work (Lång 2013: 51, Abdallah 2012). Lång (ibid.) further argues that this causes the decline of subtitling quality, as decreasing compensation causes the professionals to abandon the field, and the remaining translators are left with the intensive pressure to try and handle the increased work load while being underpaid. Another difference between the public and private sectors is how the translators are employed. Generally speaking, the Finnish public sector, Yle, has its own inhouse translators (150 in-house translators in 2013) with the addition of freelance translators, whereas on the private sector, the translators work mostly as independent freelancers (Lång 2013). Lång (ibid.) explains that this means that on the public sector, the translators are able to interact with each other, yet on the private sector, the translators work mainly independently. Understandably, the in-house translators working under the Yhtyneet agreement can be considered more involved in the production process compared to the freelancers, as the in-house translators are a constant part of the production with less intermediate actors between them and the commission. Abdallah (2007: 273) further expresses that the translators working for Yle have a solid position in Yle s organization and are therefore able to influence the processes and deadlines more and negotiate an adequate compensation for their work. However, the freelance translators working in the private sector have more intermediate actors between them and the commission, which consequently causes there to be more opinions on the quality requirements, thus complicating the quality criteria standardization when quantity is valued over quality (Abdallah 2007: 273). The subtitle production processes were further studied in Abdallah s (2012) doctoral dissertation, where the conclusion was that the translators poor working conditions, i.e. long production chains 19

25 and poor compensation, negatively affected the subtitling quality, as the translators have less time and resources to properly apply their expertise. Subtitling quality differences between in-house translators and multinational translation companies have been studied by Palosaari (2016), using episodes of the British soap opera Emmerdale as an example. In Palosaari s study, episodes translated by the inhouse translators of the show s broadcaster in Finland, MTV, were compared to episodes translated by Broadcast Text International s freelance translators after the translation production was outsourced. In the study, Palosaari (2016: 79 81) came to the conclusion that the number of translation errors discovered from the episodes translated by in-house translators had fewer translation errors than in the episodes translated by the freelance translators. In reflection to the studies on subtitle production processes and their relation to subtitling quality (such as Abdallah 2007; 2012) the studies indicated that the production processes and the translators working conditions would have an effect on the subtitling quality (Palosaari 2016: 80). Because Netflix can be considered a broadcasting network, and relatively large one at that, it is not trivial how the production of subtitling is handled. When Netflix first launched its services in Finland, it received questionable publicity when it was discovered that some of the subtitling was illicitly acquired from DivX Finland, a fan subtitling site (Pseudonym U, 2012). After the negative publicity, Netflix moved to more appropriate subtitling providers, BTI Studios and SDI Media, with BTI Studios earning the title of being Netflix s preferred vendor for timed text and originals localization (BTI Studios). The interesting aspect of the preferred vendor title in the context of this study is that the title list timed text and originals localization separately, evidently referring to subtitling and subtitling Netflix Originals as different from subtitling non-originals. This would imply that Netflix would treat content that requires subtitling bought from other networks and content produced by Netflix differently. As one of the main focuses of this thesis is to study whether the subtitling quality of the non-originals is different than the Originals, this apparent separation made by Netflix is rather notable. If the assumed separation is true, it would then imply that the subtitling production process of the Originals would be treated differently from the subtitling production process of the non-originals on some aspects. However, whether BTI Studios is the only subtitling provider to Netflix cannot be claimed, as Netflix does not appear to disclose that information even when requested (Mäkelä 2016). In his article, Mäkelä (2016) explains that even though Netflix listed translators in Wired magazine as an important actor in gaining global success, the wage level 20

26 is still under question, as it is suggested that especially freelancers would receive payment as little as under three euros per content minute from their work. In the article on Wired (Greenberg 2016), it was reported that in order to ensure the subtitling quality on Netflix s first Original talk show Chelsea, the translators went through strict screening to ensure that the jokes would be adequately translated. Chelsea aired three times a week and was broadcast worldwide in the streaming service, in other words, new episodes would be added to Netflix s content library three times a week simultaneously in the countries Netflix is available, only one day after the episode was filmed. To ensure the subtitling was of good quality, 5,000 translators were tested to ensure that the jokes, idioms and cultural jargon would be welltranslated to their target audiences (ibid.). Greenberg s (2016) description of the process gives good indication on the subtitling process in this particular case: first, the filmed content went through a voice-recognition software and generated a rough transcript of the show, the transcript was then cleaned up by an editor, noting idioms and current events necessary for the translators, and the script was translated by a team of three to six translators in each target language. According to Greenberg (ibid.), the entire process was done in about twelve hours. Additionally, Greenberg (ibid.) suggests that in a normal situation, translating a feature film, for example, would take the team about a week. On this basis alone, it could be argued that Netflix is well aware of the importance of good translation quality and its effect on the viewing experience, and thereby the subscription services success. Netflix has further verified these suggestions with their introduction of HERMES, explaining that as the work of AV translators reach millions of people daily, the Netflix wishes to ensure that the quality of subtitling is at a level that preserves the creative intent and [is] mindful of cultural nuances (Netflix5). Therefore, it would be assumed that the translation quality would be well-monitored and regarded as worth the investment by Netflix. However, Netflix further explains that while they are able to influence the quality of subtitling by choosing their subtitling supplier, they cannot ensure that the supplier strives for the same standards on quality as Netflix (ibid.). Therefore, the subtitling production processes of the providing translation companies and how the processes should be improved remains an important issue to discuss. The following subsection will explain how subtitling quality can be examined through reception studies, and how the quality has been discussed to affect the viewing experience. 21

27 Viewers quality expectations and reception With so many actors involved in the process of producing subtitling, the quality of subtitling cannot be discussed without acknowledging the viewer s importance. As the viewer is the one that requires subtitling as an aid to fully enjoy and understand content in a foreign language (Vertanen 2007b), the quality of subtitling is not insignificant, and neither should be viewer satisfaction. Especially with commercial providers, such as Netflix and the American movie industry (Remael 2010), the usability of a product and customer satisfaction would be expected to play an important role. The viewer s role is emphasized by James (2001: 152, discussed in Tuominen 2013), by stating that [o]f all the clients, viewers and their expectations are the most important. This is further reinforced by Chiaro (2008: 250, discussed in Tuominen 2013), noting that the [e]nd users should not be enduring mere adequacy but have the right to expect top quality translations on TV just as they would expect tip top services in any other walk of life, thus comparing subtitling to any product that has been produced for a purpose. With the long tradition of subtitling foreign content in Finland, it could be expected that the viewers would have their expectations on the quality of subtitling, as discussed in subsection of this thesis. This is further supported by Mäkelä (2016), by reporting that people can be unwilling to watch foreign content with poor-quality subtitling, as even though the translation errors may not be noticed by the viewer, but if the message is conveyed inaccurately, the whole viewing experience can be negatively affected by poor translation decision and poor language. This subsection discusses the reception studies executed on audiovisual translation and how the results of those studies can be used to discuss viewers expectations on subtitling quality. The quality of any product can be determined on the basis of how well the product serves its purpose. Therefore the purpose of subtitling must first be defined, as was done in subsection of this thesis. From there, we can conclude that the main purpose of subtitling is to work as an aid for the viewer to be able to follow the plot of the show, and to ensure that the viewing experience of the subtitled content is as similar as possible to watching the show without the need of subtitling (Vertanen 2007b). Added conventions and practices are designed to ensure the readability of the subtitling, so that the subtitling would appear invisible to the viewer, thus creating an illusion that the viewer understands the original dialogue (ibid.: 151). To create that illusion, the subtitling should not draw unnecessary attention to itself; instead the focus of the viewer should stay on the original audiovisual content (ibid.), or as Tuominen (2013: 31 32) further states: subtitles are not the 22

28 primary objective of the viewing experience but a facilitator for understanding, and if they demand too much of the viewer s attention, they can have a detrimental effect on the entire experience. To ensure this natural flow (Tuominen ibid.) of subtitled content, the subtitling should adhere to the conventions of Finnish language (Immonen 2005: 166), so that the viewer does not need to try and decipher the unnecessarily difficult message while simultaneously trying to keep track of the original plot. The amount of support viewers require from the subtitling cannot be objectively studied, as the individual viewing experience is always dependent on the viewer s background (Tuominen 2013: 117). English language is widely studied in Finland, and thus, as a sizable portion of the programs subtitled for Finnish viewers is in English, both on Netflix and on the traditional broadcasting networks, i.e. channels, most of Finns can therefore understand at least some of the original linguistic elements (Tuominen 2013: 67), thus being able to possibly notice the erroneous subtitling. However, whether or not the translation errors are noticed from the subtitling by the viewer, should not be the only criterion on evaluating subtitling quality, as errors in any product affect the product s usability regardless of their noticeability. In Mäkisalo s (2012) study, it was discovered that a problem in a translation that is most often noticed by the viewers were problems concerning the captions timing issues. This could arguably suggest that the false information caused by mistranslations would remain unnoticed by the viewers, still possibly affecting the viewer s experience (see, for example, Vehmas-Lehto 2005). For that reason, it could be further argued that translation errors have an effect on the viewing experience regardless of their visibility to the viewers, and therefore all errors in subtitling should be taken into account when evaluating subtitling quality. However, the noticeability aspect cannot be overlooked, as not only does the error s visibility affect the visibility of the subtitling, the noticeable errors are also the ones viewers mention when discussing subtitling quality (Heikkilä 2014, discussed in subsection 2.3.). One way to examine the quality of subtitling from the viewers standpoint is via reception study. As can be understood from the term, a reception study is used to determine how the subject of the study, here subtitling, is received by its audience. Tuominen (2013) has studied the execution of reception studies for audiovisual translations in her doctoral dissertation, where many concepts as well as possible problems regarding them are discussed. However, the expectations viewers have on subtitling quality may not be uniform, as there is a number of variables present in the process of reception (Tuominen 2013: 54 55). In her dissertation, Tuominen mentions two strategies in executing a reception study: institutional and statistical. The institutional reception strategy refers to studies that are executed via the 23

29 gatekeepers, referring to non-ordinary viewers, who are in an influential position to the distribution of the material, and are used to determine how well the translation would be received (Tuominen 2013: 53). The statistical reception strategy, in turn, refer to acquiring numbers on the distribution of translated material and reading habits, to formulate statistics and abstract scenarios (ibid.). In other words, the statistical reception strategies are used to collect data that can be generalized for a larger public to create theories on the reception of audiovisual translations. Even though these strategies may not provide comprehensive information on the ordinary viewer s viewing experience on their own, Tuominen (2013: 55) suggests that by combining the two strategies, important information can be acquired to generalize the reception of subtitled content by creating a general impression on what aspects the ordinary user requires from the translation, and thus creating a general picture of the viewer s reception and then using other empirical means to determine how well this general picture is represented in the research material. Furthermore, as Translation Studies cannot be considered as an exact science, rather often segments of foreign language can be translated in various ways, yet all of them correct and of good quality (Toader 2007), personal opinions and backgrounds can affect both the translation process and the viewing experience. Therefore, to conduct a study to determinate the quality of subtitling without a focus group, as is done in this thesis, can be difficult, yet not impossible. Reception studies have been applied to examine how translation errors are noticed by viewers by, for example, Mäkisalo (2012). In his study, Mäkisalo (ibid.) discusses that translation errors often go unnoticed by viewers, if they match the picture or the plot. However, as mentioned earlier, a large portion of the content subtitled for Finnish viewers, both on Netflix and on the traditional broadcasting networks, is in English, most Finns can therefore understand at least some of the original linguistic elements (Tuominen 2013: 67), thus, the Finnish viewer is able to possibly notice the potential erroneous subtitling. As Netflix, like many others, only offers one version of subtitling for Gilmore Girls instead of multiple options based on individual viewer s language skills, the assumption is that it should be able to supply an acceptable quality of translation for all viewers regardless of their comprehension of the English language. Therefore whether or not a translation error is noticed by an individual does not mean that the error is insignificant, as another individual may very well be able to notice it. Furthermore, even if the error is not noticed in the translation, it does not necessarily mean that it has no effect, as it can still hinder the understanding or cause confusion and draw attention away from the show (Vehmas-Lehto 24

30 2005: 64 69). Thus, this MA Thesis will not only focus on the errors viewers would notice, but all the errors found in the subtitling. Tuominen (2013: 56) further presents two levels on conducting empirical research: micro and macro. On micro-level, the translation can be studied through specific instances, such as humor, to determine how the viewer reacts to those elements, and on macro-level, the reception situation is studied as a whole (ibid.). Tuominen mentions that while these levels investigate very different aspects of a translation, still combining elements of the two is common in research. How the levels can be combined is explained next. As mentioned, micro-level research studies the translation through specific instances to investigate how the audience reacts to them instead of studying the translation as a whole (Tuominen 2013: 57). These instances can vary from linguistic or stylistic choices to humor and to the subject of this thesis, translation errors. As an example, Tuominen mentions a study conducted by Chiaro (2007), where the difference in understanding translated humor were studied by comparing a British group viewing the original content in English and an Italian group viewing the same content dubbed in Italian, and how the groups respond to the humor in the content. In Chiaro s earlier study (2004), she had similarly studied clips of originally English content that contained humorous elements, and how the dubbing of those clips affected the humor of the clip. In her study, Chiaro (2007) explains that the dubbing affected the understanding of the humor in the material so that the British group s reaction to it was stronger than the reaction of the Italian group. However, Chiaro noted that in the newer study (Chiaro 2007) the difference was minor compared to the earlier study (2004), where 75 per cent of the humor went unnoticed in the dubbed version. Therefore, it was concluded by Chiaro (2007: 150) that the quality of translation in the earlier study was poorer, which is understandable considering the results indicating that most of the humorous elements went unnoticed by the focus group. Tuominen (2013: 58) notes that these results indicate that the quality of subtitling truly does have an effect on the viewing experience of the program. Tuominen (2013: 58 59) also reports a study by Antonini (2005), where the focus of the study was shifted to subtitled material and similar results were discovered, as the poor-quality subtitling used in the study resulted in the target audience of the subtitled program to understand the humor less than viewers watching the program without the need of subtitling, thus suggesting that poor-quality translation does have an effect on the viewing experience regardless of the audiovisual form. The macro-level research, in turn, offers a more general approach to researching translation, approaching the subject with interviewing the focus groups to investigate the 25

31 general attitude towards the translation (Tuominen 2013: 64). Tuominen (2013: 67) provides a study constructed by Bairstow (2011) as one of the examples of this level of research. In Bairstow s study, one of the focus groups consisted of French-speaking persons with little language skills in English and the other focus group consisted of English-speaking persons who were fluent in French. In the study (Bairstow 2011), the participants watched a film where the dialogue was in English, both with and without French subtitling, and after watching answered questions designed to chart how much information they were able to understand from the content. In the study, it was concluded that the French-speaking participants understood the content better with the subtitling, whereas the English-speaking participants understood the content better without the French subtitling. From these results, Bairstow (2011) was able to conclude that the subtitling was helpful when watching content in a foreign language, yet hindered the understanding when excessive, thereby corroborating with the idea that the viewing experience is affected by the viewers individual language skills. Tuominen (2013: 68) further points out, this type of macro-level research can be successfully used in order to gain a general overview of the attitudes of translations. Tuominen (2013: 57) notes that the micro- and macro-levels of research do not exclude each other, rather can be simultaneously used in research. This MA Thesis applies both practices, as on macro-level, the general conventions of subtitling are taken into consideration by examining the purpose of subtitling and how conventions are designed to improve the fulfilling of the purpose (discussed in subsection ), viewers attitudes on subtitling and what types of instances are most commonly discussed affecting the quality (discussed in subsection 2.3.) and how often those instances occur in the selected material (discussed in section 5). Thus, the micro-level aspect of the study will be identifying the translation errors, and the findings will then be compared to a macro-level research on how those translation errors affect the viewing experience. As the only focus group used in this study is me, the institutional aspect of the reception study is executed by me studying the original linguistic elements with the subtitling. After the linguistic elements are studied, the statistical aspect will then be executed by applying O Brien s (2012) quality evaluation model (discussed in subsection ) to the findings to conclude whether or not the quality of the translation can be deemed acceptable. The next subsection will explain how translation errors are defined and how they will be categorized in this thesis. 26

32 2.2. Translation errors Translation errors in subtitling still remains a relatively unexamined topic. Therefore the translation error categorization applied in this thesis will be constructed by investigating the subtitling conventions (discussed in subsection ) and general discussion on translation errors that can be applied to subtitling. This subsection will explain how this will be applied in this thesis. To be able to study subtitling through translation errors, the concept of translation error must first be defined. Hansen (2010: 385) explain that generally, translation errors are misunderstandings between the original text and the translation that happen when the translation process is interfered by undesirable anomalies between the languages, such as general differences between language systems or misunderstanding context due to cultural differences causing mistranslations. In other words, the translation process becomes disturbed and therefore the translation does not convey the original text accurately. According to Vehmas-Lehto (2005: 50), it is necessary to know the intended function of a translation or the original text to know which anomalies can be classified as translation errors. Vehmas-Lehto (2005: 49) further states that translators are expected to base their decisions to a certain standard expected from the form of the translation, and deviations from these standards can be classified as errors. As has been established in subsection , the main function of subtitling is to work as an aid for the viewer so that they are able to follow the plot of a show and to ensure that people s viewing experience is not dependent on the level of understanding of the original dialogue, rather is similar to someone watching the program without the need of subtitling (Vertanen 2007b: 150). Therefore, when attempting to evaluate the quality of subtitling, this should be considered as the principal standard or guideline all subtitling should adhere to. However, to be able to further evaluate how well a subtitling fulfills this principal guideline, it should first be understood how this quality is achieved, i.e. the subtitling conventions adhered to in the process. Vertanen (2007b: 153) further mentions that in order to fully achieve this, the subtitling should be as invisible as possible, so that the attention of the viewer is not needlessly drawn to the subtitling itself, rather the focus is preserved on the plot and ideally the additional help required from the subtitling is received without paying too much attention to the action itself, thus creating the illusion that the viewer understands the original language. Therefore, it is understandable that the most noticeable errors found in the subtitling happen when the subtitling is not in sync with the original audiovisual content of the program, as has been studied by Mäkisalo (2012). 27

33 However, as Chiaro (2008: 250) states, the viewer should not be expected to settle for mere adequacy, but the subtitling should be able to meet the viewers expectations and needs on quality on all aspects. Therefore, the standard guideline that Vehmas-Lehto (2005:49) states translators ought to abide to in order to achieve the accuracy and quality required from subtitling, are the standards that are discussed in this study (discussed in subsection 2.1.), and deviation from those standards shall be considered as translation errors. For that reason, and for the purposes of this study, the term translation error does not restrict to mistranslation found in the subtitling but is extended to include all errors found in the subtitling of the research material in this thesis. To achieve the accuracy and quality in the subtitling that would adhere to Vertanen s (2007b: 150) principal guidelines, the conventions play an important role. The conventions, such as Immonen s (2005: 166) call for grammatical and linguistic accuracy in the target language and preserving the essence in stylistic and semantic issues (discussed in subsections and ) are designed to ensure that the construct of the subtitling is comprised in a way that the subtitling is as invisible to the viewer as possible, while maintaining readability and accuracy. Vehmas-Lehto (2005: 53) further mentions that anything that prevents the original message from being understood in the translation, whether the message is a piece of information or something more abstract, such as feelings, humor or artistic values, can be considered as erroneous. Hansen (2010: 386) similarly states that all errors in the translation can have a considerable impact on the quality of the translation. However, these conventions cannot always be applied exactly as Translation Studies is not an exact science due to the differences in language systems and cultures. Therefore, every word and phrase cannot always be translated identically as the context must be considered (Toader 2007), yet the applying of these conventions in an appropriate manner is essential to maintaining the function of subtitling as explained in subsection Furthermore, the main guideline applied in this thesis thus remains the semantic accuracy, preserving the essence of the original and maintaining the invisibility in the subtitling. Despite that all deviations in a translation can be considered translation errors (Hansen 2010: 386), not all translation errors affect the quality of a translation equally. According to Vehmas-Lehto (2005: 53), the effect a translation error has on the quality of a translation can be used to determine the severity of the error. Vehmas-Lehto (ibid.) further explains that when translators are trained, the translation errors that alter the original meaning are traditionally perceived as the most severe. Obviously, this does not apply to only the traditional translations, but audiovisual translations and subtitling as well. In addition to the 28

34 translation errors affecting the understanding of the original meaning, Vehmas-Lehto (ibid.) further reminds that the understanding of the original may not be limited to the meaning, rather can be related to other aspects of the original content, such as emotion, humor and artistic values. Other errors, such as those related to the style and grammar of the translation, may be considered as less severe, however they can still hinder the understanding and confuse the reader (ibid.). The severity of the translation errors will be classified and examined in accordance with Vehmas-Lehto s (ibid.) discussion on severity and O Brien s (2012) model on quality evaluation models for translations. The quality evaluation model will be applied to assess if the number and severity of the translation errors can be used to examine the overall quality of the translation, and if the quality of the examined translation, i.e. subtitled episodes, can be considered acceptable. The evaluation quality model will be discussed in subsection The examination of the translation errors and their severity can be executed by dividing the translation errors into categories. As mentioned, the translation errors occurring in subtitling is still a relatively unexamined area and, therefore the categorization applied in this thesis is executed by adapting the translation error categories discussed by Hansen (2010) and Vehmas-Lehto (2005). The translation error categorizations discussed in the aforementioned articles are errors that are generalized to all translations; therefore the application of those error categories into subtitling is executed by comparing the errors found in the subtitling of Gilmore Girls and then examining which general translation error categories can be applied to subtitling. These categories will be explained and compiled in the following subsection Furthermore, as mentioned, Translation Studies cannot be considered as an exact science, rather often segments of foreign language can be translated in various ways, yet all of them correct and of good quality (Toader 2007), personal opinions and backgrounds can affect both the translation process and the viewing experience. Therefore, a line must be drawn between personal opinions and actual errors in translation. Thus, not everything that could be translated differently or better are to be considered as wrong. However, if a verbal element is translated in the subtitling as something that does not convey the original meaning, draws unnecessary attention to the subtitling or hinders the understanding or plot-following, whether the plot in that case is the plot of the entire show or a minor subplot that is mostly limited to a particular scene, then an error has occurred. In addition, even though the material is relatively vast, it is still limited, and therefore not all possible translation errors are expected to be found in the research material of this thesis. Hence, the categorization applied in this thesis might be insufficient for other research 29

35 materials. After the material has been divided into categories, the translation errors will be further analyzed based on how severe the errors are. To summarize, in this study, subtitling errors consist of all errors found in the subtitling. In other words, any deviations from either the general conventions of translation or Yle s subtitling conventions (Vertanen 2007b) that are not superseded by Netflix s own conventions (Netflix2 & Netflix3, see Table 1 subsection ) are considered translation errors. This will be further explained below in subsection As mentioned above, in addition to Vehmas-Lehto s (2005: 53) discussion on the different level of severity for translation errors and how they affect the overall quality of a translation, the severity level categorization will be based on O Brien s (2012) article on a quality model for translation errors in this thesis. The translation error assessment model discussed in O Brien s article will be presented in subsection Translation error categorization As mentioned above, to be able to examine translation errors and the effect they have on the quality of subtitling, and thus, on the viewing experience of subtitled content, the translation errors can be divided into categories. As there is no comprehensive translation error categorization specifically targeted for subtitling, the subtitling error categories used in this study are comprised on the basis of Hansen s (2010) and Vehmas-Lehto s (2005) articles on general translation errors. A similar method to comprise categorization was used in my BA Thesis (Ponkala 2016) and in that study, the categorization comprised of these two articles proved quite useful. However, as the research material used in this study is different than the material used in the BA Thesis, the categorization is not applicable as is, therefore some adjustments needed to be made, due to the simple fact that not all types of translation errors that were found in the material of the BA Thesis, Buffy the Vampire Slayer, were found from Gilmore Girls. In Hansen s (2010: 385) article, six categories for translation errors were suggested. The categories presented by Hansen are listed below. Pragmatic The translator has not taken into account the information reader has or needs. For example, not explaining terms that are foreign in the target culture or excessive explanation of terms that are not important to the plot of the show or are globally understood. 30

36 Semantic The meaning of the source text is not conveyed in the translation. In other words, the essence of the original is not accurately conveyed in the translation. Idiomatic The translation does not follow the typical characteristics of the target language. For example, word-to-word translation of idioms. Cultural The cultural aspects of the text are not familiar to the target audience, e.g. the frequent use of names in conversation. Syntactic The sentence structure is not characteristic of the target language, for example, sentence structure follows the source language s sentence structure instead of the sentence structure of the target language. Stylistic Misspellings and typographical errors. Hansen (2010: 386) mentions that omissions and additions can at times be considered as errors. In the material of both my BA Thesis (Ponkala 2016) and in this MA Thesis, unnecessary additions were rather rare, probably due to the comprising nature of subtitling. However, in the BA Thesis (Ponkala 2016: 22 23) some translation errors were found to suit this category, and the errors were mostly involved with adding too much explanation to the subtitling, therefore not conveying the original style of a character accurately. In this MA Thesis, the unnecessary addition category has been omitted as unnecessary, thus confirming that as the material of the study, especially its genre, affects the translation error categorization. Omissions, however, are included in the categorization and are further explained below in the listing of the categories used in this study. Another categorization for translation errors used in this study are found from Vehmas-Lehto s (2005: 49 74) article. The seven translation error categories mentioned by Vehmas-Lehto are listed below. Grammatical errors Such as errors in conjugation or case governance. The words do not have the correct relation with each other, for example in Finnish: koirat haukkui instead of koirat haukkuivat, in English: dogs is barking instead of dogs are barking. If 31

37 conjugational or case governance conventions are broken in order to convey style, it is not considered an error. Sentence structure The sentence structure of the translation does not follow the conventions of the target language or the emphasis of the sentence has been shifted. Idiomatic errors The translation does not follow the typical characteristics of the target language. Cohesion The text is hard to follow or the translation is lacking in context. Connotation The word choice in the translation has a different connotation than the original word choice. Style The style of the original is not accurately conveyed to the translation. Pragmatic The translator has not considered the information reader has or needs. As can be seen from the listings above, the categories both Hansen (2010) and Vehmas-Lehto (2005) present are somewhat similar to each other, yet both mention categories not found in the other. By combining the categories mentioned in these two articles, this study should be able to provide a more profound overview of the translation errors found in the Netflix subtitling of Gilmore Girls. Regardless of the inclusiveness of Hansen s (2010) and Vehmas-Lehto s (2005) translation error categorizations, especially when they are combined, the study in this MA Thesis needed two more categories to be added: continuity and convention. Due to the fact that the subtitling on Netflix is assumedly done by multiple freelance translators, the continuity of recurring elements, such as nicknames and other recurring jokes, suffers in some cases despite Netflix mentioning the continuity requirements in its general guidelines (Netflix2) stating that consistency across episodes and seasons should be ensured. As issues in continuity can be regarded as an error in other translations and not only an error related to Netflix s conventions, this has been separated from the convention errors. Other convention errors are considered to occur when the translation is in direct conflict with Netflix s guidelines regarding, for example, technical and stylistic aspects that would not be otherwise regarded as errors in the translation. In other words, the convention errors are errors when 32

38 either the layout of the subtitling or how certain abbreviations or titles should be translated is not in accordance with Netflix s guidelines. Furthermore, to be able to better analyze how these translation errors affect the viewing experience of the program, the translation error categories will be further divided into two umbrella categories discussed in Vehmas-Lehto (2005): overt and covert errors. According to Vehmas-Lehto (2005: 64 69), the overt errors are the errors that do not convey the meaning of the source text as it is intended, the errors that the reader would be able to notice and would disturb the reader. The covert errors, in turn, are the errors that convey the meaning correctly, but still have problems in linguistic aspects (Vehmas-Lehto 2005: 64 69). The covert errors cause the text to become unnatural and might make the understanding of the translation difficult. Vehmas-Lehto (2005: 64) mentions that if the translation is affected by covert errors, the reader might notice that there is something wrong in the translation, but might be unable to identify the specific error. The translation error categories applied will hereby be divided into the two umbrella categories, overt and covert errors, with subcategories subdivided under them. The overt errors will include the four translation error categories listed below. Omission The omission errors will only include the omissions that are either unnecessary or affect negatively on the viewing experience, e.g. when something that is important to the plot of the show is omitted. This is in accordance to the nature of audiovisual translation requires a certain amount of omission, so that the subtitling does not become too strenuous. Cohesion Cohesion errors are found when the subtitling is hard to follow, such as lacking in context, but yet again the nature of audiovisual translation does not always require context to be written down, as there are audiovisual aids conveying the context, only when the context is not either seen or heard on screen is there a cohesion error in the translation. Semantic errors A semantic error appears when the original meaning is not conveyed to the translation, this includes mistranslations and translation errors that seemingly could be translated correctly, yet the word choice in the translation changes the original meaning of the sentence. 33

39 Stylistic errors The stylistic errors include errors in spelling and typography. The four translation error categories that are considered to be covert errors are listed below. Continuity The continuity in, for example, jokes or nicknames suffers from the changing of translators between episodes, i.e. the continuous elements are translated differently between episodes thereby affecting continuity and plot-following. Idiomatic errors The translation does not follow the characteristics and conventions of the target language, for example, word choices or idioms are not translated in a way that a native speaker of the language would choose, or if the sentence structure of the translation does not follow the conventions of the target language or the emphasis of the sentence has been shifted. Convention error This category includes errors that are not translation errors per se, but errors in the subtitling process. These types of errors are errors in the stylistic conventions (not to be confused with stylistic errors), such as not adhering to Netflix s guidelines (Netflix2, Netflix3) regarding conventions in dividing a sentence into two captions not visible on the screen at the same time. If a convention error is found in an episode, there are two ways of calculating it in the error analysis; if a convention error is continuously and invariably found in an episode, it will be considered as one error in that episode, but if the application of that error varies throughout the episode, every new convention error will be considered as a separate error. This is done because convention errors are errors that would not necessarily be considered errors in the subtitling were they not especially advised against in Netflix s guidelines (Netflix2, Netflix3), and therefore it cannot be assumed that the conventions were made known to the translator by the commissioner(s). Connotation error Connotation errors are translation errors that occur when a character s mannerisms and style are either not included in the subtitling when they would be necessary for the plot or a character s style, or are incorrectly translated, for example, instances, when a character uses a mild expression or profanity and the level of the profanity exponentially increases in the subtitling. An example to clarify a connotation error 34

40 would be an instance, where an annoyed character would use an expression, such as Oh, shoot!, and the translation of the expression in the subtitling would be Voi vittu!, a commonly used translation, and a near equivalent, for Oh, fuck!. All of these translation error categories will be further explained and analyzed with more detail and examples from the research material in section 5. When the number of these errors has been discovered from the material, each of the error will be further categorized in terms of the severity of that particular error. The severity categorization of the translation errors will be based on Vehmas-Lehto s (2005: 53) statement that the severity of an error is related to the effect the error has on the meaning or essence of the original being accurately conveyed to the translation. The severity errors are then calculated separately and compared to O Brien s (2012) article on quality evaluation model for translations to determine whether the quality of the translation can be deemed acceptable. In other words, the translation error s effect on the translation quality will be based on Vehmas-Lehto s discussion on the effects of different translation errors, and the errors will then be divided into severity categories based on O Brien s article. O Brien s translation quality evaluation model will be explained next in subsection O Brien s quality evaluation model To examine the effect of a translation error on the overall quality of subtitling in the research material, the number and severity of the errors will need to be examined and calculated. For the assessment, the translation errors will be assigned a severity rating based on Vehmas- Lehto s (2005: 53) discussion, and O Brien s (2012) article, where she discusses the quality evaluation models for translations. In the article, O Brien introduces commonly used translation quality evaluation models that are suitable for evaluating the quality of translations using severity ratings to assess the translation errors occurring in translations. As this MA thesis only takes into account the errors found in the subtitling to examine the overall quality instead of examining the well-executed aspects of the translations and contrasting them with the errors, the quality evaluation models summarized by O Brien is the most suitable quality evaluation model for this study. Based on O Brien (2012), the severity of the translation errors found from the subtitling will be divided into three categories; minor, major, and critical (O Brien 2012: 62). O Brien explains that the severity of the errors can be divided into these categories based on commonly accepted criteria (ibid.). The listing below will present the categories and their 35

41 features according to the commonly accepted criteria explained by O Brien (2012: 62) and applied to the special features of subtitling. Minor The minor translation errors are the errors that may be noticeable for the viewer but do not affect the viewing experience negatively by either confusing or misleading the viewer. These errors, for example, can be small misspellings and other typographical errors that do not have a negative impact on the viewer s ability to follow the plot. Major The major translation errors are the errors that are likely to have a negative effect on the viewing experience or the ability to follow the plot. For the purposes of this study, the major translation errors will include any deviations that could affect the assumed viewer s ability to follow the plot. Critical The critical translation errors the errors that have a considerable effect on the product s usability, and therefore for the purposes of this study, the translation errors will be deemed critical if they severely impact the viewer s ability to follow and understand the plot, or critically mislead the viewer. As O Brien mentions, the penalties, i.e. how much the errors are expected to affect the understanding, applied with the severity ratings, vary, and therefore the context must be taken into consideration to evaluate if the translation can be deemed acceptable. Due to the nature of audiovisual translation, and especially in association with the medium and content selected for this study, the severity ratings will be heavily relying on the assumed viewer s ability to follow the plot. In her article, O Brien (2012: 58) mentions that all of the error-based quality evaluation models are used to define whether a translation is an acceptable quality by using a pass/fail threshold. O Brien (2012: 63) further explains that although there are several strategies in applying the severity ratings in different quality evaluation models, there is one commonly accepted agreement in how to apply the discovered data to this pass/fail threshold. According to the agreement, the number of errors will be made proportional by examining the number of each severity rating category in relation to 1,000 words. On average, for the quality of a translation to be of an acceptable quality, or a pass on the threshold, there should be fewer than four minor errors and one major error per thousand words, and even one critical 36

42 error in a translation would result in an automatic fail. This agreement will be used as a basis for the quality evaluation in this study The quality of subtitling in AV translators and laymen s perspective As mentioned earlier and expressed by Lång (2013: 52), translators are still usually considered as the sole culprit of translations errors found in subtitling. However, producing subtitled programs involves many other actors in the process than solely a translator; the most prominent being considered as the commissioner (broadcasting network), translation company, and translator (Abdallah 2007). As the final customer, especially with profitable businesses such as streaming subscription services, can be considered to be the viewer, their views on translation quality must be taken into consideration (James 2001: 152). As studied by Bairstow (2011, discussed in subsection ), the subtitling had plausible effects on the understanding of the content, and by comparing Chiaro s (2004 & 2007, discussed in subsection ) studies, it could be concluded that the quality of an audiovisual translation does affect the viewing experience. Therefore, even though some of the translation errors in subtitling often go unnoticed if the subtitling is otherwise properly synchronized (Mäkisalo 2012), it does not mean that translation errors, whether they are noticed or not, have no effect on the viewing experience. Vehmas-Lehto (2005: 53) further mentions that errors in a translation occur most severely when the translation does not convey the original message, whether the message is a piece of information, emotion, humor or other artistic values, altering any element of the original text that is important for the reader (or viewer). However, other less severe errors, such as errors in grammar or style, can still have an effect on the understanding or annoy or confuse the reader (or viewer) (ibid.). To further examine how these translation errors in subtitling can possibly affect the viewing experience, understanding how the errors are conceived by the viewers may be beneficial. In addition to the viewers perception, equally beneficial might be to examine how the quality and translation errors are perceived by the AV translators, to examine how their attitudes on the quality of subtitling differ from the laymen s perceptions and how the AV translators themselves discuss the possible reasons behind the errors. This subsection will explain how translation errors in subtitling are discussed by both laymen and AV translators to understand what type of translation errors are noticed by the viewer and how they possibly affect the viewing experience on the basis of the discussion on Maija Heikkilä s (2014) MA Thesis. 37

43 In Heikkilä s MA Thesis Tekstitysten laatu av-kääntäjien keskusteluissa ja blogeissa (The quality of subtitling in AV translators Internet forums and blogs) (2014), Heikkilä has examined how the quality of subtitling is discussed by both AV translators and viewers on the Internet. For the MA Thesis, Heikkilä studied the comments AV translators and laymen had made on Internet forums and blogs maintained by AV translators. The study included three forums: the Finnish av-kääntäjät.fi, the American Proz.com and the German XING Übersetzer Lounge, and two blogs: the Finnish av-kääntäjät.fi and a blog maintained by a German AV translator. The forums and blogs examined by Heikkilä were all specialized in AV translating. The blog av-kääntäjät.fi is founded as a respond to the defects in the AV translation industry, improve the AV translators status and inform others on the profession (Heikkilä 2014: 29), Proz.com is the world s leading online community for translators, defining their mission as to provide tools and opportunities that translators, translation companies, and others in the language industry can use to: network, expand their businesses, improve their work, and experience added enjoyment in their professional endeavors (proz.com), and XING Übersetzer Lounge is a networking and community service for German translators (Heikkilä 2014: 30). Of course, due to the field of these forums and their standpoint, the conversations can be considered as rather narrow in some aspects (Heikkilä 2014: 30). However it could be argued that on some aspects, due to the conversationalists expertise, they offer broader views on the subject, since as Vertanen (2007a: 322) states that the conversation on AV translators usually concentrates on the funny translation errors, typically removed from context. To gain a better understanding on the significance translation quality has on the overall viewing experience, the focus of the conversation should shift from ridiculing the translation errors to the overall quality and how the translation quality is affected by external issues, such as production processes. The effect translation processes have on the translation quality has been rather comprehensively studied (see, for example, Abdallah 2012), but it is arguably equally important to examine how the translation quality affected by the process further affects the viewing experience, to gain a better understanding on why improving the translation process, and thus improving the translators status, would affect the improvement on the viewing experience and thereby customer satisfaction. Through discourse analysis, Heikkilä (2014) studied what the forums and blogs revealed about the attitudes towards the AV translations and how the laymen s and AV translators attitudes towards the AV translations differed from each other. According to the study (Heikkilä 2014: 37) the AV translator s comments on the forums and blogs often discuss the reasons behind the translation errors. One frequently mentioned reason was the 38

44 commissioners (such as broadcasting networks), especially often expressing frustration towards commissioners attitudes, implying that the most common reason for bad quality is the commissioners perception that a competent AV translator offers fast rather than highquality translations. Especially the AV translators themselves therefore call for more feedback from viewers, both positive and negative, so that the commissioners would begin to appreciate quality over quantity (Heikkilä 2014: 54). In contrast with the AV translator s comments and discussions, according to Heikkilä (2014: 34), the laymen s comments focus mainly on more visible aspects in subtitling, such as wording, grammatical correctness and timing. Heikkilä mentions that it is understandable, considering the people referred as laymen are the ones that have no personal experience in audiovisual translation other than as a viewer. In addition, Heikkilä mentions that it is frequently mentioned in the posts that AV translators often have limited time to produce the translations. According to the posts, this is due to the fact that the compensation AV translators receive from multinational translation companies is still so low that AV translators are not able to invest enough time in providing high-quality translations. The bloggers and conversationalists also believe that changing the AV translator in the middle of a television series affects the quality and especially the continuity of the subtitling negatively, unless the translators collaborate with each other (Heikkilä 2014:55). In the MA Thesis, Heikkilä comes to the conclusion that the laymen mostly notice the poor quality in subtitling through individual translation errors, which then have a negative effect on their perceptions regarding the translator s competence. In contrast, AV translators regard poor subtitling as one that is not synchronized with the picture, that the viewer finds difficult to follow, includes foreign sentence structures, i.e. is not idiomatic, or is otherwise foreign and unnatural in its structure (Heikkilä 2014: 63 64). According to the comments, high-quality subtitling is something that is as fluent and unnoticeable as possible, as also mentioned in Vertanen (2007b, discussed earlier in subsection 2.1.). As mentioned before, the fluency and noticeability criterion will be used as the main guideline in investigating the translations in this MA Thesis and to discuss how the translation quality is connected to the views of both laymen and AV translators. As has been explained earlier (see subsection 2.1.), the viewers expectations and quality requirements cannot be ignored. However, in addition to the laymen s opinions, it is important to hear the AV translators own opinions considering subtitling quality and especially the circumstances affecting it, since the AV translators still have not received the credit they are due in 39

45 producing foreign content for Finnish viewers, yet the laymen still consider the translator a sole culprit for the translation errors. As mentioned by Remael (2010:15), New technical and socioeconomic developments are enhancing that process [of audiovisual translation], sometimes to the detriment of translators status and working conditions, and a focus on quantity rather than quality. It can be argued that once the AV translators receive the credit they are due, the quality of subtitling can only improve. In other words, this means that even though Netflix has worked towards ensuring the subtitling quality on its content by setting guidelines and testing translators abilities, it should be ensured that the proved and competent AV translators receive appropriate compensation and working conditions to be able to properly execute their professional skills and provide translations of good quality. 40

46 3. Research material This section will present the two main subjects of this study: Netflix and Gilmore Girls. Subsection 3.1. will offer a brief introduction on Netflix and its practices and content, followed by subsection 3.2., which will introduce Gilmore Girls and information about the show, such as special features that needs to be taken into consideration when evaluating the subtitling of the series, and subsection 3.3. will then present the research material used in this study Netflix Netflix was founded in the United States of America in 1997 by Reed Hastings and Marc Randolph (Netflix). It started out as a website where customers could rent movies online, and the physical copy of the movie would be sent via mail to them. In 2007, Netflix began offering an online streaming service (Netflix), which meant that the subscribers no longer had to wait for the movie or other show to be delivered to them, but instead could instantly stream it online. Netflix is the leading online streaming subscription service in the world, and the numbers of both operating countries and subscribers have been steadily growing. In 2005, the streaming service had over 57 million subscribers around the world, and the number has risen to over 81 million by 2016 (Heyman 2015, Netflix letter to shareholders 2016) and the numbers finally reached 100 million subscriber in 2017 (Statista1). Netflix operates in over 190 countries worldwide, and at the time this thesis was written, the service investigating possibilities to operate in China, and the other locations Netflix was not available were the Crimean Peninsula, North Korea and Syria due to corporal limitations on the countries set by the United States (Netflix). To retain its status as the world s leading online streaming subscription service and to gain leverage among other streaming services and broadcasting networks, Netflix announced in 2011 that it was bringing the service s first self-licensed and produced series to the service with House of Cards (Sarandos 2011). The series premiered February 2013, and was initially considered a gamble with Netflix investing 100 million US dollars to the production. However, House of Cards turned out a success, earning numerous Emmy nominations and increasing Netflix s visibility (CNN 2014). Since the success of House of Cards, Netflix has continued to produce Netflix Originals, many of them receiving both popularity and critical acclaim. The investment on content has been estimated to be 6 billion 41

47 US dollars for the year 2017 (Castillo 2017), rising up to 8 billion US dollars in the year 2018 (Koblin 2017). From the total content budget, Netflix has reported that 25 per cent has been spent towards producing Original content, with expectations to the to grow even larger in 2018 (Koblin 2017). Netflix s online streaming service has been available in Finland since 2012 and has since become the most popular international streaming service in the country, reaching the number of 500,000 Finnish subscribers in 2015 (Karkimo 2015). The content library includes anything from movies and television series to documentaries and stand up specials, however, the selection varies by country (Netflix). As has been established earlier, a large portion of the Finnish viewer s viewing experience includes subtitling. When Netflix s global success is taken into consideration, it becomes apparent that many of the countries Netflix operates in are not English-speaking countries, and to ensure the global success, the subscribers overall viewing experience, including subtitling, should be taken into consideration. The importance of the quality of subtitling has not gone unnoticed from Netflix, as was discussed in subsection about the services first Original talk show, Chelsea. As Netflix has continuously increased the portion of budget spent towards Original content (Koblin 2017), it would be assumed that some of the investment would be used towards subtitling. However, subtitling quality has not always been a priority for the subscription service as was discussed in subsection After the questionable publicity Netflix received in its early days in Finland, the service has then acquired the subtitling in more admissible ways, from the two largest translation agencies offering audiovisual translations, BTI Studios and SDI Media, nowadays awarding BTI Studios the title of Netflix s preferred vendor (BTI Studios). In addition to changing the subtitling providers, Netflix has made additional proceeding in order to ensure the quality of subtitling would serve the service s goals and meet the subscribers expectations in the best way possible. These proceedings include the guidelines examined in subsection and HERMES, a test designed to examine the translators proficiency in the English language, translating idioms to their target language, identify linguistic and technical errors, and subtitling (Netflix5). An additional aspect of HERMES has been in issuing the translators an H-number to imply their subtitling proficiency, so that the data can be compiled, and the subtitling commissions can be offered to the translators that would be best suitable for them (ibid.). Netflix has since closed the possibility to take the test, announcing it has reached its capacity on all required languages due to the test s popularity among translators (Netflix6). It has not been announced if the test will reopen if the need for more translators should return. As the test was implemented after 42

48 the research material of this study was acquired, it will not be taken into account when discussing the AV translators levels of expertise. However, the attempts Netflix has made in order to improve the subtitling quality in its content will be taken into account when discussing the subtitling quality and the possible issues affecting it Gilmore Girls Gilmore Girls was originally aired between 2000 and 2007 by the WB and the CV (IMDb1). The series follows the close relationship between a mother, Lorelai Gilmore, and her daughter, Rory. Lorelai is portrayed as a former teen mom, who had her daughter when she was only 16 years old, left her childhood home to raise Rory as a single mom and through hard work managed to rise from a maid to a manager at an inn, to eventually opening her own inn. Her daughter, Rory, is an intelligent and ambitious young woman, who loves reading and aspires to become a journalist. At the beginning of the series, Rory herself is 16 years old, and the previously aired episodes of the series followed their lives from Lorelai s decision to transfer Rory from public school to a private high school, in order to help Rory best achieve her full academic potential, until Rory graduates from university. The mother-daughter-duo are arguably most well known for their intelligence and wittiness, and for the dialogue being a substantial part of the series and often delivered in a fast pace, as further enforced by Honeywell (2011), stating that when most of the television series of the same length would have scripts of pages, a script in Gilmore Girls were known to hit 80 pages per episode. The fast dialogue would often be filled with inter-textual and pop-cultural references, often obscure to even the native viewer (Bednarek 2010: 33). One of the most notable aspects of Gilmore Girls, and arguably one of the reasons the series has stayed in people s minds, is the characters are defined by their characteristics instead of gender. Even though many of the most prominent characters in the series are women, the female gender is not presented as a homogenous unit, rather each of the characters are individuals with their own specific features. Those specific features are best appreciated through the dialogue; from Lorelai s and Rory s witty back-and-forth to the clever insults of Rory s friend, Paris, and the backhanded compliments of Lorelai s mother, Emily. The series was praised for its portrayal of the female characters, breaking the conventional image of girls and women, who yearn for higher education and knowledge, falling to the clichéd geek image of being intelligent yet either socially inept or masculine to compensate for the intelligence (Westman 2007). Instead, as Westman (2007: 23) incisively 43

49 exclaims it, in the world of the Gilmore Girls, women are brilliant, education is cool, and intelligence is a team sport, thereby expanding the portrayal of female characters, especially the intelligent ones. Due to its refreshing views on female intelligence, the series and its characters have been the subjects of many a publication (see Westman 2007, Bednarek 2010). Thus, it is hardly surprising that the series was not forgotten by viewers either, and in January 2016 it was finally announced that after nearly a decade since the original series ended, Netflix would bring it back as a four episode miniseries (Ausiello 2016). The new season Gilmore Girls: A Year in the Life follows the main characters, Lorelai and Rory Gilmore, for a year, separated into four seasonal episodes (IMDb2). The Netflix Original miniseries followed faithfully in the footsteps of the original, preserving the wittiness as well as the amount and speed of dialogue that have become such a distinctive feature for the series. As mentioned in subsection , the main focus of subtitling is that the viewing experience of a subtitled program is as similar to the original program as possible. Therefore, the subtitling of Gilmore Girls and Gilmore Girls: A Year in the Life should reflect the characteristics found in the original dialogue. However, as the subtitling should still preserve its readability, preserving the amount of text from the original to the translation would prevent the purpose of subtitling to be fulfilled. Therefore the audiovisual aspect of translating discussed by Immonen (2005) is applied: as the viewer is able to hear the characters talking a lot and fast on screen, the subtitling should be able to compress what is being said, preserve the essence of it, and convey the essence and style correctly to the viewer. As mentioned, this idea is the guiding principle applied in the analysis of translation errors in this MA Thesis Research material in this study and its validity From the series, three episodes were randomly selected from the previously aired seasons and two episodes from the new seasons. The three episodes from the previously aired season are: episode 16, Star-Crossed Lovers and Other Strangers from season 1, episode 7, They Shoot Gilmores, Don t They from season 3 and episode 18, Hay Bale Maze, from season 7. The two episodes from the Netflix Original season are: episode 1, Winter, and episode 2, Spring. These episodes will be later identified by their season and episode numbers, following the custom of SxEy, where the variable x will identify the season number and the variable y will identify the episode number, for example, the 18th episode from season 3, Hay Bale Maze, will now be 44

50 identified as S3E18. For separation from the first season of the previously aired episodes, instead of addressing the Netflix Original episodes as S1E1 and S1E2, the episodes will be addressed as S8E1 (Winter) and S8E2 (Spring). The material was mostly collected by transcribing straight from Netflix. The three episodes from the previously aired seasons had fan-made transcriptions found on the internet (crazy-internet-people.com), and those merely needed to be looked over with the audio. However, the two episodes from the new season did not have any transcriptions readily available, so those had to be transcribed from the audio. The subtitling was transcribed from the screen on all episodes, which even though may seem as overly laborious, was probably the best solution for myself, as I was able to continuously compare the original lines to the subtitling with the audiovisual aspects, thus ensuring that the audiovisual aspect would, indeed, be taken into consideration when looking for translation errors. I believe that due to the context having been constantly present when evaluating the subtitling, this also adds to the validity of the analysis of this study. In addition to the consideration of the audiovisual elements with the verbal elements, I would argue that another aspect adding to the validity of this thesis is the amount of research material and how it is selected. The episodes examined in this study have been randomly selected, and instead of the material consisting of only clips where a translation error is present, this thesis will examine the entire selected episodes to ensure that the analysis is as accurate as possible, and the amount translations have been taken into account when discussing the translation errors. Additionally, this means that the material examined in this thesis is relatively extensive, as the full transcribed material consists of over 80,000 words (both original lines and subtitling) and circa 300 minutes. 45

51 4. Research methods As the theoretical background and the research material of this study have now been established, this section will explain how the study is executed. The main subject of this study is to examine the subtitling quality on Netflix through the number and severity of translation errors, and if the added investment on Netflix Original content extends to the Finnish subscribers viewing experience through subtitling. The analysis in this thesis will therefore be divided into three parts, each part examining a different aspect of the subtitling, yet still all connected with each other to gain perspective on the overall quality of subtitling. The first part of this analysis will identify and examine the translation errors found in the subtitling of the research material, by examining the subtitling and dividing the discovered errors into categories discussed in subsection of this thesis. The error categorization will be used to determine how many translation errors there are on the material, and what is the possible effect those errors have on the viewing experience. The second part of this study will examine if the number and severity of the errors found in the first part of this study. The number and severity of the translation errors will be used to determine if the subtitling quality of Gilmore Girls on Netflix can be deemed acceptable based on O Brien s quality evaluation model discussed in subsection of this thesis. The third part of this study will examine the differences in the subtitling quality of Netflix Original and non-original content. The difference will be examined through the number of translation errors found in the subtitling by applying the numerical data discovered in the first two parts of the analysis. The data will be examined through statistical significance model to investigate if the quality of subtitling can be considered different in Original and non-original content. The following subsections will explain how the analysis is executed in the three parts of this study and how they are connected to the existing theories on subtitling and its quality discussed in section 2 of this thesis. Subsection 4.1. will explain how analyzing of translation error categories is executed, subsection 4.2. will explain the severity level of the translation errors, and finally subsection 4.3. will explain how the statistical significance method is applied to the discovered data. 46

52 4.1. Translation error assessment As discussed above, the first part of the analysis will examine the translation errors found in the subtitling of Gilmore Girls. The analysis of the study will be executed by examining the subtitling of the selected episodes of the series to discover what types of errors there are and the number of each of translation errors in each translation error category. The transcriptions of the spoken lines for S01E16, S03E07 and S07E18 were found on the Internet (crazy-internet-people.com) and were verified by comparing them to the audio on Netflix. The amount of work that came with the transcribing was definitely lightened by the transcriptions of the non-original episodes being found on the Internet, but as the Netflix Original episodes were so new at the time this study began, the transcriptions were not readily available, so the spoken lines from S08E01 and S08E02 needed to be transcribed from the audio. After the transcriptions of the spoken lines were finished, the subtitling of the episodes were transcribed in the same document as the original, so that the original elements could be easily compared with their translations. As the subject of this study is in audiovisual form, the subtitling could obviously not be compared with just the original spoken lines, therefore to examine which captions contained a translation error, each caption was examined in context. As the main guideline applied in this study is Vertanen s (2007b: 150) idea, that the subtitling should work as an aid to ensure that the viewing experience in subtitled content is as similar as possible to the original by conveying the elements in a foreign language accurately, each caption was compared to the guideline. The translation errors discovered from the research material were then categorized according to their specific features. The translation error categorizations were based on the Vehmas-Lehto s (2005: 49 74) and Hansen s (2010: ) articles on translation errors (presented in subsection ) and will be explained and discussed in more detail in the analysis of each category. This study is both quantitative and qualitative. The quantitative part of the study includes the numerical distribution of the errors, i.e. the translation errors discovered in the research material are divided into the selected categories and the number of translation errors in each category is counted. The translation error categories applied in this study are two umbrella categories based on Vehmas-Lehto s (2005) article, overt and covert errors, which are then further divided into the eight subcategories presented below. 47

53 Overt errors o Omission o Cohesion o Semantic o Stylistic Covert errors o Continuity o Connotation o Idiomatic o Convention Both the umbrella categories and their subcategories as well as the subcategories relation to the umbrella categories will be thoroughly explained in section 5. The quantitative side of the study will show how many errors there are in each category, both the number and their percentage of all errors. The qualitative part of the study will describe the errors in each category in more detail and examples, and also discuss the effect the errors are likely to have on the audience s viewing experience based on Heikkilä s (2014) and Vehmas-Lehto s (2005) research discussed in subsections 2.3. and , respectively. Due to the multimodality of audiovisual translation, subtitling cannot be compared to the original lines only word-to-word, rather they are compared utterance-tocaption while taking context into account. In this part of the study, the quantity of translation errors will be in relation to the number of translations. In this part of the study, the term translation refers to an individual caption uttered by one speaker, visible on the screen at one time. In other words, the captions are divided into individual translations according to how they are shown on the screen. Each caption is analyzed on their own as an individual translation, and those captions that have more than one speaker are further divided so that each speaker s line is analyzed as an individual translation. For example, both captions in examples 1 and 2 are analyzed as individual translations, even though one only has one sentence and the other one has two sentences. All examples used in this thesis are retrieved from the material of this study and are presented the way they appear on screen. Example 1. Äkkiä taivaalle ilmestyi tähtivyö. 48

54 Example 2. Jokaisen nimen pääte on ski. Kuinka se on mahdollista? However, captions appearing in the screen that have two different speakers are analyzed as two separate translations, as in example 3. Example 3. -Onhan se iso. -Hyvin iso. Furthermore, if one speaker s line is divided into separate captions, such as in examples 4 and 5, they will again be considered as two separate examples, as the division of one utterance to two or more translations, or combining two or more utterances to one translation, is considered a cognitive decision made by the translator. Example 4. Example 5. Heitä erottivat välimatka ja vanhemmat, jotka eivät hyväksyneet liittoa ja nuori pari uneksi ajasta jolloin he olisivat yhdessä. To separate the original lines from the captions in the examples, the examples presenting both the original line and its caption are on separate columns one beside another, the original line being on the left with the caption beside it on the right, the captions being presented in the same form as they are seen on screen. Example 6 below will illustrate this in practice. Example 6. Separated by distance and by parents who did not approve of the union, Heitä erottivat välimatka ja vanhemmat, jotka eivät hyväksyneet liittoa For the purposes of this study, the lines that are spoken by the characters but not translated will also be taken into account in the quantitative part of the study. Those lines will be counted as an individual translation when the speaker changes, i.e. no matter how long the spoken line is, or how many sentences it entails, it will be considered as one translation (that has been omitted) when the line is ended by either a long enough pause or another character speaks. For example, if a speaker should utter three untranslated sentences at once, it would be considered as one translation being omitted, whereas if the three lines should be uttered by two or more alternating speakers, the sentences will be considered as many translations as 49

55 there are changes in speaker. This is done because due to the omission, it is not known how the uttered lines would have been divided into captions by the translator. Discussing the translation errors in relation to the individual translations is done to show how the translation errors are distributed in the subtitling, to further illustrate that the occurrence of the errors is separate from each other. Furthermore, if a translation includes more than one occurrence of the same translation error category, it will still be counted as one erroneous translation. However, if a translation contains two translation errors of two separate translation error categories, for example, a stylistic error and a continuity error, the translation will then be included in both categories and counted twice. Thus, it is important to further note that from the entire research material, only 7 translations were discovered to contain two separate translation errors, and it has been taken into account in the report of erroneous translations. However, the translations with two errors do not affect the overall number of translation errors, as the total amount of translation errors in the material is counted by the number of translation errors, not the erroneous translations. After the translation errors are analyzed in their categories and the number of translation errors is discovered, the quality of the subtitling will be examined by analyzing the severity of the errors. The next subsection will explain how this part of the study is executed Quality evaluation After the quantity and category of the translation errors have been identified, the quality of the subtitling will be analyzed based on the translation errors severity. The severity of the errors and the number of them will be compared to O Brien s (2012) quality assessment model for translations (discussed in subsection ) to determine if the overall translation quality can be deemed acceptable on the model s standards. After the translation errors have been discovered from the subtitling and categorized in the first part of this study, the severity of each translation error will be evaluated based on O Brien s (2012) three-level scale, assorting the translation errors into three severity categories: minor, major and critical. The categorization of the translation errors will be based on O Brien s (2012: 62) explanation on each of the severity category (discussed in subsection ) and Vehmas-Lehto s (2005: 53) account on the hindrances of different translation errors in understanding and perceiving a translation, and how they relate to the severity of them. The categorization will also consider other aspects of subtitling quality discussed in this thesis, such as Vertanen s (2007b, discussed in subsection 2.1. of this 50

56 thesis) report on Yle s subtitling conventions and their purpose and Heikkilä s (2014, discussed in subsection 2.3.) study on how translation errors are noticed and discussed by AV translators and laymen. The principal argument that will be applied in the categorization of the translation errors will be the assumed viewer s ability to follow the plot (discussed in subsection ). The severity categories and how the translation errors are divided into the categories will be more thoroughly explained in the analysis of each of the categories, as it is done according to the results discovered in the first part of the analysis, by examining each translation error and its effect on the subtitling individually. When the severity category of each translation error has been established in an episode, the number of the errors in a category will be proportioned to the word count of the subtitling in the episode. This will be used to determine if the subtitling quality of an episode can be deemed acceptable according to O Brien s (2012) evaluation model. In accordance to the evaluation model, as discussed in subsection , the number of errors in each severity category will be counted and calculated so that they can be reported in the number of errors per 1,000 words format. When the relation has been calculated, this will then be used to determine if the translation quality of the episode will be acceptable by comparing the results to the commonly accepted agreement, or the pass/fail threshold, discussed by O Brien (2012: 63), stating that for the translation to pass the quality assessment, the amount of errors per a thousand words should be less than four in the minor error category and less than one in the major error category, and even one critical error in a translation would lead to an automatic fail on the translation quality evaluation model. When the quality of the subtitling of each episode has been evaluated according to the model, the third part of the study will be conducted. In the third part of this study, the differences between the subtitling in Netflix Original episodes and the non-original episodes will be examined through statistical significance to further investigate if the number of translation errors discovered in the first two parts of the study can be concluded to be in relation to the origin and producer of the episode. The next following subsection will explain how this investigation is executed Quality difference evaluation through statistical significance When the quality of the subtitling has been evaluated with the methods presented above, the third and final part of this study will examine if the findings can be used to express a correlation between the origin of the content and the quality of its subtitling. This will be 51

57 examined through statistical significance using the discovered numerical data. This subsection will explain how the statistical significance can be discovered through a statistical hypothesis test. Once the numerical data can be collected from the translation error categories, the data will be run through a statistical hypothesis test to find out if the difference in the quantity and severity of translation errors between the non-original episodes and Netflix Original episodes has a statistical significance. When implementing a statistical hypothesis test, must the study first constitute a null hypothesis, or a hypothesis that there is no correlation between the samples (Routio 2007). The null hypothesis is typically set as the opposite of the hypothesis that is actually being studied, the original subject then being named as the alternative hypothesis (ibid.). The collected data on the subject will then be run through the test to discover the p-value for the hypothesis, which can be then used to determine the probability of the null hypothesis (ibid.). The p-value, or the probability of the null hypothesis, can then be used to determine the statistical significance of the result, divided into the three levels listed below. Non-significant If the p-value is higher than 0.05, no correlation between the samples can be statistically proven, therefore the null hypothesis is deemed correct, as there is higher than a 5% chance that the correlation is a pure coincidence. Significant If the p-value is between and 0.05, there is less than a 5% probability that the correlation between the samples is a coincidence; therefore the null hypothesis can be abandoned in favor of the alternative hypothesis. High significance If the p-value is less than 0.001, the probability that the correlation between the samples is less than 0.1%, the null hypothesis can be abandoned and the alternative hypothesis can be deemed correct. In this thesis, the null hypothesis will be that there is no relationship between the producer and the quality of the subtitling (or the number or errors), therefore the alternative hypothesis of the statistical significance test is that there is a correlation between the origin of the content and the quality of the subtitling. To confirm the null hypothesis, the correlation between the producer and the number of errors must first be examined. This will be done with a statistical dispersion with the two variables, producer and errors. Depending on the 52

58 correlation coefficient, the study will then move forward in any of the three possible directions listed below. 1. There is a positive correlation between the variables (Netflix Originals are translated with fewer errors). 2. There is no correlation between the producer and errors (the Original episodes and non-original episodes have the same amount of translation errors). 3. There is a negative correlation between the variables (Netflix Originals are translated with more errors). Since the sampling is randomly selected from a larger set, the subset can be run through a statistical significance test. In the statistical significance test, the p-value will be set to 0.05, as the value has been commonly suggested to be sufficient. This means that when the received p-value is less than 0.05, it can be concluded that the quality of subtitling does increase when the translated content is produced by Netflix. The translation errors will be examined through the severity of the errors, by comparing if the severity and number of the errors can be influenced by the origin of the content. This will be done to discover if the assumed added investment of Original content on Netflix in order to gain leverage in the growing market of subscription services extends to ensuring the global viewing experiences by providing the content with a subtitling of high quality and to provide information on if the subtitling quality of Netflix has evolved from the time the non-original episodes were translated to the time the Original episodes were translated, and if the evolvement can be deemed statistically significant. As one of the main focuses of this study is to examine if the subtitling quality on Netflix Originals is different the non-originals, the quality difference evaluation through statistical significance has been separated as its own subsection in the analysis. This is done to emphasize the initial focus of this study and to provide a more distinct analysis on the issue, instead of combining the three parts of the analysis and discussing the issues together. 53

59 5. Analysis In this section, the number and type of the translation errors discovered from the research material will be analyzed, and the results will be applied to evaluate if the subtitling quality can be deemed acceptable and subtitling quality is in relation to the origin of the content. The analysis has been divided into the three main subsections discussed in section 4. First, subsection 5.1. will present the amount and category of the translation errors discovered from the material by discussing the translation error categories with examples from the material. After the translation errors have been analyzed in their respective categories, their severity and the quality of the subtitling will be analyzed in subsection 5.2. Then finally, in subsection 5.3. will analyze if the quality of subtitling is different in the Netflix Original episodes than in the non-original episodes Translation error analysis As mentioned before, the errors found in the randomly selected episodes of Gilmore Girls have been categorized into eight subcategories that fall under two umbrella categories, overt and covert errors. This subsection will present the categories, reporting how many translation errors were found from the research material belonging in each of the categories, present examples of the translation errors in each category and explain suggested reasons behind the translation errors when possible and how the translation errors might affect the audience s viewing experience. Among the 8,025 translations in this study, a total of 578 translation errors were discovered. From the material, seven translations had two separate translation errors, therefore 571 of the of the translations were erroneous, which means that an error occurred in 7.1 per cent of the translations. The distribution of the translation errors is presented in Table 2 below. However, it must be noted that due to the multiple separate translation errors in the seven translations, Table 1 does not take into account multiple errors in a single translation in the total number of the umbrella categories and the overall total, therefore the percentage of the umbrella categories and total percentage are not included in the table. 54

60 Table 2. The number and percentage of translation errors in each category Error category Number of translations with this error Overt errors 362 Percentage of all translations omission cohesion semantic stylistic Covert errors 216 continuity error connotation idiomatic error convention Total 578 To further demonstrate the distribution of translation errors by category, each translation error category s percentage of all translation errors discovered from Gilmore Girls are presented below in Figure 1. Figure 1. Percentage of translation error categories Percentage of all translations omission 19.7% 9.5% 3.3% 9.9% cohesion 9.9% semantic 24.6% stylistic 8.5% 4.8% 8.5% continuity 4.8% connotation 9.5% Overt errors 63.1% Covert errors 36.9% Overt errors 62.6% Covert errors 37.4% idiomatic error 19.7% convention 3.3% As can be seen from Figure 1, the most of the translation errors discovered from the material were overt errors, which constituted 62.6 per cent of the errors. Of the overt errors, the most 55