|
|
|
|
|
|
|
Dissertation Contents.
1. Why the Assistant Programme project was conceived. 1.2 Personal interest in this area of study. 2. Description of the Assistant programmes. 2.1 Programming languages and external software used.
2.2 The Assistant Programme layout.
2.3 Selecting phrases for the email. 2.4 Tackling questions of style in textual expression. 2.5 Search for phrases not found in the General Email Subjects section.
2.7 Assistant Trainer: for more advanced students. 2.8 Building the Corpora for the Assistant Programmes.
3.1 Machine Translation systems and how they work. 3.2 Practical applications of the MT processes. 3.3 Description of a translation software package.
4. Assistant Programme effectiveness 4.1 Translation effectiveness. 4.3 The methodological approach of the Assistant Programmes as a self-study tool.
5. Demand and practical implementation. 5.1 Comments on the ELAN report: “Effects on the European Economy of Shortages of Foreign Language Skills in Enterprise (2006)”.
***********
|
The philosophy behind the creation and development of the Assistant Programme project. Dissertation.
4. Assistant Programme effectiveness 4.1 Translation effectiveness. Although the principal intention of the Assistant Programmes is essentially as a learner's tool of an L2, I also emphasise the importance of its efficacy as a sound translating machine into the L2 albeit in the limited field of business email communication. Indeed, my argument is that the Assistant Programmes will be less successful as teaching tools if they fail to provide practical help for straightforward translation. A practical tool is essential at the employee's workstation, a purely didactic tool will be left for moments when the user has more time - if those fortunate moments ever occur! In this section, I shall appraise the effectiveness of today's machine translation but in comparison to the Assistant Programmes translation abilities. The purpose of this analysis is to determine whether MT tools could substitute the Assistant Programmes translation engine and therefore render it unnecessary as an inferior tool. In other words, could the Assistant Programmes translation tool be dispensable and be replaced by an existing MT tool and would the latter prove equally or more effective for the teaching objectives required of it?
4.1.1 Appraisal 1: Comparing texts. Assistant Programmes versus Systran. Systran is one of the world's leading providers of machine translation tools and offers powerful translators to the business world in a large selection of languages. Any attempt to compare translation efficiency of a Systran translator based on algorithm functionality and the limited glossaries of my Assistant Programmes may sound presumptuous. I must point out from the onset that a Systran translator accepts and deals with any text pasted into its textbox translating tool and offers an instant translation - at least of some sort. The Assistant Programme translation ability, on the other hand, which is the result of selecting from a limited list glossary is very modest in comparison. The users of an Assistant Programme may not find the lexis required to express their needs or they could be forced to compromise the original intention of an email due to a lack of suitable phrases available. Nevertheless, where comparisons can be made is at face value, resulting translated text if and when the user has successfully located suitable phrases from the Assistant Programme in their L1.
4.1.2 Results: Firstly, let us consider the translations into Spanish:
Secondly, the translations into English:
4.1.3 Conclusions to this test: The errors made by Systran translator in both translations were not of the same type. The translator seemed to have more problems with the translation Spanish - English than the other way around. There is an obvious difficulty to correctly assess the subject due to the Spanish linguistic custom of ellipsis. In general, however, the main error lies in overall consistency; this translator cannot recognise the necessity for register cohesion in one single text. The Spanish text starts with Muy señor mío (second person) and finishes with a mention of usted (second person) in the last line; despite this, Systran translator has elected, quite illogically, the third person impersonal it in the second paragraph and third person masculine him in the last.
4.1.4 Translation flaws in the Assistant Programmes Admittedly, although in this instance the Assistant Programme performed well, there are circumstances where the compiler's translations provide stylistic imperfections:
a) When compiling phrases from various parts of Email Compiler words can repeat leaving a text which is awkward to read:
Fig. 21
b) If the user has no knowledge of the L2, register inconsistency may occur. Usually, this is avoided by a like register in the L1:
Con el sentimiento de no poder resultarle de mayor utilidad, agradezco el interés en nuestra empresa. (With regret that I cannot be of any further help I thank you for your interest in our company - formal.)
Siento que no pueda ofrecerte más ayuda pero se agradece el interés en nuestra empresa. (I'm sorry that I can't be of any further help to you but thank you for your interest in our company - more informal.)
However, during the writing of this programme it was found imposible to always reflect the register in the translation process with separate and often short, out of context sentences. One reason for this is that the second person pronoun and verb variations in Spanish do not necessarily reflect a certain register as some Spanish-speakers may think. The use of USTED may be formal in many parts of Spain but in South America it could be combined with informal use. This is especially true of VOSOTROS which is non-existant in some South American countries and where USTEDES takes its place. Therefore a phrase like: ¡Vengan, lo van a pasar bien! may cast doubt in a Spanish speaker's mind as to whether selecting it might result in an unsuitable register in English. (It sounds informal in structure and meaning but the USTEDES use could confuse.) Considerations such as creating different programmes for different Spanish-speaking countries will not necessarily solve the problem: an example could well be the one we have just discussed - many Andalusian speakers might use the USTEDES form for informal colloquial speech while a speaker from Madrid would not contemplate the informal use of USTED.
4.1.5 Solving the problem of flaws in the Assistant Programmes. The only way to avoid errors such as these happening is to offer the user instruction in the L2. And here I come again to the essential characteristic of the APs: unlike machine translation, the Assistant Programmes translation success must necessarily lie in its intrinsic didactic properties - accurate translation which circumvents stylistic and register idiosyncrasies is only achievable by at least a minimal amount of instruction which provides insights into the nature of the L2. This inevitable characteristic could be regarded as negative if the user requires nothing more than an accurate translating machine or positive if, on the contrary, the user has an interest in learning or improving L2 skills.
4.1.6 Appraisal 2: Considerations regarding Hutchins' three categories. It is interesting, I believe, to analyse and distinguish between the approach to MT from the viewpoint of a professional working in MT as a researcher and developer (such as Hutchins) and the lay user of MT. Hutchins perceives MT software systems “as is”. Rather than emphasising the limitations, he prefers to highlight its practical applications among professional translators and lay users. However, I find it difficult to imagine that any MT tool was purposefully developed for any of the three uses Hutchins defines: assimilation, interchange and dissemination. MT developers surely strive to create a machine that can match the quality of output that professional human translators are capable of. That goal is proving a problematic objective. Hutchins, on the other hand, seems to be saying that the current MT “inter-product” provides a valid service in its presumably immature state. The lay user, though, appears to place a value on MT tools which is directly proportional to that tool's capability of rendering a perfect translation. Poor tools are criticised rather than considered useful for “assimilation” and more accurate translation tools are more highly regarded.
These perceptions regarding the lay users' opinions on MT I have gleaned from reading software forums on the Internet. I cannot offer detailed questionnaire results or statistics to support my claims except that there is some common sense underlying a theory that MT software may not be neatly divided into the three categories that Hutchins suggests. The very name “machine translation software” leads the lay user to assume it should do just that and the user is presumably disappointed if the tool fails to translate effectively. I can find no software called “gist translation” or “assimilation translation” etc. on the Internet. Neither is Hutchins' term, “gisting”, an accurate one. The dictionary meaning of “gist” is: the main or general meaning of a piece of writing, a speech or a conversation. To write the gist of any given text would be to list the principal points and ignore detail. There is no MT tool which is able to carry out such a complex linguistic task. When a text is translated in one of the free translation tools listed earlier, the resulting text is the product of a machine attempting to provide a perfect translation which fails. If the errors in the MT tool's output are not excessive or comprehensible then the user may be able to glean some meaning related to the original text. This meaning could be the main points or the detail, there is no guarantee that the gist will be conveyed. Neither is there any guarantee that the translation will not convey a “gist” which is erroneous and entirely misleading vis-à-vis the original text.
I present as evidence that a lay-user's opinion of MT software tends towards a qualitative appraisal, by quoting the comments in one particular forum whose participants are apparently of university education and who appear to have had experience with an ample range of MT software. The forum was a Google group entitled: “English to Spanish Translation Software for Mac and PC”.
The question was:
“Has anyone had any personal experience or have any knowledge of reliable, (Carolyn J. Williams. Graduate Student Arizona State University)
First reply:
Yes. I have. First of all, nothing beats a real live individual for good
There are essentially 3 categories of translation applications. The prices
0$ to $150 I consider these almost useless, EX: Belitz, Parsons, Power
$250 to $500 These do a mediocore job. The translations will need some
$1000 and up: These do a vastly better job. They 99% of the time produce
(Eduard L. Frerking, Sunnyvale, CA, Spanish Kindergarten Teacher, Hillsdale Year Round School Network Admin.)
My comment: (Interestingly, and totally coincidental of course, Mr Frerking defines three categories as Hutchins does. Conversely, however, Frerking's categories are purely qualitative; rating the software according to how close to perfect translations they are capable of.)
Second reply:
There is still not a satisfactory Spanish translator available for a (Andrew B. Wolff, Department of Spanish, Italian and Portuguese, N342 Burrowes, Penn State University.)
My comment: (Mr Wolff here inadvertently hints at Hutchin's category of Dissemination, though also offers reasons to discard the idea of revamping MT tools' results suggesting that, in many cases, it would be faster for a good translator to start from scratch. I can only add support from my own experience of professional translating by saying that I agree with Mr Wolff. Furthermore, I mention the common professional practice of human translator “correction”, whereby one human translator corrects another human translator's work. Even in these circumstances, unless the first draft contained very few mistakes, I found shorter texts were better re-translated from the original foreign language text - ignoring the translation with errors altogether.)
4.1.7 Conclusions to Hutchins' three categories appraisal. I consider that, on the whole, Hutchins' categories of MT applications are, at best, hopeful and optimistic. It does seem true, apparently, that Assimilation usage plays an important role in the comprehension of foreign texts. Google's webpage translator tool has presumably been researched to decide whether a demand for the service exists. However, the question is not the demand for machine translation but rather MT tools which offer a reliable and comprehensible text - at the moment tools immediately accessible to the Internet user such as free or cheap online services, do not necessarily always provide reliable gist translations. If Assimilation is an authentic application of MT, it is rather because this is better than nothing at all and not because users are actively seeking approximate translations as Hutchins appears to be suggesting.
It would be interesting to discover to what extent MT is really used as an integral part of Translator Workstations (Dissemination). It is indeed true that, approaching the problem from the other direction, MT software is borrowing tools from Translator Workstations to improve overall accuracy in output. Systran Business Translator (priced from $199 to $379) summarises its benefits like this:
Those feature underlined, I consider to be supplementary tools separate from the MT engine. They are, in fact, Translator Memories (TMs) as described earlier. The question here is whether features such as “creating personalized User Dictionaries” can be included within a definition of MT; the operation of such features must require the user to have some previous foreign language knowledge or at least linguistic knowledge. In other words, although this product is marketed as MT software this is not in fact an accurate description.
Professional human translators may find MT useful for large texts which can later be corrected but as has been suggested earlier, text correction creates multiple problems for the translator and it may often be better not to attempt MT translation at all. Further research on this subject would be interesting to reveal just how much MT is used by human translators.
Lastly, we should consider Hutchin's application of MT in Interchange. Another interesting study to carry out would be what proportion of users requiring translation of emails for communication purposes are content to relay ideas to native speakers via a poorly expressed machine translation - especially when there is no guarantee that the sent message actually even expresses the author's intention. Hutchins' quote: “...there is little objection to the inevitable poor quality.” deviates from the point. By the term “poor quality” Hutchins must be referring to punctuation, spelling and occasional lexis or grammar errors or such errors that do not impede at least a gist rendering of the original text. Perhaps such errors that a low level student in the foreign language might produce and that the receiver could divine and thereby compensate for while reading the resulting text. In the case of machine translated texts, the errors are not of this kind but rather computer-brained “reasoning” errors regarding lexis, syntactical, morphological and grammar choice that most human-brained lay readers find impossible to decipher. Therefore, surely the user's objection to sending MT texts would be overwhelmingly negative if he/she suspected that the translation being sent was possibly incomprehensible and perhaps risible or ridiculous.
Admittedly, more expensive and therefore better equipped translating machines will produce a higher percentage of more accurate translations. The question is, as the user is presumably not conversant in the foreign language, he/she cannot check or amend the resulting text before sending it in the email so will a user be satisfied that even one in ten translated emails sent conveys the wrong message and unbeknown to the user (comparable perhaps to a game of Russian roulette) which email was the bogus one? In all likelihood, the importance of the message being sent will influence the user's concern as to its accuracy. In business operations, orders, product offers, invoicing issues and other communication requiring the utmost precison, the user will be most reticent at using MT. Another area where language precision is imperative is where the user demands a correct register to transmit a message where subtlety is essential such as in product and service complaints and appraisals.
Below, I offer a selection of likely business-type neutral/informal register phrases which might be sent by email (taken form English Assistant Business) which have been translated by the free version of Alta Vista's Babelfish English to Spanish. The selection is only those phrases I found where meaning could be seriously impaired (this was not a random input selection):
Sure, I'll get on to it right away! ¡Seguro, le conseguiré encendido enseguida! (Human translation from the AP: ¡Por supuesto. Me ocuparé del tema de inmediato!)
I'll pass the message on. Pasaré el mensaje encendido. (Human translation from the AP: Le daré el recado.)
Sorry, I can't help you out there. Apesadumbrado, no puedo ayudarle hacia fuera allí. (Human translation from the AP: Siento que no pueda ayudarle.)
OK, please go ahead. La AUTORIZACIÓN, va por favor a continuación. (Human translation from the AP: Por supuesto. Adelante.)
We've got to iron out our differences if we are to make a deal. Tenemos planchar fuera de nuestras diferencias si debemos hacer un reparto. (Human translation from the AP: Hace falta superar nuestras diferencias si vamos a hacer negocios.)
For the bilingual linguists reading this dissertation I believe what will be immediately evident is that the translating machine performs poorly with the English phrasal verbs in the Spanish translation: GET ON = conseguiré encendido; GO AHEAD = va por favor a continuación; HELP YOU OUT = ayudarle hacia fuera; IRON OUT = planchar fuera etc. If we re-write the last phrase avoiding phrasal verbs and idiomatic language we achieve a comprehensible translation from Babelfish:
We must agree before we can do business. Debemos convenir antes de que poder hacer negocio.
There are three important points to make here:
Any adaptations a user might make to “help out” the MT engine could impair the register the user wants to write in. For example, We must agree before we do business is a more formal register than the more natural or neutral: We've got to iron out our differences if we are to make a deal.
4.1.8 Conclusions to Translation Effectivess appraisals.
Firstly, it is important here to reinterate the essential feature of the Assistant Programme translation engine. As a didactic tool in acquiring an L2 it is imperative that the resulting translation be accurate or that sufficient help to the user is provided so that the resulting translation could be accurate by applying accessible explanations in L1 to render accuracy. For example, a user finds a suitable sentence in Email Compiler with the (?) variable, which must be filled in by the user. This example taken from English Assistant Business (for learning English):
Please send (?) as soon as possible.
There should be L1 (Spanish) help, in this case, as to how to locate the required word or words to fill the (?) gap:
Please send some samples as soon as possible.
The help required and provided here was:
The Email Compiler section of the Assistant Programmes should provide accurate translations if:
Having established how translation accuracy can be obtained in the Assistant Programmes, we can now consider whether MT would serve as a useful alternative within a language learning programme from the point of view of translation effectiveness.
MT tools even at the highest technological standards can produce inaccurate translations. Unlike the APs' translation engine, MT translation mistakes are not necessarily caused by errors in the programme that can be rectified; the results are due to the limitations of MT software technology at the current time. The dynamics of translation procurement are MT's own downfall. The user is allowed to insert whatever text he/she likes into the translation text area with the consequence that the software must produce results for an infinite array of lexis “clumps”. The Assistant Programme translation dynamic is to choose from the lexis clumps provided and therefore controlled by the programme as these have been pre-translated. Naturally, the limitations in lexis choice are a major drawback to the Assistant Programmes and only by drastically decreasing the user's possible lexical needs by reducing the expanse of the subject matter to basic procedure business emails is it perhaps possible the user will find lexis to communicate his/her required message. In this respect there exists a similarity between the needs for better SMT, EBMT and an improved Assistant Programme translation engine: larger pre-translated texts or glossaries (corpora) would provide improved translation performance. In the case of MT, larger bilingual corpora (and faster computers) would yield more accuracy in translation and regarding the Assistant Programmes, increased bilingual glossaries would offer more lexis choice for the user. But despite the limitations in Assistant Programme translation the pre-requisite of accuracy could, in a perfectly written programme, be offered to the user. In this respect then, an MT tool would be inferior to the Assistant Programme translation engine for the purposes of accurate translation.
4.2 Teaching potential.
Here I shall attempt to analyse any capacity which existing MT programmes may possess to provide the user with a language acquisition context and compare that quality with the APs' translation engine in the same context. MT programmes are not advertised as language teaching devices but purely as translation software. However, I find this area essential to discuss as the rationale behind the Assistant Programmes comprises a translation tool which teaches albeit a translation tool combined with help features to achieve a faithful translation. The question which begs, then, is whether an MT tool together with help tools (such as the Workstation offered in Systran's Business Translator - see above for description) could improve on the didactic role of an Assistant Programme. Firstly, I shall describe in more detail the didactic context and methodological focus of the Assistant Programmes before finally making a comparative assessment of pedagogical quality.
4.2.1 The language-learning context of the Assistant Programmes - student types.
I find it difficult to identify a similar medium/tool for foreign language teaching to that which the Assistant Programmes provide so the only comparisons which can be made will be with reference to previous teaching methods and approaches that in some way bear resemblance to the teaching approach of the Assistant Programmes. This means that an argument regarding the nature of the ideal language-learning context of these programmes must be partly through conjecture. My thoughts on this subject, therefore, are theoretical until empirical studies can be carried out.
4.2.2 Assistant Programmes and low-level students.
Here, I shall discuss whether these programmes could prove useful as a self-study tool for students at a beginner or pre-intermediate L2 level for language acquisition purposes (as explained earlier, even users with no L2 skills can write emails in L2 via Email Compiler but here we are concerned with any potential progress in language learning that may result). In a pedagogical context the Assistant Programmes may well not be very useful to low-level students of L2. I believe that such students will need to struggle with grasping more rudimentary elements of grammar and vocabulary to fully comprehend the sentences they are structuring in Email Compiler. Furthermore, being more inexperienced linguists, they will not recognize and appreciate those essential language constituents which differentiate business email language from other types of written and oral language such as colloquial oral and written styles usually studied up to intermediate level in schools and academies or quasi-business terminology that adult employees at beginner level often encounter on in-house company classes. Indeed, although the Assistant Programmes offer grammar and vocabulary help at all levels of Spanish and English language learning, they cannot be considered sufficiently comprehensive in didactic material to provide a self-study teaching medium for low-level language students. The often-cited new language acquisition formula of “i + 1” would not be met as there would be too many unknown language elements between the student's L2 knowledge and the complete, authentic sentences of Email Compiler resulting in frustration on the student's part. Krashenist theory, that “i + 1” is not necessary as long as L2 input is comprehensible (Krashen, 1983), can neither be applied here. Interestingly, and seemingly paradoxically, a parallel foreign language translation of an L1 text cannot be considered sufficiently comprehensible to a low-level language student for acquisition purposes. The student is provided only with a solution but if he or she fails to see how that solution was obtained, it cannot be deemed useful. Consider the English - Russian translation below:
If the reader of this dissertation is unfamiliar with Russian, you will appreciate that the translation offers little evidence as to how it was derived. Presumably, the first word means “unfortunately” but any more assumptions that the L2 reads word for word as a mirror image of the L1 would be facile and misleading. Admittedly, an Assistant Programme would provide more analysis assistance than an MT translation output text area (such as the example above). An Assistant Programme translation output is broken down into smaller clumps:
Despite the presence of brackets, the variations in syntax in Spanish and English do not always allow consecutive bracketed section order eg. si (nos) (enviara) and if (you) could send (us); and the variables do not always match: (Estaríamos agradecidos) and (We'd) be grateful. This translation analysis in the Assistant Programmes, though useful for higher level language students, may not be sufficient for low-level students. Furthermore, is an analysis of this kind conducive to L2 acquisition? I should suggest that there would still be too much new lexis and grammar material in Email Compiler output for low-level students to analyse easily and efficiently; too far beyond the formula of “i + 1”. To substantiate this claim one could consider that if the Russian text shown above were divided into bracketed clumps (and the user had no or little knowledge of Russian) acquisition difficulties would still not be removed. “Comprehensible input”, in L2 acquisition terms, must therefore mean that not only more linguistic evidence is provided but a large proportion of the input is already familiar (has already been acquired) by the learner. To sum up, low-level students of English could use the Assistant Programmes without teacher guidance to successfully write and send emails but owing to the large number of unfamiliar language structures the learner would encounter, L2 acquisition would prove slow, laborious or even impossible.
The same low-level students as mentioned above may benefit more from the Assistant Programmes if a language teacher was involved in instruction. A teacher could be selective about the areas of language to work with so that learners need not become exasperated at the sheer amount of material available. Teaching business language from beginner levels is now very much accepted as sound language teaching methodology as is reflected in the ample selection of business English course books which can be found in the TEFL market. (See Bibliography).
4.2.3 Assistant Programmes and post-intermediate students.
I believe that where greater effectiveness for potential L2 acquisition would be reached is for learners at an intermediate or post-intermediate level. Students at intermediate and above levels are presented with a much reduced learning agenda when working with the Assistant Programmes. The formula “i + 1” is more closely respected. Here I am considering either a self-study context (a teacher led class could be more controlled). The main new linguistic elements to be learnt are:
For higher level students, acquisition of the material involved in the above listed tasks is not ominous. If the learning objective is to equip students already conversant in the essentials of general L2 with the linguistic knowledge to make the adaptations required for business correspondence, then I believe the language-learning goals are more straightforward and realistic and acquisition could take place especially if the learner is allowed to use the programme outside the classroom and directly as a tool for communicating via email with genuine L2 speaking customers. This vitally important context of a tool for practical use is the principal rationale behind the Assistant Programmes conception and which I shall now attempt to analyse.
4.2.4 The language-learning context of the Assistant Programmes - methodology.
The Assistant Programmes can be used with a teacher as a blended learning tool where Assistant Programme instruction takes place combined with teacher input in the classroom or via Computer Mediated Communication (CMC) with other materials and resources. In fact, as the degrees of teacher intervention when working with a PC programme can be very varied the different teaching approaches possible would be numerous. It is not my intention to delve into discussion on the subject of blended learning because my comments would probably be generic of most other types of didactic software with teacher support and therefore not be specific to the Assistant Programmes. I prefer to concentrate on any unique didactic capacity the Assistant Programmes hold as a self-study tool.
| ||||||||||||
