On reading an article in a Finnish philosophy magazine about a new collection of Walter Benjamin essays translated into Finnish a few months, I was struck by the richness of his ideas on language and linguistic trends over time.

Following some initially unsuccessful efforts to understand certain aspects of the text, I realised Benjamin wasn’t just talking about language and culture — he was also talking about me, about my process of understanding a text in a foreign language, and about the seemingly inexorable drift we are seeing towards a disembodied language — the language of machine translation output and AI.

And since translators today are frequently called upon to revise such texts generated by machine translation and AI tools, Benjamin’s insights are now more relevant than ever, I believe.

The essential idea I gleaned from the article was the contemporary — that is to say, occurring about 100 years ago — decline in the art of storytelling. Particularly in “The Storyteller” (Der Erzähler), his essay devoted to the Russian author Nikolai Leskov, Benjamin describes and laments a transition from the epic genre, storytelling and the “exchange of experience” — firstly to printed texts and the novel, and then ultimately to mere “information”.

This transition from relationships to information provides a very useful and suggestive framework for describing linguistic communication on all sorts of different scales, and on many different levels, from the most general — the evolution of language as such — to the most specific — ie what happens every time we open our mouths or listen to what someone is saying to us.

It seems to me that the starting point on Benjamin’s transition — the genuine exchange of experience — is basically where speech comes from, where it originates.

This can be seen in the process of a young child learning to speak and converse. I remember watching a dad pushing a toddler in a buggy, with the two of them saying and repeating to each other “we’re going to auntie’s place”, or some such. The toddler was mimicking and imitating the dad’s tone of voice and actual words, with a greater or lesser degree of accuracy. This interaction between parent and child was not really “about” anything — whether they were going to auntie’s, uncle’s or the betting agency around the corner, for example — but was rather part of the process of the child learning to talk and converse.

So what I intend to address is the contrast between interaction and information, or an “interactional how” and an “informational what”. I see the former as preceding and transiting into the latter.

For example, in preparing this article as a conference presentation, my starting point was the imagined experience of speaking to an audience of friends and colleagues. Then, on the basis of that anticipated interaction I generated the “informational what” that I intended to deliver, as set down in writing in this text.

It was my anticipation of an interaction that generated the words on the page as I was writing it. The process would then start again in the delivery, as I adjusted the words of that written text to the audience I actually saw before me. Once again, an interactional how can be seen as preceding and underpinning the informational what.

On a much more general level, and on a vastly longer time scale, we might reflect that human languages themselves also negotiate a version of this transition. The mechanism involved here might be something like this: every human utterance converts unarticulated, preverbal content into an articulated, verbalised form. So, every utterance represents a sort of mini-victory of the verbal over the preverbal, of articulation over that which was not yet articulated. Over time, the aggregation of billions or trillions of utterances in the language in question would slowly but surely pull it from the interactional exchange end of the continuum towards the “informational what” end of said continuum.

This process might shed some light on the debate on the relative merits of indigenous knowledge in Aotearoa New Zealand — mātauranga Māori — and Western “science”. I see English as particularly far advanced towards informational whatness and away from interactional howness, Accordingly, mātauranga is imbued with a real exchange of experience in Benjamin’s terms — from the Māori perspective, this is knowledge of “our” ancestors that is now being shared or exchanged with members of “our” community today.

In contrast, Western science has no clear sense of belonging, origin or ancestry. Discoveries and advances may originate from this or that great scientist or this or that national culture, but that ancestry has now largely faded into the background. The cord has been cut, so that science is knowledge that simply “is”, irrespective of any parties it originated from or is being exchanged between.

In fact, I believe that all languages (and cultures) can be seen as lying somewhere on a continuum between kōrero (or “speaking”, “story-telling”) cultures and information cultures. Māori is definitely towards the kōrero end, whereas English has done a runner to the information end. Hence the emergence of AI and chatbots, where there is no real exchange, and no experience to be exchanged. They didn’t just emerge from nowhere.

But my primary here is located at a much more personal and specific level of the transition from interaction to information — ie the process that takes place every time we say something. Clearly it is always the interactional how that comes first, forming the substrate for the informational what that then emerges.

Similarly, the listening recipient first senses the nature of the communication relationship (as a child I seem to remember knowing I was in trouble with my mother before I had any idea of what I had actually done wrong), then more and more precisely decodes the informational content of the communication.

When we read a written text, the process is the same. At least for a text written in our mother tongue, we almost immediately sense the nature of the intended interaction or mode of exchange between the writer and us as the reader. Once again, interaction can be seen as preceding information.

But what happens when we translate from a “foreign” language (to us) into our mother tongue? This is where the process can get a bit messed up.

At this point I want to transit to the very personal story of my efforts to understand the final paragraph of the Finnish article I was reading that day on a Wellington suburban train.

This is the original text:

Seminaaripäivän kenties mieleenpainuvin puheenvuoro onkin tarina Pariisin ruumishuoneista. Ensin en tajua, mistä on kyse — eihän aihe liity Benjaminiin lainkaan. Mädäntyvien kalmojen haju ja kuoleman ihmittely täytävät kuitenkin pian hapettoman huoneen, ja unohdan, että kaikesta kuullusta tulisi olla jotain hyötyä. Tämän tarinan myös kerroin eteenpäin heti kotiin päästyäni. Lopuksi tarinankertoja kieltää itse lukeneensa Benjaminia, ja kutsuu meitä niin tehneitä hörhöiksi.

And this is my translation:

Perhaps the most memorable presentation during the day’s proceedings was a description of the morgues of Paris. My first reaction was, what on earth did this have to do with Benjamin? But the odour of rotting corpses and the dread fascination of death soon filled the stuffy seminar room, and I forgot the rule that all the day’s presentations should have some relevance or value for the participants. And I have to admit that this was the first paper I talked about later on returning home. In any event, the presenter concluded by telling us he had never read a word of Benjamin’s writings, and more fools we if we had.

Initially I found this paragraph very hard to understand. There I was, sitting in the train, letting my thoughts drift somewhat aimlessly over these sentences. Then, just as I arrived at my destination, I “got it”.

What did I get, and how did I get it? I realised that my obtuseness had been because of my failure to intuit the experiential exchange, the interactional how, that underpinned the text — which changed from one sentence to the next, as we shall see.

To highlight these changes, it may be useful to formulate some kind of narrative framing expression for each sentence.

The first sentence in the paragraph is a quasi-objective general statement, so the introductory narrative framing phrase might be something like, “In retrospect, I can tell you that…,”. This sentence is essentially presented as a fact, by a narrator looking back on the day’s events.

In the second sentence, we are taken inside the narrator’s mind as she responds to the situation then and there. Indeed, the narrative framing phrase indicating this is already there in my translation of the text: “My first reaction was …”. This is a much more immediate “exchange of experience”, if you will.

In the third sentence, we jump back to describing the seminar as it proceeded. We are still inside the narrator’s mind, but in a less immediate mode — these are remembered impressions rather than in the moment, here and now. “But as time went on I found that …” could be the narrative framing statement here.

The fourth sentence is a parenthesis, jumping forward in time to tell us what the narrator did later when she got home. In this case, the first words of my version, “And I have to admit”, do not appear in the Finnish, so here my narrative framing phrase is explicitly inserted in the translation.

And what of the final sentence? This is effectively the punch line, but in terms of the narrative mode it jumps back into the previous context, and reverts to the quasi-objective tone of the first sentence. We are back in “I can tell you that …” mode.

My conclusion is that because Finnish is still very much a foreign language for me, it took some time and effort to guess and intuit the interactional how of the text but, once I had done that, the meaning seemed to jump out of the page for me. As I say, I “got it”.

I then hypothesised that the experiential, interactional component that was so fundamental for my understanding of the text might be the Achilles’ heel for an AI-based machine translation programme tasked with translating this paragraph.

I anticipated that MT would nail the informational what of the text with supreme confidence — far outperforming my puny efforts — but its “what” would not be based on an intuited “how”; and it might not always convey the “how” of the text effectively.

I proceeded to run the paragraph through an MT programme. And here is the result:

Perhaps the most memorable talk of the seminar day is the story of Paris morgues. At first I don’t understand what it’s about — it’s not related to Benjamin at all. However, the smell of rotting corpses and the wonder of death soon fill the oxygen-free room, and I forget that everything I hear should be of some use. I also told this story as soon as I got home, Finally, the storyteller denies having read Benjamin himself, and calls those of us who did so fools.

This is quite an impressive translation. The software has instantly and effortlessly reproduced the informational what of the text, in a form and style that is very far from laughable.

I further surmised that the differences between my version and the MT output would be very similar to the kinds of changes I would have made as a post-editor.

There are only two obvious phraseological changes to be made: “oxygen-free” to “stuffy”; and “wonder of death” to “dread fascination of death”. In fact the former change, from “oxygen-free” to “stuffy”, can also be seen in narratological terms, since it is the nature of the exchange of experience that seems to call for an emotional (subjective) rather than a scientific (objective) term.

But most of the other changes can be understood as changes to the interactional how. Here are some examples:

1. The change from historic present to simple past has been made because in English, to my ear, the use of the present tense in the MT version gives the text a vivid, suspenseful feel that is out of keeping with a description of past events presented as essentially true and accurate;
2. Instead of “I don’t understand what it’s about”, I have used indirect speech: “what on earth did this have to do with Benjamin?”. This change is prompted by my sense of being inside the narrator’s head at this point — it’s not a matter of “correct English usage”;
3. “Everything I hear should be of some use” becomes in my translation “the rule that all presentations should have relevance or value for the participants”. The Finnish original actually reads “everything heard”, an agentless passive, if you will (heard by whom?). And of use to whom? The problem of supplying the missing person in such phrases is well known to all translators working from Finnish. The trick is to provide a convincing and coherent agent entity. Again, there’s no right or wrong here — it’s just a matter of getting the interaction between the writer, text and reader to feel right.
4. “I also told this story as soon as I got home” has in my version been picked out as a parenthetic statement by means of the inserted introductory phrase “And I have to admit,…”. This was my response to the tense switch in the Finnish from present historic to simple past at this point. Since I used the past throughout, I felt the need to pick this sentence out from the context in some other way.

As I look at the two versions, I also note my instinctive use of clichés — for example, to get the right relationships going within the text, and thereby catch the right note for the relationship between writer and reader. But the point I want to make is, firstly, that the machine translation programme has instantly nailed content that was giving me pause as a human translator (particularly because of my less than secure knowledge of Finnish); and, secondly, that what would have been my post-editorial changes were largely matters of narratology rather than grammar and syntax, on the level of the interactional how of the text rather than its informational what.

So perhaps Benjamin’s notion of a transition from storytelling, as a genuine exchange of experience, to mere “information” might shed some light on how we understand verbal communication in general, and in particular on how we translators understand texts in a language that is still quite foreign to us. It might also show us how to set about post-editing machine output that has already perfectly replicated the what of the original.

And it clearly does provide an excellent model for what the machine translation post-editing task entails — which is to put the “story” back into the “information”.

In a linguistic environment increasingly dominated by AI, it might show the way ahead — or a way ahead — for translator training programmes. We cannot hope to compete with the speed and in many cases accuracy of AI-based translation tools — but we can hope to complement them through a sharpened awareness of what I have called the “interactional how” of a text.

In other words, I believe that narratology in some form or other should play a major part in translator training programmes in future, if this is not already the case now.

Maybe we need to recall Roman Jakobson’s list of three forms of translation: intralingual translation; interlingual translation; and semiotic translation. We translators have perhaps been guilty of focusing almost completely on interlingual translation.

Maybe the time has come to look at intralingual translation as a major part of our future as a trade or profession — because that is precisely what the humanisation of AI texts is.

My quest is to get an unintelligible text to “talk” and tell a story, to put the kōrero back into the executive summary — and I believe that human translators are uniquely equipped to do just that.

So rather than despising post-editing and similar tasks, we can just as easily approach them with a sense of vocation. What greater challenge could there be than to translate non-human into human, to breathe fresh vitality into a desiccated husk of information? Practising a “love thine enemy” strategy, we translators might be able to find a profitable niche in our brave new disrupted ecosystem, as specialist humanisers.

This article is adapted from a presentation delivered at the conference of the New Zealand Society of Translators and Interpreters held in Tāmaki Makaurau/Auckland on 7-8 September 2024.