The latest goal of researchers at Google is to create a computer that can translate poetry. This ‘Al Complete task’ is “as difficult as anything we can attempt in artificial intelligence”, states Dmitriy Genzel, a research scientist at Google. What particular difficulties are there in the translation of poetry?
Google is not alone in the development of artificial intelligence with so-called ‘human dummy’ potential. IBM recently revealed ‘Watson’, a computer that is capable of understanding human speech and searching through a bank of pre-programmed knowledge in order to compete as a ‘contestant’ on the US game show ‘Jeopardy’.
However the contrast between the two companies’ systems is that whereas Watson possesses the factual and objective human qualities of data assimilation and regurgitation, the Google poetry translator would have to possess the subjective and emotional qualities of humanity that directly respond to our reading (and translating) of poetry. If Watson is our artificial head then the Google translator would need to be our artificial heart.
Genzel presented a paper outlining all the difficulties of artificially translating poetry at a recent conference, alongside possible solutions to some of these issues. One issue that he claimed could be solved was that of structure. The structure of a poem is often seen as academically ‘unique’ and constitutes a wealth of interrelated features that work to both individual and overall ‘effects’. These structural elements include the graphology of a poem, comparative line structure, syntax or word composition and the rhythmic and rhyming patterns of syllables and sounds within the words of each line.
In a way tackling structure could be seen as the best starting point for a system such as Genzel describes. Teach the computer the rules of a poem, its metre, stresses, rhyme, rhythm and so on, and you have a controlled -if not quite objective- framework in which to build. You could take an Alexandrine, for example, and tell the computer to use a frame of twelve syllables per line, with stresses on the sixth and twelfth syllables (as well as other accent and caesura rules). Genzel purports that the computer could easily understand this and as such translate a poem.
In terms of difficulties Genzel states that “the hardest thing to do is rhyme”. Rhyme isn’t a paralinguistic feature, what rhymes in one language may not necessarily rhyme in another, furthermore meaning is not always derived from the sound being ‘rhymed’. However, couldn’t it be said that structure also has this ‘subjective’ problem and by extension so do all the other features of poetry. You can objectively look at a poem and say that there are four lines per stanza or extended repetition of the letter ‘w’, but you cannot study these features fully without subjectively commenting on their effects and their overarching contribution to the emotional purpose or “feel” of a poem.
Genzel admits that it would be difficult (perhaps he means impossible) to recreate this purpose or “feeling” using computer translation; but he still underplays the problem by stating that there is a “big aspect of poetry translation that machines can do pretty well”. Except a poem is not an individualistic work, features cannot be seen simply to exist or be used, there is always an effect being created even if you aren’t focusing on the poem as a holistic piece.
Translating poetry or ‘literature’ is hard enough even with a human understanding of the nuances of language, a famous example being Cinderella swapping vair (fur) boots for her rather more impractical verre (glass) counterparts as the faerytale was translated into English. With poetry often being considerably shorter than other ‘literary’ works the effect of each word a poet uses is precious and can affect the tone or meaning of an entire stanza or work.
There is no point in retaining the framework of a poem if it no longer creates the effect it was supposed to do. As John Millington Synge appropriately says “A translation is no translation…unless it will give you the music of a poem along with the words of it”.