Are we going somewhere, or simply going? Ask your translator.

‘I am going to the mall’ / Yo ies la mall.

Where did the directional preposition “to” go?

The English-to-Romanian translation of many directional phrases seems at first both incorrect and radically oversimplified. Instead of using a preposition like “to”, in this case, Romanian uses the verb “a ieși” – essentially, “to go to [somewhere]”. These uniquely inherently directed verbs, also known as inherently telic verbs, are present not only in Romanian, but also in various verb-framed language types that give information regarding the direction of motion through verbs. These languages, called “verb-framed” languages (as opposed to their counterpart “satellite-framed” languages) are discussed in-depth in Dan I. Slobin’s, “The Many Ways to Search for A Frog”. Due to my understanding of English, a satellite-framed language, as well as French, Spanish, and Romanian, all Romance and verb-framed languages, I found this reading the most interesting as it inspired me to juxtapose the variety of linguistic typology in regards to motion events.

The differences in syntactic expression of motion events between S-languages and V-languages are also particularly evident in the case of expression of manner of motion.

Since “[S-language] lexicons provide a large collection of verbs that conflate manner with change in location,” (Slobin 220) S-languages such as English or German may seem more elaborated in dynamic path descriptions. For example, one could say, “The bug crawls home,” and express both direction and manner because of the word “crawl” which conflates location and manner of movement, whereas in Romanian one would have to say “insect merge acasa pe brânci”. Evidently, an adverb is needed to translate this V-language’s expression to an S-language with the expression of both manner and direction. Clearly, these languages not only differ in vocabulary, but also in syntactical representation of motion events. 

These linguistic differences may seem minute, however, when put into the context of artificial intelligence, machine learning, and automated translation systems, these crucial syntactical differences lead to a glaring conclusion: in order for an accurate translation of a motion expression to occur between S-languages and V-languages, the machine must have the ability to gain a semantic understanding of both manner and direction of the expression, store that encoding, and then process it into the other language’s syntactic composition. This discounts the accuracy of generic word-for-word translation methods in favor of recursive methods with high semantic processing performance. 

Conversely, the ability to selectively pick between near-synonyms based on syntactic environment and attributes, as seen in Atkins’ and Levin’s “Building on a corpus: A linguistic and lexicographical look at some near-synonyms*” further reinforces the need for increased syntactic analysis in translation rather than reliance on semantic understanding. In this reading, near-synonyms of “shake” that are common semantically, coined “the ‘Shake’ verbs”, were proven to have differing “internal vs. external causation [that]…account[s] for the[ir] apparent idiosyncratic [syntactic] behavior[s].” (Atkins & Levin 107) In order for an AI translator to be able to correctly pick the grammatical near-synonym (from the “shake verbs”), for example, it must have an encoded understanding of which verbs in its lexical corpus are transitive and intransitive and recursively examine which type is required for that particular location in the sentence.

These challenges in translation are exacerbated by the myriad word-forms and compounds discussed in Hapselmath’s 1 and 2; without an innate understanding of which part of a complex word in a language forms the root lexeme and which morpheme modifies it, a language translator does not have the ability to fully understand the semantic meaning of that sentence, process it, and then translate it syntactically to another language. This is made all the more difficult if the languages are being translated from S-languages to V-languages, and vice-versa – which brings us back to the difficulty in comprehending the broad range of linguistic typology regarding motion events. 

Evidently, machine translators have a tough job ahead of them – the necessity of recursively looping semantic and syntactical processing of text in order to make sentences correspond grammatically in other languages is only the tip of the iceberg of what is needed for accurate translation.