I'll get to that chart in a bit, but first let me relate a little anecdote. A while back a Japanese friend of mine had me give a speech at his wedding. He wanted me to do it both in Japanese and then in English; the vast majority of his guests were Japanese speakers, but there were a handful who only spoke English and I was the only speaker who they'd be able to understand. So I was giving the exact same speech in Japanese and then in English.
My usual method for giving a speech is having a list of points I want to cover and then just making sure I hit them as I speak; the purpose of the list is only to remind me of what I want to say and not to provide me with the actual words. As this speech was going to be in both Japanese and then in English, I was language agnostic when preparing the list and it ended up being a hodgepodge of both languages. Here's a sample:
① same high school, 教科書に載ってない言葉
② invited me to his house
③ vegetarian: 食べれる大盛りのパスタ
Even if you speak Japanese, that will probably only make vague sense to you, but for me it directed me right to what I was going to cover in the speech.
After giving the speech, a few people who spoke both Japanese and English enough to understand what I had said in both came up to me and asked me how I had managed to make the speeches so similar in both languages—nuances, jokes, etc., were all exactly spot on in each language (a wholly separate matter is how some jokes worked better in one language than the other, but I digress). Their assumption was that I had translated it word for word and committed it to memory. In fact, I had used my little list to simply point my memory to a concept and then just expressed that concept however it should be expressed in the appropriate language.
And that gets me to Arkady Zilberman and his concept that "we think in a code language of images and associations".
Arkady lays out this idea in a guest post on Kirsten Winkler's blog:
We don’t think in our native language; we think in a code language of images and associations (this language we use in our dreams) which is connected to the native language. So it creates the illusion that we think in a mother tongue. People who speak fluently in a few languages have a rare ability to connect the code language of thoughts and feelings to different speech centers in the brain.That seems to explain nicely what I did with that speech. Although I only call myself "fluent" in Japanese reluctantly, within the confines of the vocabulary I know I certainly am fluent, so I'm pretty sure I've got this "speech center" he speaks of for Japanese. So, for the speech, the notes directed me to this "code language" which I then processed through the relevant speech centers for each version of the speech. Since the underlying code language was exactly the same, the speech came out pretty much exactly the same in both versions.
Arkady goes on to apply this to what he sees as the problem of most language learners:
In my opinion, most adults cannot escape from the habit of subconscious translation into their native language. In other words, they try to add a foreign language to the existing native language speech center.So to get back to the graphic at the top, this is how a fluent speaker will process language:
A speaker will interpret non-linguistic input (an image, a concept, etc.) directly in the code language. If it's then going to be expressed linguistically in some way, it'll go through a speech center to become output (speaking or writing). Linguistic input (listening or reading) travels in the other direction; it comes in via the appropriate speech center to be interpreted as code language.
On the other hand, this is Arkady's take on how a non-fluent speaker will process a language:
So the big difference is that there's no foreign-language speech center and the foreign language is interpreted through the native language before making it back to the code language. The inefficiencies this entails are obvious when you're considering actively translating something in your head, but he holds that they are also a substantial impediment even when it's done subconsciously and you don't think you're translating.
While I think these concepts are true when describing a complete language-learning noob and someone who's already fluent, I think the concept is much grayer than either "you are" or "you are not" going via your native language. Rather, it's a gradual process to go from one chart to the other.
When you first start learning a foreign language, you certainly are interpreting it via your native language. However, piece by piece, vocabulary from your target language is getting to the point where that trip through your native language is unnecessary. For instance, I'm skeptical that a native-English Spanish learner will be processing "¡Hola!" through "Hello!" every time they use it, as it's used so frequently.
Arkady might argue that the translation is just happening subconsciously, but this argument is undercut by words that don't exist in your native language. A great example of this is ごちそうさまでした gochisou sama deshita in Japanese. This is said in Japanese at the end of the meal. It's a formulaic phrase that indicates that you're done eating and, in more formal situations, has a connotation of being thankful. There's just nothing in English that plays the same role.
Or you could just take a simple noun that no one knows in the learner's native tongue. 法被 happi in Japanese is one example. It's a piece of clothing, of a certain shape and look, that's worn at festivals in Japan. (Wikipedia calls it a coat, but I'd say it's more of a shirt.) Trying to process a word like this through your native tongue would be utterly pointless as there's just no corresponding word. Even a person just starting out with Japanese that came across this word would just be forced to learn it without any influence from his or her native tongue. So, if we assume that every language has some things that your native tongue does not have, then it will be impossible to subconsciously translate 100% of your target language via your native language.
Let's turn back to the slightly tougher question then: words that do have a sufficiently corresponding word in your native language. When still relatively new to a language, I would agree with Arkady; you will be processing this through your native language (or some other language you know, but let's leave that wrinkle aside). But what is your native language in this case but a stand-in for the code language? Given that a code language cannot be written in words, your use of the native language is really just a link to the actual meaning of the word in code language.
In other words, your use of your native language to remember a word in your target language is nothing more than a mnemonic device to remember the actual meaning of that word in the code language. And just like with any other mnemonic, once the link you're trying to establish has been sufficiently established, you won't need the mnemonic device any longer and it will gradually fade away into oblivion.
To put that in terms of the above charts, the tool used to make the link (the word in your native tongue in your native-language speech center) eventually won't be necessary; you'll just remember the link between the word in your target language (in your foreign-language speech center) and the actual meaning (in code language). Word by word, you'll build a your very own language center, as demonstrated by this crappy GIF animation, showing how you move from one chart to the next:
Arkady goes on to say that 95% of adults lose their ability to connect foreign-language words directly with code language. As the use of native-language words is nothing more than a mnemonic (and typically a powerful mnemonic), I find this assertion particularly dubious (wholly aside from the other times I've discussed why kids have no particular edge in language learning). If he had said that 95% of language learners use some crappy, inefficient method of language learning (*cough* classes *cough*) and get poor results, then I might be on board.