Behavioral Signatures of Memory Resources for Language

Looking beyond the Lexicon/Grammar Divide
Published: 18 Nov 2022

For long, research on language was guided by the assumption that there is a crucial difference between the forms or structures of a language a.k.a. grammar, and the lexicon or the words and expressions of a language. This is reflected in pedagogical materials: we tend to learn a foreign language by learning rules and using those rules to string words together into sentences.

The same division also plays an important role in linguistic theories: generative theory, for example, posits a rather strict division between grammar and the lexicon. Usage-based theories see grammar and the lexicon as different ends of a cline that runs from concrete form-meaning mappings as we see in lexical items on the one end, to abstract form-meaning mappings as we know them from grammar on the other end.

This division has a nice parallel in models of memory. Memory researchers generally assume that there are different types of memory. You will all have heard of long term memory and short term memory, or working memory. Within long-term memory there are two further distinctions: there is declarative memory, which stores facts, events or experiences which can be consciously recalled. And there is procedural memory which harbours unconscious memories such as skills, the knowledge of how to do something, like forming the shapes of letter or driving a car. The declarative memory system is assumed to be used by the lexicon, while the procedural system would handle grammatical knowledge. While declarative memory deals with idiosyncratic experiences where each experience needs to be mapped and there is no commonality (link here of lexical items that each have their own unique meaning), procedural memory deals with patterned experiences, or experiences that have a lot in common (think here of -ed that marks past tense in English).

Over the years, quite a bit of evidence has amassed to support this view, but virtually all of the evidence stems from research on English, which is a language with very little grammatical marking: most words in a sentence do not have any marking at all; they look exactly the same as the dictionary entry. What happens when we take a language that is very different and marks every single noun and verb for a variety of information (such as whether it is the subject or object, or whether the subject or object is male, or female, or neuter, etc). Is it still possible to distinguish so clearly between grammar and lexicon, and between the memory systems that handle the information?

There’s nifty little, but rarely used, experimental paradigm that should show which memory system is active. It’s called the dual task paradigm. As the name would make you suspect: you’re asked to do two things at the same time. And if those two tasks compete for the same memory resources, processing is slowed down significantly and becomes more variable. So, we ran a dual task paradigm that would impede access to declarative memory, but not so much to procedural memory.

First, in the baseline condition, we asked participants to listen to sentences containing a range of grammatical and lexical phenomena. After stimulus presentation, participants were given 5 seconds to judge the accuracy of the sentence that is to tell us whether the sentence they heard was correct or not, and we recorded how long it took them to take a decision.

Next, in the concurrent task condition, we asked participants to do the same time WHILE keeping in mind a sequence of three digits. So, first we showed them three digits, three six and seven for example, and they had to keep this sequence in mind while listening to the sentence and rating it. After they had rated the sentence as correct or incorrect, they were asked to write down the number sequence. Keeping a number sequence in mind would impede access to declarative memory, which harbours the lexicon, but not so much procedural memory which harbours grammar. So participants should struggle more to judge things that are lexical  in nature in the dual task condition, making their judgments less accurate, slower and more variable.

We analysed the time it took our participants to reach a decision, how accurate their decision was, and how consistent their judgment was. One of our four linguistic types, lexical information, is, indeed, mainly declarative in nature: judgments were slower, more prone to error and more variable. But all other types showed traces of procedural memory, albeit to different extents. Interestingly, syntax differed least from the lexicon under memory load conditions - and that is the opposite of what had long been assumed!

In sum, we found that the default mapping of language systems to memory systems may not accurately reflect language-use in healthy L1 users who speak languages that are formally more complex than English, and that’s the majority of languages! Our findings also confirm both dominant linguistic theories are partially right and partially wrong: there is a division between lexicon and grammar, but the division falls in a different place than assumed, and the distinction is graded: the hypothesised grammar (rule) – lexicon (idiosyncrasy) opposition appears as a continuum on which linguistic abstractions can be placed as being more or less ‘ruly’ or ‘defiant’, and more or less amenable to automatisation.

You can read all about it in our paper which has just appeared in Cognitive Science and can be read free of charge using this link.

In line with the Open Data policy of our project, data can be downloaded from the University of Birmingham edata repository UBIRA and code is on our GitHub page.