Is our mental lexicon structured like a tag-cloud system or hierarchical?

Question

Thinking about this discussion on meta i was reasoning about simple self-experiments you can do in psycholinguistics, where you don't need great background knowledge in Cognitive Psychology or Neuroscience. But where you can investigate and find out different phenomenological factors and properties, how our mental lexicon is structured. (Neurolinguistics tries to explain this phenomenology structurally)

So i gave Hauser (myself) 1 min time and a tag-keyword to say as many words as fast/spontaneous as possible and recorded them (e.g. Windows Sound recorder on your Desktop). I started with verbs. 1 min time. This went pretty good, around 100 of verbs i recorded. Same with nouns. Also pretty good, not surprising. How about adverbs. Hmm, it gets tricky. My brain didn't seem to have a crystal clear tag here. Additionally it was hard to think just of single adverbs and not adjectives (to deduce the adverb) or verbs often used with a adverb. How about locations? Pretty easy, many cities i could tell or places i have been. Then i thought about terms like Zeitgeist. You cant really define a good clear tag-name for words of this category. PROBABLY, because it is a very abstract term with contextual meaning. A lot of scientific terms are abstract, so how about quantities in natural sciences. Went good, as i have some background here. Interestingly, on the "verbs" tag it seems common to list a lot of related verbs for maybe 10 seconds (running, skiing, climbing, walking...) till you stop and then quickly switch to verbs related to e.g. the sub-tag "cooking" (cook, eat, smack,...).

So what can I deduce from such results (?) There seem to be words pretty good structured and tagged in clouds and sub-clouds (like verbs, nouns), also abstract words of your personal experience (education, hobbies etc.). Even if you might not use them more than others (e.g. prepositions) ,they seem to be tagged, while prepositions you use in nearly every sentence, are hard to think of spontaneously and better recognized in a distinct phrase they are often used in, there doesn't seem to exist a preposition tag in the mental lexicon (?). So visual things and activities (objects in real world and motor activities) seem to be much better tagged vs. abstract and not directly perceivable words.

To me it was a nice little experiment I would like to continue, with some tips, how this can be done the most productive and objective way to gain further phenomenological hints, how words and phrases are stored in the mental lexicon and contribute to linguistic topics like sentence building or TOT.

---

So what I'm asking for (which fits IMHO one question, because of link of theory - exp. method):

• What are current most investigated and favored models (technical termini) concerning the question "How is our mental lexicon structured?" that are similar to my self-invented experiment above recording a "spatial" map of tags/word-groups of my personal mental lexicon

• How do professional linguistic research experiments differ from my self-invented exp. described above? What are common phenomenological properties/quantities investigated in such experiments (e.g. semantic proximity), which exist on different human mental lexicons, in such tests? I would think, there have been done experiments recording really big mental lexicons of students and plotting them 2-dimensionally?

Nathan's answer is spot on and exactly what I'm asking for. Maybe you or anybody else can add some further details (papers), which cover similar experiments (to get a foot in the door), esp. how diff. mental lexicons differ between single persons. I would like to know what the phenomenological similarities and differences on those lexicons are and maybe map my own mental lexicon more systematically, build my own linguistic fingerprint.

Answer 1

I agree with the commentators that your question is a little confusing. However, I think I can get at an answer, and perhaps help you clean your question up a bit.

For a while, semanticists thought that our vocabulary was arranged in a logical, hierarchical way. Nouns are arranged so that they fall under each other, with 'sparrow' coming under 'bird' coming under 'animal' coming under 'life' coming under 'thing'. Something similar takes place for verbs. This is a very optimistic vision of semantics, as it implies that our brains work in a very logical, ordered way.

However, this turns out to be false. There are a number of classic experiments that can be performed that demonstrate this. For instance, you could show people names of different things and ask them to say if they are a bird or not. If you time their responses, you find that people are quicker to respond when shown typical birds, like 'sparrow', then they are for 'ostriche' or 'penguin'. This was taken to show that our semantic structure was a web, with archetypical words at the center and less "fitting" words further out.

Another experimental paradigm is priming. We show the subject a series of word and nonce words, such as "blurg" and ask them to determine whether or not they are words. We can prime a subject's response for a particular word, that is, decrease their response time, by showing them a related word before. For instance, if I show you "salt" and then "pepper", your response for "pepper" will be faster than if I had previously shown you 'President'. Cross-modal priming is when you prime someone for a written word by saying a related word, or vis-versa.

You can get really fancy with ERP (event related potential) experiments and fMRI scans and the like, and talk about activation patterns in the brain, but as far as your question goes we don't really need to.

Answer 2

A caveat: I'm experiencing the tail end of a three day migraine. This answer may not be as clear as I'd like.

The nature of the cognitive structures that make up linguistic knowledge is still a very open question. In fact, the nature of cognitive structures in general is still hotly contested (and rightly so).

Before I answer your questions, I think a little clarification is in order. Models of mental processes (at least from my perspective) are not likely to be accurate representations of the actual, real world functioning of the brain. Like any scientific theory, they're only as good as the predictions they make. So just because a model has support does not mean that it captures the nature of cognitive structures. Rather, it means that it accurately predicts behaviour. What I'm getting at is that different processes can produce the same behaviour, and just because a model makes accurate predictions, does not mean that it necessarily is an accurate model of the cognitive processes involved.

With regards to your first question. It's important to consider what type of language phenomena you're interested in. Different models attempt to capture the functioning of different processes, and each model has its own set of assumptions about the associated mental functions. Judging by the task you set yourself (which, by the way, is referred to as a verbal fluency test), you're interested in speech production, so we'll look at that. There's two influential theories of speech production, Dell's (1986) spreading activation model and Levelt et al's (1999) WEAVER++ model*. This article, by Levelt, is a great primer on the different types of speech production models. Both models assume that lexical knowledge is instantiated in a network, but make different assumptions about how processing proceeds through the network, and the different steps involved. I won't spend any time describing them since the Levelt article does a much better job of that.

Returning to my first point, this isn't to say that these models are accurate descriptions of the mental lexicon. Rather, they're influential models that do a good job of predicting linguistic behaviour.

On to your second question...

At a the level of experimental design, psycholinguistics experiments differ from what you describe in that they usually:

• Involve random allocation of participants to condition
• Have a large number of participants
• Have control conditions

That is, what you describe is not an experiment at all.

Some sources of information used to evaluate models of speech production are:

• Observational (non-experimental) studies of speech errors where the rate and type of error are recorded (Dell uses data from these as a basis for his model).
• Experimentally induced instances of the tip of the tongue (TOT) effect (e.g. Harley & Bown, 1998)
• Naming tasks in which participants are simultaneously presented with two pictures and the relationship between the pictures is manipulated (semantic, phonological etc). Participants have to name a target object in one of the pictures. Reaction times are recorded. (Meyer & Damian, 2007)

These data are usually used to support or weaken the argument in favour of a particular model. In the case of the models discussed, they are designed to address predictions stemming from assumptions about the stages of speech production, and the order (if any) in which these stages are processed. For instance, WEAVER++ assumes that syntactic and semantic information is accessed prior to phonological information. Thus the TOT effect occurs when phonological information is difficult to access. Harely and Bown (1998), showed that words with an unusual phonological form are more susceptible to TOT, evidence that supports WEAVER++.

As an aside, the 'priming' studies that Nathan refers to are in fact uses of the masked prime paradigm with a lexical decision task. These are used to investigate the processes involved in visual word recognition (i.e. reading), and are a great way to investigate the information we can access very quickly from written language. For instance, that we compute phonological information from brief (50ms) presentations of non-words (Kinoshita & Norris, 2012). What such information tells us about the mental lexicon is unclear.

This isn't meant to be an exhaustive account of speech production. I've undoubtably presented some contentious information, and excluded some important information. Regardless, hopefully I've given you enough resources to make a start on your own research. Most of the articles I've mentioned are accessible through google, at the very least, Dell (1986) and Levelt et al (1999) definitely are.

*Don't be put off by the length of Levelt et al (1999). Behavioural and Brain Sciences articles often include commentary by researchers in the same field, so the article itself is only about 35 pages long. The commentary is a great way to a get a feel for how a theory fits in with the rest of the field. Personally, I think it's a great idea.

Dell, G. S. (1986). A spreading-activation theory of retrieval in sentence production. Psychological Review, 93(3), 283-321. doi:10.1037/0033-295X.93.3.283

Harley, T. A. & Bown, H. E. (1998). What causes a tip-of-the-tongue state? evidence for lexical neighbourhood effects in speech production. British Journal of Psychology, 89, 151.

Kinoshita, S., & Norris, D. (2012). Pseudohomophone priming in lexical decision is not fragile in sparse lexical neighborhood. Journal of Experimental Psychology: Learning, Memory and Cognition, 38, 764-775.

Levelt, W. J. M., Roelofs, A., & Meyer, A. S. (1999). A theory of lexical access in speech production. Behavioral and Brain Sciences, 22(1), 1-38. doi:10.1017/S0140525X99001776

Meyer, A. S., & Damian, M. F. (2007). Activation of distractor names in the picture-picture interference paradigm. Memory & Cognition, 35(3), 494-503. doi:10.3758/BF03193289