The Ascent of Babel: An Exploration of Language, Mind, and Understanding
Page 9
In the following chapter, we leave behind the question of what, if anything, constitutes the access code, and consider instead what `recognizing' a word actually means. In terms of the analogy with searching through a written dictionary, at what point (in time) do we read off the information contained within the lexical entry, and what determines when we reach this point? In addressing this issue, we can remain agnostic with respect to the nature of the access code. At the end of the day, more important than the denomination of the notes is what you can buy with them.
Words, and how we
(eventually) find them
The history of science is littered with examples of analogies that do not work. Often, they are simply inappropriate, simply wrong, or simply confusing. But even when inappropriate, they can prove useful. For instance, it is not unnatural to think of our knowledge about the words in our language as residing in some sort of dictionary. The Oxford English Dictionary (OED), all 20 volumes of it, is as good an example as any-its purpose is to provide, for each entry, a spelling, a pronunciation, one or more definitions, general knowledge about the word itself, and perhaps a quotation or two. Getting to this information is relatively efficient. You scan down the page, ignoring the definitions of the words you are uninterested in until you come to the word you want. Of course, if you are lucky enough to possess the dictionary on CD-ROM, you do not even need to scan down the page (let alone pull the appropriate volume off the shelf); just type in the word, and up pops everything you ever wanted to know about it. But so what? Why should anyone care about how the OED works, CD-ROM or no CDROM? In the last chapter we saw that the analogy between accessing a written dictionary and accessing the mental lexicon is at best fragile. So why carry on with it? The answer to this is simply that it provides a useful starting point from which to proceed, using a vocabulary that is easily understood to describe a process (accessing the mental lexicon) that is easily misunderstood.
Describing the mental lexicon as `something like you've never imagined' is probably accurate, but certainly useless. At least a conventional dictionary can be imagined, and is therefore a useful place from which to start our exploration of the mental equivalent. Most importantly of all, the questions one can ask of a dictionary such as the OED, and the questions one can ask of the mental lexicon, are remarkably similar. The answers, though, can be surprisingly different.
Before delving into the mental lexicon and looking at how we retrieve words, we need first to address an important, if basic, question: what, exactly, are these things we call `words'?
Words and what they contain
The purpose of language, and communication in general, is to convey meaning. In spoken language, the word is the smallest stand-alone thing that can do this. It is not, however, the smallest verbal gesture capable of expressing meaning. An `s' added onto the end of `fact' also expresses meaning; namely that we are dealing with more than one fact. So words can generally be broken down into even finer units, called morphemes. This last sentence contains 13 words, but 21 morphemes-the word `units', for instance, consists of the morpheme `unit' and the morpheme `s', and the word `morphemes' consists of `morph'+'eme'+'s'. Figuring out which kinds of morpheme can be stuck onto which other kinds, and how this affects the meaning of the resulting word, has been studied within a branch of linguistics called morphology.
There are different kinds of morphemes: stems (e.g. `unit', `word', or `speak') and affixes (e.g. `-s', or `-ing'). Languages differ in terms of where they put their affixes-in English, the most common affix is a suffix, coming at the end of the word, but we also have prefixes which come at the beginnings of words (e.g. the `pre-' in `premature'). Some languages (for instance, Tagalog, the language of the Philippines) have infixes too; these are affixes that are inserted in the stem. Mark Aranoff, a morphologist, has an entire section in his book Word formation in generative grammar devoted to one of the few English infixes-'fuckin' as in `fan-fuckin'-tastic'. But affixes are not the only device we can use for modifying the meanings of words; the irregular past tense in English`run-ran', `speak-spoke', and so on-is a remnant of a time when the past tense was produced not by adding the suffix `-ed' onto words, but by modifying a vowel in the stem. In a Semitic language like Hebrew, this is the rule, rather than the exception.
To complicate matters further, different kinds of affix do very different things. Some affixes are called inflectional; these include the plurals ('-s'), and the various ways of inflecting verbs (e.g. for `govern': `governs', `governing', `governed'). Inflectional affixes do not change the meaning of the word, but convey additional information relevant to it. Derivational affixes, on the other hand, do change the meaning; they are used to derive new words, so from `govern' we can get `governor', `government', `governance', `governable', `ungovernable', 'governability', and so on. Although related in meaning, each of these words means something different, and in the case of `governable' and `ungovernable', they mean exactly the opposite. But not all derived words are related; `casual' and `casualty' are unrelated in meaning, as are `depress' and `express' (although a glance at the OED will reveal their common historical ancestry). Another complication (there are several more) is that although the majority of stems can be free-standing, there are some inflected words which do not contain a free-standing stem, such as the verbs `permit' and `submit'. And whereas the meaning of `ungovernable' can be deduced by stripping away the affixes and recovering the meaning of the stem `govern', the meaning of `permit' and `submit' cannot be deduced from the meaning of `mit'.
So words are complicated things. And knowing whether it should be called a word or a morpheme, an affix or a stem, a prefix or a suffix, an inflection or a derivation, matters far less than knowing that words have an internal structure. Somehow we have to strip off the excess (but important) baggage, and reveal the word's core. And sometimes it looks as if we ought to do this, but in fact we should not (as in `permit', `report', and so on). Linguistics has told us an enormous amount about how words are structured and how the meaning of a word is dependent on the meanings of the different morphemes it is composed of, it tells us which kinds of morpheme can be combined with which other kinds, and in which order. But that is just the periodic table again-it tells us what the result is, but it does not really tell us how the result comes about. It does not explain how the brain comes to acquire the conventions that tell us that 'un-' on the front of a word results in a meaning that is the contrary of whatever the word meant in the first place. It does not tell us where this knowledge is stored, or how it is stored, or how the brain takes a complex word like `unspeakable' and breaks it down into its components, or even whether it does break it down at all. All these questions fall under the remit of psycholinguistics. And sadly, there are few definite answers, only hints. But of one thing we can be certain: at the root of each word is a meaning, and recovering this meaning is precisely what a dictionary is for.
Of course, we knew all along that words convey meaning, and that the mental lexicon is a store of word meanings. But so is the OED, and yet physically they could hardly be more different. But what, if any, are the consequences of the physical differences? The fact that they evidently are different does not mean that they are necessarily used any differently-for instance, the OED in book form could hardly be more different from the OED on CD-ROM, and yet there are aspects of their use which are common to both of them. Apart from the fact that the OED and the mental lexicon are physically different, what else is different about them?
Accessing our dictionary
We already know that not all dictionaries are the same, and that depending on which dictionary we use, we can access the words (and narrow down the search) on the basis of how they are spelled, how they are pronounced, what they rhyme with, what they look like, how long they are (as in a crossword dictionary), or even how frequently they occur in the language at large (as in some specialized dictionaries used by psycholinguists). But crucially, however we do it (see Chapter 5 for some discussion
of this), it is an inevitable consequence of accessing the dictionary that we will encounter, during the search, other words that share certain features with the word we are ultimately interested in finding, whether they share their spelling, pronunciation, rhyme, shape, length, or frequency. It is in this respect that our intuitions about what we do with a written dictionary are quite at odds with what we actually do with our own mental lexicon. For instance, although we do not burden our minds with the definitions of the other words that we pass as we scan down the page of a written dictionary, the same is not true of the process by which we access the mental lexicon. We do burden our minds with the contents of the neighbouring words we encounter as we narrow down the search. Our intuition that we do not is wrong, and our expectation on the basis of what appears to be a similar process (using the OED) is also wrong. The challenge, of course, is to prove that these intuitions are wrong.
It seems somewhat unreasonable to access the meanings of the words `ram' and `ramp' simply because they are encountered during the search for `rampart'. It would be equally unreasonable for an Australian listening to the word `acoustic' to access the entry for `acubra' (a traditional Australian hat) just because they start off sounding the same, or for a naturalist to access the meaning of the word `pichiciago' (a kind of armadillo) just because it starts off like `pitch'. It would surely make sense only to look up the definition of the word being looked for, as we do with written dictionaries, and not to look up the definitions of all the other words that just happen to overlap in their first few sounds. So why, when searching the mental lexicon, do we access the meanings of neighbouring words? And how can we, as psycholinguists, be so sure that this happens? As we shall see, it is unclear how things could possibly happen any other way.
During the 1980s, William Marslen-Wilson demonstrated that we can recognize a word even while it is still being heard (before, even, the speaker has finished saying it). We therefore access the lexical entry of a word well before the corresponding physical stimulation has ceased (that is, before its acoustic offset). In one of the first demonstrations of this, people were asked to repeat aloud as quickly as they could what they heard over headphones (to shadow what they were listening to). MarslenWilson found that often they would start to vocalize a word before it had finished playing on the tape. This was not simply some blind repetition of the sounds they heard, because if the words were jumbled up so that they made no sense ('up words jumbled he they so no sense the made'), people could no longer shadow as fast-so they were clearly interpreting what they were listening to, and were therefore recognizing the individual words before repeating them. In other experiments, he asked people to press a button as soon as they heard a particular word on a tape (word-monitoring, similar in spirit to the syllable-monitoring task mentioned in Chapter 5). He found that once you took into account the time it takes to decide to press a button, and the time it takes to press it, people were responding so fast that they must have been initiating their response well before the end of the word.
Marslen-Wilson found that the time it takes to recognize a word correlates very well with how much of the word has to be heard before it becomes uniquely distinguishable from all the other words in the language that share the same beginning. So `slander' becomes uniquely distinguishable only when the /d/ is encountered. Before then, the input would be compatible with `slant'.
An important component of the account of lexical access developed by Marslen-Wilson is that the entries in the mental lexicon are not simply accessed, they are activated. The idea that information is activated has a long established history in psychology, although its application to word recognition became more widespread in the late 1960s following the work of John Morton, now Director of the Medical Research Council's Child Development Unit in London. One way to think about this is to remember that ultimately, all the information in the mental lexicon is stored within the neural structures of the brain. When a pattern of light enters the eyes, or a sequence of sounds enters the ears, those stimuli do not access anything within the brain, even if they result in the recognition of, for instance, a politician speaking or a baby babbling (or both, if they are indistinguishable). Instead, the stimulation passes through the neural circuitry of the brain, being modified by, and in turn, stimulating (or activating) different parts of the circuit. Only certain kinds of stimulus will provide the right kind of stimulation for some particular part of the neural circuit-the stimulus is a key that can activate a part of the circuit, and depending on which part is activated, we experience `seeing a politician' or `hearing a baby'. There will be more of this later, but for now, the important point is that nothing is accessed; it is activated. And although we might just as well continue to refer to lexical entries, we shall return later to the idea that the mental lexicon is in fact a collection of highly complex neural circuits.
So what has this to do with why we access/activate the meaning of anything but the intended word? Why does this suddenly make it reasonable to suppose that we start to activate words and their meanings even before they become uniquely distinguishable from their neighbours? The answer has to do with the quite reasonable assumption that sounds entering the auditory system (i.e. the ear and beyond) stimulate the neural circuitry as they enter the system-a sequence of sounds is much like the combination to a safe; the tumblers in a combination lock fall into place as the correct sequence of rotations is performed, without waiting until the sequence is complete. Similarly, those neural circuits which require a particular sequence of sounds (before a particular word is `experienced') will become activated as that sequence enters the system. So the neural circuits that encode what we think of as lexical entries could quite reasonably become activated on the basis of a developing (but not yet completely developed) sequence of sounds- /slan/ would activate the neural circuits associated with (and hence would activate the meanings of) both `slander' and `slant'. But so much for what is possible in principle. What actually happens? Where is the proof?
What we need is a way of establishing which meanings of a word have been activated, and when. The priming task (first mentioned in Chapter 1) does just this. The task here is to decide whether a word that has just appeared on a computer screen is a real word in their language (e.g. `broom'), or a nonword (e.g. `broom'). How long it takes people to make a response (a lexical decision response) depends on all sorts of things. Nonwords that are similar to real words take longer to say `no' to than nonwords that are very different, and real words that are used infrequently take longer to say `yes' to than words that are used frequently. But the recognition of a real word can also be faster if a related word has been seen beforehand-lexical decision times to `broom' are faster following `witch' than following the unrelated control word `pitch'. This effect is called priming; `witch' (the prime) can prime `broom' (the target), `doctor' can prime `nurse', `bug' can prime `ant', and so on. Activating the prime causes the target to be activated faster. Conversely, if a target word is activated faster (primed), you can be sure that the priming word must have been activated.
In the mid-1980s, a student of William Marslen-Wilson's, Pienie Zwitserlood, used a version of the priming task called cross-modal priming to explore when, during the sound sequence, words are activated. In cross-modal priming, the priming word is presented in the auditory modality, and the target is presented visually. Zwitserlood and MarslenWilson reasoned that if lexical entries are activated before the end of a word, and if this activation is all it takes to get priming to related words, it should be possible to find cross-modal priming effects when the visual target word is presented on the screen part way through the auditory presentation of the priming word. So people would hear only the first part of a word, and at the end of that part, a related word (or unrelated control word) would be flashed up on the screen. The actual experiment was performed in Dutch in The Netherlands, but it translates very easily into English.
Zwitserlood used the Dutch equivalent of `captain', and played people a recording
of this word up to and including the /t/. At this point, the sound stopped, and a related word (e.g. `ship') appeared on the screen. Sure enough, she found priming-the word `ship' was responded to faster than the corresponding control word. Of course, this simply shows that the lexical entry for `captain' can be activated before the entire word has been heard. But the clever thing about this experiment was that `captain' was not the only word compatible with the fragment played to people; `captive' is just as good a continuation (the words can only be discriminated between on the basis of the final phoneme). And crucially, Zwitserlood also found priming to words related to these alternative continuations. In other words, the two alternatives that were compatible with the auditory input were both activated. And just to really prove the point, Zwitserlood demonstrated that if the visual targets were flashed up on the screen at the end of (and not part way through) `captain' or `captive', then only the related target words were primed-there would be no priming to words related to `captive' when presented at the end of `captain'.
So it looks pretty cut-and-dry; as the acoustic input enters the system, we activate all the lexical entries compatible with the input that we have heard so far. This is exactly what Marslen-Wilson's theory had predicted. And as the input becomes incompatible with certain alternatives, so those alternative entries begin to de-activate. But there was a further aspect to Zwitserlood's experiments that was important. Recall that one of the determinants of response times in lexical decision experiments is the frequency of occurrence of that word in the language at large; the more common words are responded to faster than the less common words. And although the more common words tend also to be the shorter words, it has been shown that this is not simply a length effect-once length is held constant, it is still the case that more frequent words appear to be recognized faster than less frequent words. The priming words used by Zwitserlood ('captain' and `captive' being just one pair) did not have the same frequencies; one member of the pair was always more frequent. And what Zwitserlood observed was that there was generally more priming from the more frequent word than from the less frequent word. It is as if the lexical entries corresponding to the more frequent words become more strongly activated on the basis of similar acoustic input than their less frequent neighbours. Again, this had been predicted by the theory.