Greg Detre
Tuesday, 21 November, 2000
for tute on Wed @12
Evaluate the downfall of the Derivational Theory of
Complexity. Did it really fail?
Is it appropriate to characterise syntactic knowledge as
involving a separate processing module?
Is word recognition achieved before the syntactic
structure of an utterance is computed?
How do we recognise words in sentences?
duality?
Words are not sufficient for language � their meaning is altered and given huge extra dimension by being bound in a syntax
But only if this knowledge is shared by speaker + listener = grammar
Why don�t we notice ambiguities? �time flies like an arrow� could mean any of over 50 different possible meanings (all grammatical), more if we are listening to the words spoken aloud one by one. it is because we rely on contextual clues and expectations of what the speaker is probably saying that we unconsciously filter all these out
Grammar is simply a set of conventions, adhered to within a given language community
Word order matters
politicians think that the public don�t know
the public know that politicians don�t think
Having said that, arranging the same letters of a word in a different order alters the meaning � what�s special about grammar? the difference is that grammatical correctness is determined by infinitely generative rules rather than rote-learnt set, finite combinations stored in a mental lexicon. the meaning of a sentence is constituted by the particular combination of its parts.
The gross constituents of a word, the phonemes, taken together simply identify the word. the constituents of a sentence, the words, all have a different function � they each play a different role, e.g. subject, object, verb etc. these roles inter-relate. different languages use different rules for identifying the roles each word is playing, e.g. English uses the �subject-verb-object� word order; English uses the �noun-adjective� order, which is reversed in French
Other languages use case-marking rather than word order to indicate the roles that words are occupying.
But often, a given sentence (especially as you hear it word-by-word) might be conforming to more than one grammatical convention, or some of the words may have more than one meaning � in such (frequent) cases, the sentence will be ambiguous, i.e. there is more than one grammatical interpretation of the sentence. For example, �time flies like an arrow�, could be interpreted as time flying through the arrow like an arrow, or time flies, a variant on the common house fly, having a penchant for arrows. Indeed, to the listener, the problem of distinguishing words from the auditory jumble is made more difficult since there may well be a person called �Anne Arrow�, or a rare verb, �to flizelike�.
Sometimes, the only way to pin down the roles being played in a sentence is by reference to the words� meanings. For instance, one can eat a pizza with friends/family or with your fingers/fork or with extra anchovies. The three senses are very different, and this is only signalled by their very different meanings.
But sometimes, one meaning being more plausible than the other is not enough???
Sometimes though, the ambiguity passes unnoticed (at least initially), with unexpected results:
For those of you who have children and don�t know it, we have a nursery downstairs.
Crowds rushing to see Pope trample six to death
And sometimes, it can prove very difficult to spot an ambiguity, to the extent that we might decide that a sentence is actually ungrammatical:
If someone read this sentence thought it was ungrammatical because it missed an �and� between �sentence� and �thought� they would be wrong.
The key to understanding this is to realise that there is someone who is being read the sentence, i.e. �read� is in the passive rather than being the main verb.
The fact that this mistake is made almost universally, despite there being a correct grammatical interpretation, would seem to indicate that there is a consistent preference underlying our grammatical analysis that rules out the alternative to the extent that our first response is to assume the sentence is ungrammatical. This mistaken reading need not hinge on a word with two possible meanings, like �read�.
This universal preference for one (incorrect) reading over another may be invaluable in understanding each other as well as we do without becoming bogged down in the possible ambiguities at every point. However, theories of language understanding get engtangled in explaining phenomena like this.
It could be that we learn to prefer certain conventions over others
Given that the less preferred interpretations tend to require more complex descriptions, they may require more complex, and so less obvious, mental processes too. We might automatically plump for the simplest option
We might be opting for the most frequently used constructions (which may also be the simplest)
Such a frequency-based choice might depend on the frequency of a particular construction for the words being used, e.g. �read� in the past tense is more frequently used than �read� as a passive
Yet, counter-intuitively, none of these explanations depend on the meaning of the word.
Context plays a critical role in our understanding. Consequently, it needs to be taken into account during experiments. Until Stephen Crain�s work with relative clause comprehension and production in young children, it was assumed that they were unable to differentiate between the elephant bumping the giraffe and vice versa in sentences like:
The elephant that a giraffe bumped against lay down and went to sleep
The elephant that bumped against a giraffe lay down and went to sleep
When Crain tried to get children to act out a similar story with toy animals, he found that children were able to understand and use such relative clauses, but only if there was more than one elephant. Taken in this context, the roles of subject and object are being much more clearly signalled by the relative clause, and the meaning was clearer to the child.
Various experiments in adult sentence processing were designed to investigate whether some of the preferences described would be affected by context.
Subjects� unconscious eye movements were measured with an infra-red beam while reading for the first time:
Sam told the writer that he couldn�t understand to get some help from a decent editor
Our eyes spend more time on problematic regions, or return to earlier parts of the sentence. The experiment was designed to test how problematic the sentence was by analysing eye movements when reading �to get�.
In order to have a control to see how our eye movements cope with a less ambiguous sentence, separate measurements were conducted on eye movements reading:
Sam asked the writer that he couldn�t understand to get some help from a decent editor
Here, �asked� replaces �told�, since it does not lead us to think that �Sam told the writer [something of great importance]�. For instance, we don�t ever say, �Sam asked the writer that [his new book was awful]�. As might be expected, eye movements in the sentence with �asked�, where there is only one possible interpretation, are different.
But if the ambiguous �told� case is embedded in a context where there is more than one writer, and it is that particular writer that Sam is addressing, then the problems on encountering �to get� are eradicated, mirroring Crain�s experiment with children.
Thus, it seemed as though certain kinds of ambiguities resolved themselves when the meaning could be distinguished on the basis of context. Subsequent experiments have given frequency a place in the explanation once more, alongside context and a variety of other factors. Experiments have shown, for instance, that different verbs associated with alternative structures but with different frequencies influenced the particular interpretation chosen.
frequency of occurrence of the different meanings assoc with each word???
We find then that if one kind of information is missing, e.g. context, we rely on other types, e.g. frequency of occurrence.
But which takes precedence over which? surely, in a connectionist model, it�s all probabilistic??? constraint satisfaction
Spoken vs written. Written contains no ambiguity about what individual words are, i.e. their spelling, allows you to easily distinguish between similar sounding words. However, spoken gives you prosody (intonation, rhythm, pausing etc.), which helps disambiguate the sentence greatly. However, it doesn�t always help.
Does emphasising a word disambiguate a sentence, or does it add alternative meanings? Possibly, it can do either. And does prosody analysis come after or before all these other factors?
What sort of mental representations do we form? Do they mirror the kinds of categories that linguists use? The answer affects the question of which mental representations are more complex, which we said earlier might affect our preference for a given reading over another. This would require a mental representation divorced from the individual meanings normally conveyed.
Heavy debate about just what form such internal representations of the �who-did-what-to-whom?� take.
Choosing between ambiguities is just part of the battle of understanding a sentence by deciding which roles should be assigned to which participant words.
Words have meanings in the same way that objects around us have meanings � we use them according to how they work and what they are for. Driving a car, as an analogy, is not just a matter of knowing what its components do, but also using them according to conventions to tell us which side of the road to drive on, what to do at a red light etc.
what about how to use the clutch and accelerator together - are they part of the conventions too???
words refer to things � sometimes they refer to other words (e.g. pronouns), or they point to which of a number of particular objects we mean (relative clauses)
some psycholinguists talk to the universal (except Frisian, and then only when no ambiguity would arise) distinction between reflexive and non-reflexive pronouns as evidence of innate linguistic knowledge, although children do take a long time before they learn to get them right consistently
perhaps it just stems from children�s early instinct not to assign two words for the same meaning
pronouns in questions � questions are like statements with gaps, and we answer the question by figuring out which role would be in the gap
how do they tell whether the subject reading the sentence thinks a word is implausible (Tanenhaus)?
see the words come up individually on the screen every time you press the button � the more implausible, the longer it takes to read the word
Which woman did Bertie present
a wedding ring to ___?
Which horse did Bertie present a wedding ring to ___?
Or Marta Kaus found that an identifiable pattern of brain activity can be seen on an EEG when the subject comes across a contextually implausible word
In fact, we process the words as we go along, and we assume after �which woman did Bertie present� that the woman is being presented
So what does this show??? maybe that �present� is used more commonly to present that thing to something, rather than to present something to that thing
We try to assign words to roles as we hear/read them, rather than waiting until we have found the �gap� in the question, which may only come at the end � this is why we sometimes have to re-read a sentence, to establish new roles for words in the light of what comes at the end
Unlike the him/himself universal clean distinction, different languages have different rules about where you can form a question by adding a gap. for instance, in German, one cannot have a gap after a proposition, though this is often permissible in English. However, even when these conventions are flouted, the sentence is easily understandable.
These seemingly arbitrary constraints on what can and cannot be done may reflect underlying, non-arbitrary conventions. Here is one example of how this might work for English: usually, to interpret questions like, �Which woman did Bertie marry ___?�, you try to identify which participant is intended by establishing first what role it played. It should be possible to see which word best fits the roles entailed by the verb. Finding the gap in the sentence later will tell you which role should have come there, and this should be enough to wholly disambiguate the roles of the entire sentence.
This may help when we now look at examples where adding a legitimate-seeming gap results in an ungrammatical question:
What did she think Bertie�s
offer of ___ was long overdue?
The marriage that she thought Bertie�s offer was long overdue took place last week.
In both these cases, the gap is in a position which does not receive a role from the verb. The arbitrary-seeming rule can be seen in the light of an underlying, wider rule, regarding how the roles are assigned by the verb.
Tension between linguistics and psycholinguistics � overlapping remits but differing agendas. Linguistics focuses on the grammatical, while psycholinguistics focuses on the processable.
There are limits to the processing capacity of the human sentence processor. For instance, almost everybody stumbles over double center-embedded relative clauses like:
The rat the cat the dog worried chased ate the malt.
Even though this is wholly grammatical, we cannot seem to hold the sentence structure complete in our heads. If we simplify, by naming �the cat the dog worried�, �Socks�, then we find:
������ The rat Socks chased ate the malt
comparatively easy to process. The complexity of this sentence can be seen in its very deep tree structure.
The processing difficulty of the original sentence is exacerbated by its containing two �empty categories�. These are seen by linguists as sections of the sentence that effectively exist, but are not pronounced or written. The listener has to deduce the existence of and properties of an empty category. An illustrative example is the sentence:
John flew to Paris, and Mary to Chicago
which contains the �missing� verb, �flew�, in the second half of the sentence. This is an empty category, because the tree diagrams for both halves of the sentence are identical, with an empty category completing the verb phrase.