Default vs. frequent; Other kinds of evidence
Mark McConville
Henry S. Thompson
9 March 2012
1. Regularity: Majority or defaultness?
There are two other important ways in which English plural morphology is similar to English past tense morphology.
Firstly, there are many more nouns that take the regular plural suffix than form irregular plural forms.
- Around 100 irregular plural nouns
- But thousands of regular ones.
Secondly, the regular plural suffix is the
default way of forming the plural, used when an
irregular plural form cannot be found, for whatever reason:
- new words - "wugs", "faxes",
- quotations - "I found three 'man's on page 1" (not "three 'men'")
- names - "We are having the Childs over for dinner - Julia and her husband" (not "the Children")
- foreign words - "mongooses" (not "mongeese"), "paninis", "cappucinos", "talismans" (not "talismen")
- acronyms - "RBIs", "MBAs", "SOBs"
- other rootless/headless compounds - "lowlifes", "still-lifes", "Mickey Mouses", "sabretooths", "bigfoots", "Walkmans", "Mack the Knifes", "the Maple Leafs".
2. Default vs. frequent, cont'd
So, there are two distinct ways of understanding morphological regularity:
- majority regularity - the regular pattern is the most frequently occurring
- psychological regularity - the regular pattern is the psycholinguistic default
These two different interpretations coincide in English, with both past tense and plural inflection
- there are vastly more regular verbs than irregular verbs
- there are vastly more regular nouns than irregular ones
This "coincidence" is why it is possible to train R&M-style
pattern associators to mimic English inflectional
morphology
- And also why they can be made to mimic the U-shaped curve of infant language acquisition.
What evidence can we find from other languages?
Are there languages where the psycholinguistically default plural is not in the majority?
3. German past tense inflection
German past tense morphology is similar to English
- Regular verbs take the dental "-te" suffix
- Irregular verbs use semi-systematic stem vowel changes (e.g. singen/sang, lesen/las)
But the relative proportions of regular versus irregular verbs differ.
German has more irregular verbs than English
- English: 86% of the 1000 commonest verbs are regular
- German: only 45%.
So, based on one counting method at least, German regular verbs appear to be in a minority.
However, the regular past tense suffix is clearly the psycholinguistic default:
- novel verbs - faben/fabte
- rare verbs - löten/lötete
- unusual sounding verbs - rilken/rilkete
- headless/rootless compounds - frühstücken/frühstückte, haushalten/haushaltete
- etc.
4. German plural morphology
German has eight different ways of forming plural nouns
- No suffix: e.g. Daumen/Daumen (thumbs)
- Umlaut + no suffix: Mutter/Mütter (mothers)
- -e: Hund/Hunde (dogs)
- Umlaut + -e: Kuh/Kühe (cows)
- -er: Kind/Kinder (children)
- Umlaut + -er: Wald/Wälder (forests)
- -en: Strasse/Strassen (streets)
- -s: e.g. Auto/Autos (cars)
The first seven classes are reckoned by German linguists to be
completely irregular.
- They have to be memorised independently in the lexicon.
- Attempts to formulate rules end up with large numbers of exceptions.
- Any semi-regularities (e.g. phonological or gender-based) are an artifact of associative memory.
The eighth class (-s) is the least frequent
- Only 1% of the 200 most common nouns form their plural using -s.
- Only 4% of nouns in the largest German dictionary form their plural using -s.
But -s appears to be the default plural ending, just like -s in English:
- Foreign loanwords: "Cafés", "Kiosks"
- Names: "Die Manns, Thomas und seine Frau" (not "die Männer")
- Headless compounds: "Tunichtguts" ("do-no-goods", i.e. "ne'er-do-wells")
- Abbreviations and truncations: "GmBHs", "Nazis"
- etc.
German linguists have called -s the "Notpluralendung" - the emergency plural.
Thus, the German -s plural is a textbook example of the distinction between the two senses of morphological regularity
- It is the psycholinguistically default option.
- But is extremely infrequent, compared to all the other plural suffixes.
Since the -s plural is so infrequent, pattern associators
cannot learn to mimic its default behaviour.
Pinker takes this as evidence that regularity cannot be reduced
to a pattern stamped on children's minds by repeated exposure to
words.
- And that symbolic rules are a necessary aspect of cognition.
5. Correlation or causation?
Connectionist explanations of regular morphology may well have got things back-to-front.
The connectionist argument runs as follows:
- Most English verbs form the past tense with the -ed suffix.
- Therefore, the -ed suffix is the default past tense marker.
But the evidence from German suggest the opposite direction of causation:
- The -ed suffix is the default past tense marker in English.
- Middle English imported lots of foreign verbs from Norman French.
- Therefore, most English verbs nowadays form the past tense with the -ed suffix.
The reason that German has fewer regular verbs (and regular
plurals) than English is simply that Germany was never conquered
by a French speaking ruling aristocracy.
6. The causal arrow, cont'd
Similarly, the following argument is a common misconception:
- Old English had many more irregular verbs than Modern English does.
- Therefore, languages always evolve from irregular to regular.
But we have already seen many examples of regular verbs becoming irregular:
- ringed ==> rang
- hitted ==> hit, quitted ==> quit etc.
- maked ==> made, haved ==> had
- goed ==> wended ==> went
So, languages don't consistently evolve in either direction
- Different psycholinguistic (and sociolinguistic)
processes constantly create and destroy both regular and
irregular inflection.
Same is true more widely about language change
- Languages change, but they aren't headed anywhere
7. Data from other languages
Dutch (another West-Germanic language):
- Distinction between regular and irregular past tense inflection (just like in English).
- The regular past tense suffix is the psycholinguistic default.
- Two distinct regular plural suffixes (in addition to a small number of irregular plurals):
- -s: nouns ending in an unstressed vowel, [r], [l] or [n]
- -en: other nouns.
- Each plural suffix behaves as the psycholinguistic
default for its phonological domain.
French (a Romance language - still Indo-European):
- The regular plural inflection is a "silent" '-s'
- e.g. chaise [ʃɛz] ("chair") => chaises [ʃɛz] ("chairs")
- There are a few irregular plurals too:
- e.g. nouns ending in -al often form a plural in -aux
- e.g. animal [ani'mal] => animaux [ani'mo:]
- The regular silent -s plural is the psycholinguistic
default, even for nouns ending in -al.
8. More languages
Hungarian (non Indo-European):
- Like German, has a number of distinct ways of forming plurals.
- But one suffix in particular, -ok, appears to be the psycholinguistic default
- It is used with names, e.g. "Aranyok" - the Arany family
- It is used with loanwords, e.g. "telefonok" - telephones
Arabic (Semitic - non-European):
- Two ways of forming plural nouns.
- "Broken" plurals (semi-systematic): kitabun (book) => kutubun (books), madrasatun (school) => madarisu (schools)
- The "sound" plural - add "-uun" to masculine nouns and
"-aat" to feminine nouns. This is rare for core Arabic nouns
but appears to be the psycholinguistic default.
Hebrew (also Semitic):
- Regular plurals add "-im" to masculine nouns and "-ot" to feminine nouns.
- Around 200 irregular nouns do exactly the opposite.
- But the regular suffixes are clearly the psycholinguistic defaults.
9. Still more languages; Conclusion
Chinese (Sino-Tibetan - isolating i.e. no inflection):
- All nouns need a classifier when used with a numeral
- cf. English - "three head of cattle", "four pieces of fruit", "ten sheets of paper"
- Different classifiers are used for differents kinds of object:
- "zhi" - long thin cylinders (e.g. pens)
- "qun" - animals
- "tiao" - long, flexible objects (e.g. fish, strips of paper) or news items
- But there is one special classifier "ge"
- Used as the default case, when speakers cannot retrieve
an appropriate classifier from memory.
- So: even a language with no inflection can manifest
psycholinguistically default behaviour.
Arapesh (New Guinean):
- Thirteen genders
- One gender is the default - used when the gender of the noun cannot be determined
- e.g. in conjunctions - If X is masculine and Y is
feminine, then what is the gender of the conjoined noun "X
and Y"?
- e.g. sex-neutral nouns, like "friend" or "grandparent"
Moral: all languages appear to have the characteristic signs
of rules - the existence of default words or
morphemes to fall back on when memory lets speakers down.
So the dual mechanism account is strongly supported
- At least at some level of explanation
10. Cognitive neuroscience
Psychology has traditionally been the study of the human mind as a black box
- We try to figure out what is going on inside by examining the inputs and outputs
- What does the whole person do when presented with certain words, pictures or instructions?
New techniques allow neuropsychologists to study the living brain without invading it
- Computerised Axial Tomography (CAT scans)
- Positron Emission Tomography (PET scans)
- Functional Magnetic Resonance Imaging (fMRI scans)
These techniques allow us to study living brains in action, including injured brains.
DNA testing allows us to pinpoint genes responsible for inherited psychological conditions.
Together, these advances have led to a new field of study: cognitive neuroscience.
Can we isolate words and rules in different parts of the brain?
Can we learn anything from cases of direct neurological damage?
11. Brain as computational system
Can we identify a bit of brain devoted to words and a bit devoted to rules?
Unfortunately, the brain is not the kind of organ in which
every function needs to be carried out by a chunk of tissue with
a recognisable shape.
The brain is a computational system
- What is important is how information flows within it
- Not how the system takes up space.
Also: we cannot force experimental subjects to just relax and think pure past tense thoughts
- We need to study them performing particular tasks
- And these tasks will involve lots of other mental modules
- So we cannot expect to see putative rule circuits glow in an otherwise inactive brain.
So to prove that regular and irregular verbs rely on different
parts of the brain
- we need to look for more indirect clues about
their behaviour
12. Human brain anatomy
The human brain consists of billions of neurons, connected by trillions of synapses.
These are all organised into dozens of lumps, sheets and strands in a convuluted 3D structure.
The brain has two hemispheres
- Language/grammar is typically in the left hemisphere
- At least as far as right-handed people are concerned.
Each hemisphere is divided into parts by the central sulcus and the Sylvian fissure
- the frontal lobe
- the parietal lobe
- the occipital lobe
- the temporal lobe
![[no description, sorry]](../1/BrainAnatomy.jpg)
The major language areas are on the banks of the Sylvian fissure
- Broca's area -
- planning of speech
- verbal short-term memory
- comprehension of complex sentences
- Wernicke's area -
- connecting the sounds of words with their meanings
- different parts appear to encode different semantic fields (e.g. colours, animals, tools).
So, at a very rough level we can postulate:
- in front of Sylvian fissure = rules
- behind Sylvian fissure = words.
13. Double dissociation
The supersized human brain is a vulnerable organ, which can be
damaged by a variety of causes - tumours, infection,
malnutrition, blocked/burst arteries, falls, bullets, car
accidents.
Brain-damaged patients are a crucial source of evidence for
cognitive scientists.
Imagine that a patient with an injury to brain part X can no longer do feat Y but can still do feat Z
- It is tempting to assume that X is the neural centre responsible for Y.
But this is too simple
- It may just be that Y is an intrinsically harder task than Z, and the weakened brain stumbles on it.
Neuropsychologist Hans-Lukas Teuber pointed out that links
between mind and brain should be based on a double
dissociation, involving two kinds of patient and two kinds
of task
- Patients with damage to brain part X but not to part Y
who cannot do feat Z but can still do feat W
- And: patients with damage to brain part Y but not to
brain part X who cannot do feat W but can still do feat
Z.
This would suggest that X and Y differ in the kind of work
that they do, and that the kind of work done by X involves task Z
whereas the kind of work done by Y involves task W.
- As in the visual system double dissociation we saw early in the course
14. Double dissociation and spelling rules
Spelling rules give us a nice example of double dissociation involving language.
Consider the following two conditions:
- surface dyslexia - inability to pronounce irregularly
spelled words such as "yacht" and "aisle" (i.e. [jatʃt] and [azəl])
- phonological dyslexia - inability to pronounce regularly
spelled non-words such as "wug" and "dax"
Some brain-damage patients have surface dyslexia but not phonological dyslexia.
Other brain-damage patients have phonological dyslexia but not surface dyslexia.
So we can conclude that the brain contains two distinct routes from printed text to sound:
- memorised words and their pronunciations
- default letter-to-sound rules
Each of these routes can be damaged independently of the other.
15. Double dissociation and connectionist models
Connectionist models such as R&M's try to capture regular and
irregular forms in a single pattern associator memory.
- Regular and irregular associations are smeared across a single set of connection weights.
- This eliminates the need for separate modules for rules and exceptions.
This approach makes it impossible to simulate double dissociation.
Connectionist modellers generally simulate damage by eliminating/weakening connections at random.
This leads to a single dissociation involving inability to handle irregular words
- since irregularities depend on a small number of particularly strong connections
- but regularities depend on a diffuse set of weaker connections.
Claims that connectionist models have been able to exhibit
double dissociation are only valid for artificially small toy
models
- If regular connections are spread over a very small
number of connections, then the chance that they can be damaged
randomly is higher.
16. Agrammatism
Aphasia is an impairment of language following an injury to the brain
- and hence an important source of information about the
organisation of linguistic knowledge in the brain.
Agrammatism is a symptom of some forms of aphasia, where patients have difficulty
- Forming words into phrases and sentences
- Putting the right grammatical suffixes onto stems
- Understanding complex sentences.
Agrammatic aphasics have less trouble retrieving individual
words and idioms.
Examples of agrammatic speech:
- Son ... university ... smart ... boy ... good ... good.
- Lower Falls ... Maine ... paper ... four hundred tons a day!
Agrammatism is often caused by damage to Broca's area, and
surrounding regions of the frontal lobe.
17. Agrammatism and inflection
Agrammatics have particular problems with inflectional
suffixes either leaving them out or using the wrong one
- e.g. an agrammatic might read out "smiled" as "smile" or "smiling".
But studies have shown that such patients make
fewer such errors with irregular past tense and
plural forms than they do with regular ones
- This suggests that the damage has been done to the rules
involved in processing regular inflection, rather than to
memorised words
- And hence that words and rules are found in different
parts of the brain.
18. Agrammatism cont'd
Similar studies have discounted alternative explanations, e.g.
- That agrammatics have trouble pronouncing -ed or -s at the end of a word.
- That agrammatics stop reading from left to right as soon as they encounter a completed word.
- That agrammatics have more trouble with infrequent words than with frequent ones.
Studies involving language production among agrammatics have shown similar results
- They are better at producing irregular past tense and plural forms than regular ones.
- They cannot inflect novel words like "rick" or "wug".
- They never overregularise (e.g. "digged" or "foots").
19. Anomia
The other half of the double dissociation involves anomia
- difficulty in retrieving and recognising words despite fluent and generally grammatical speech.
Anomic patients generally have words stuck on the tip of their
tongue, and resort to circumlocution, pronouns, and generic words
like "something" or "stuff".
For example, here is an anomic patient trying to talk about his elbow:
- "That's the part of my body where, my hand and my
shoulders, no, that's not it. I just can't get it. Isn't that
terrible?"
And a clock:
- "I know that. It's the thing you use for counting, for
telling the time. You know, one of those ... It's a ... I just
can't think of it. Let me look in my notebook."
Anomia is often caused by damage to posterior
parts of the brain, in and around Wernicke's area.
20. Jargon aphasia
Some anomics exhibit jargon aphasia
- They speak in their own neologisms, e.g. "nose cone" instead of "phone call".
Interestingly, jargon aphasics often inflect their jargon with
regular suffixes ("a self-administered wug test"!) -
- "Waitrixes. A backland and another bank. For bandicks I
think they are. I believe they're zandicks. I'm sorry, but
they're flitters landocks."
- "She wikses a zen from me."
- "He mivs in a love-beautiful home."
Again, this suggests that regular inflection is computed in a
different part of the brain from the parts in which words are
stored and retrieved.
Studies have shown that anomic patients:
- Have more trouble inflecting irregular verbs than regular ones.
- Are relatively good at inflecting novel verbs like "rick" or "plam".
- Make overgeneralisation errors (e.g. "digged").
So, anomic patients appear to be the mirror image of
agrammatic patients.
21. Priming and double dissociation
Recall the priming effect in lexical decision tasks with
regular and irregular past tense forms
- both regular and irregular past tense forms priming their verb stem
- but regular past tense forms are more effective at
priming than are irregular past tense forms.
William Marslen-Wilson and Lorraine Tyler investigated priming
effects in agrammatic and anomic patients.
In two agrammatic patients
- regular past tense forms failed to prime the relevant verb stem
- but irregular past tense forms were effective at priming verb stems
- and semantically related words primed each other (e.g. "duck" and "goose")
And in one anomic patient
- irregular past tense forms failed to prime verb stems
- and semantically related words failed to prime each other
- but regular past tense forms were effective at priming verb stems.
Again we see a classic double dissociation, suggesting that
regular and irregular inflection are handled in different parts
of the brain.