Psycholinguistics
James Myers
April 9, 2004
The mental lexicon

OVERVIEW:

1. Lexical access
2. Morphological structure
3. Models of lexical semantics

=============================================================

1. Lexical access

	Psychologists study processes; thus the study of the lexicon is the 
	study of how people access (reach) it.

1.1 There are two kinds of access (perception/comprehension vs.
	production), so maybe there are two lexicons...?

	For example, Caplan (1993:120-1) points out that:

		There are brain-damaged patients who are normal in 
		comprehension tasks and normal in picture-naming 
		(production) tasks      -- BUT --
		they make SEMANTIC errors in repetition (both perception
		and production):

			"Eisenhower" (US president in 1950's) --> 
			"Krushchev" (USSR leader in 1950's)

		This suggests they have a normal comprehension lexicon and
		a normal production lexicon, but the link between them is
		damaged.  This in turn implies that there are two lexicons!
		(... or does it...?)

1.2 The effect of modality:  reading versus listening

	The lexicons for reading and listening are probably very different
	as well.  For example, when reading a single Chinese character,
	readers probably access it all at once, but when listening to a
	spoken syllable, listeners may instead access it in pieces, over
	the time they hear the syllable.

1.2.1  For example, Gaygen & Luce (1998) found that subjects were
	sensitive to whether a word was more familiar
	from reading or from listening.)

1.2.2	A specific theory about accessing spoken words:  the COHORT model.
	(e.g. Marslen-Wilson 1987):  as you hear a spoken word from
	beginning to end, you activate all words in memory that are
	consistent with what you've heard so far.  This "cohort" of
	competing words shrinks until only the correct one is left --
	which may occur before you have heard the word all the way to the
	end.  For example, if you are listening to the word "elephant",
	the cohorts are:

		HEAR:	el...		...e...		...ph...
		COHORT:	elk,		elevator,	elephant!
			Elvis,		elephant
			elevator,
			elephant...

	But the evidence for this mainly comes from off-line GATING
	experiments, where subjects are played pieces of words, e.g. "el",
	then "ele", etc, and they are asked to guess what the word is.  It
	is not at all surprising that they end up naming words in
	"cohorts", but does this really describe what happens in on-line
	word access?

1.3 Now let's discuss two important, well-established and
	experimentally useful effects regarding lexical access, and then a
	third point illustrating these two:

		Frequency effects
		Priming effects
		Is word access modular?

1.3.1 Frequency effects

	The lexical frequency of a word is how often it is encountered
	(heard, read, written, and/or spoken).

1.3.1.1	Frequency in Chinese is particularly tricky:

	  What counts as a "word" in Chinese?  Which of these examples are
	  single words, and which are two words combined in a phrase?

		(1)	¶}¤ô			(6)	®à¤l
		(2)	¶}¤M			(7)	¤ý¤l
		(3)	¶}Ãö			(8)	­ì¤l
		(4)	¶}©l			(9)	¦Ñ¤H
		(5)	¶}ªù			(10)	¦ÑÁó
 
	  Chinese also shows very large differences between frequencies
	  for written and spoken language (a problem, since most frequency
	  counts are calculated from written language).  For example, ©M is
	  the usual word for "and" in speech, while »P is a very common word
	  for "and" in writing.

1.3.1.2	In spite of such problems, people have pretty reliable intuitions 
	about the frequency of words.  Guess which of each pair of words 
	below is more common, and you'll probably be right (you can 
	check your answers on the web at the Academia Sinica corpus site:
	http://www.sinica.edu.tw/SinicaCorpus/).

		Å¥¡H  ÆU¡H
		Å¥¨ì¡H  ¬Ý¨ì¡H
		¤ë¡H  ¦~¡H

1.3.1.3	Frequency effects are similar to other perceptual judgments:
	they follow a logarithmic function (Rubenstein & Pollack, 1963).

	  At low levels, small differences in loudness, brightness, etc
	  are easily perceived; at high levels, small differences are harder
	  to perceive. 

	  Likewise, small differences in frequency are easier to perceive
	  for low-frequency words than for high frequency words.

1.3.1.4	Frequency effects in the NAMING task:

	  Subjects see a written word or a picture and must "name" it as
	  quickly as possible. 

	  Key measurement:  reaction time (RT), also called response time
	  or response latency.

	  RT's can be affected by other factors, of course, e.g. the
	  length of the word (high frequency "banana" vs. low frequency "id")

1.3.1.5	Frequency effects in the LEXICAL DECISION task:

	  Subjects are given a list containing both real words and fake words
	  (nonwords, pseudowords, nonce forms), presented either visually or
	  auditorily; subjects must decide as quickly as possible whether a
	  given item is a real word or not.

	  Subjects are faster to decide if a visually presented string of
	  characters is a real word or not if the real words are higher 
	  frequency (assuming that length, etc, is controlled).
	  [Note:  some have argued (e.g., Hung, Tzeng, & Ho, 1999)
	  that the fuzzy status of the Chinese "word" makes lexical
	  decision tasks questionable for Chinese.]

		English example:
		  Try the "experiment" described in Carroll, p. 120.
		  (It might not work if you don't read English a lot!) 

		Chinese example:
		  Time yourself for each list below.  Do you find a 
		  frequency effect?  (from data used in Ahrens 1998)

			Higher frequency words (mixed with fake words):

			Ū¥õ	¬Q¤Ñ	¤¤¤ß	±M·~	¹D¯S	«p´ú	¥Ë´µ
			¨ý¼M	¯Þ²~	µ¹ÁÚ	§ä¨ì	°ª¿³	ÆF¸T	³ø®×

			Lower frequency words (mixed with fake words):

			»¡²z	¬ü­Ì	Án±a	¤åªZ	®È¯u	Â÷¥@	¦Û¦h
			°Ê²Î	»´ÆF	ÂåªÌ	¹ê¥R	²§¦P	¥Ö½L	»G©É

			Notice, by the way, how hard it is to make "fake 
			words" in Chinese!  Many of the nonwords are in fact 
			impossible, e.g. ¬ü­Ì violates the morphology of 
			Chinese.  Moreover, Chinese subjects often aren't sure 
			what you mean by "word" anyway.  It's easier to tell 
			them to make a meaningful/nonmeaningful judgment.

1.3.1.6	Frequency is an arbitrary, unpredictable fact about a word, and so
	must be "stored" somehow in the lexicon.

	  Frequency can therefore be used as a diagnostic of lexicality:
	  if X shows frequency effects, then X must be stored in the lexicon.

	  Example:  Sereno and Jongman (1997)

		Question:  Do English speakers generate regular inflection
		by rule, or is regular inflection included in the lexicon?
		For example, is the word "dogs" in the lexicon, or only "dog",
		so that people have to use a rule to add the /s/ when
		speaking or writing, or subtract the /s/ when listening or
		reading?  (Connectionists say "no rules"; others, like
		Steven Pinker, say "yes rules".)

		Task:  Visual lexical decision task.

		Materials:  Two kinds of nouns were chosen:  nouns whose
		singular form is more frequent than the plural form
		(e.g. river-rivers); and nouns whose plural is more frequent
		than the singular (e.g. window-windows).

		Procedure:  Half of the subjects got a list of singular 
		nouns (plus an equal number of fake words).  The other half
		got a list of plural nouns (plus fake words).  

		Prediction:  If regular plurals are stored in the lexicon,
		then there should be a frequency effect.  Specifically,
		subjects in the SINGULAR condition should be faster with
		"river" than with "window", while subjects in the PLURAL
		condition should be faster with "windows" than with
		"rivers".

		Results:  The prediction was supported.  This implies that
		English users do NOT use a rule to add or subtract plural /s/.

1.3.1.7	Models of lexical access and frequency effects:

	  One obvious but silly model (Forster 1976):
		words are "listed" in memory from highest frequency to lowest,
		so it takes longer to search for lower frequency words.

	  ACTIVATION models (includes the old "logogen" model and current
		connectionist models):

		each time you encounter a word, the "activation" of its
		lexical entry is increased, and this activation helps you
		access it later; thus frequent words are easier to access.

	  Note that both models imply that frequency is an inherent part
	  of storing a word.  Thus they predict that no task will show a
	  lexical effect (e.g., real words are faster to respond to than
	  fake words) without also showing a frequency effect.  But this
	  isn't true.  Naming tasks using English often find that naming
	  real words is faster than naming fake words, but there is still
	  no frequency effect (see e.g. Paap et al. 1987).  However, this
	  doesn't seem to be true in Chinese, which implies that the
	  English effect is due to subjects using pronunciation rules
	  which can be used independently of the lexicon.

1.3.2 Priming effects

	PRIMING:  the presentation of one stimulus (the PRIME) affects the
	speed (usually speeding up) of the response to another stimulus
	(the TARGET).

	  "X primes Y" means "X is a prime for Y" or "the presentation of
	  X speeds up (or slows down) the response to Y relative to a
	  control"

1.3.2.1	Words prime themselves:  subjects are faster to name or make a
	lexical decision about a word if they've been presented with that
	same word recently.

	Words are primed by semantically related words:  subjects are
	faster to make a lexical decision about "BUTTER" if it's preceded
	by "bread" than if it's preceded by "nurse."

1.3.2.2	How to model priming:  By far the best models use a connectionist
	or spreading-activation approach (e.g. Carroll, p. 115 or p. 116):

	  This model claims that the lexicon consists of a network of
	  related words.  If a word is "activated", its activation may spread
	  to words that are linked to it:  SPREADING ACTIVATION.

	  Words prime themselves because repeated exposure to a word
	  increases its activation level; thus it will be easier to
	  respond to the next time.

	  Words are primed by semantically related words because of
	  spreading activation. 

1.3.2.3	Thus priming effects can be used to study the mental
	representation of words:  if you know what A is like, and you know
	that A primes B, then you know that B must be represented in the
	lexicon in a way similar to A. 

	However, only semantic priming is reliable and consistent. 
	Phonological priming and morphological priming are much more
	complex and are not well understood.  In fact, phonological
	"priming" can actually cause the response to the target to slow
	down ("inhibitory priming").  (Any guesses why this might be...?)

	Priming effects can be used to study the time-course of lexical access;
	e.g. in processing a sentence, what words get activated at what time?
	(More on this shortly.) 

1.3.3 Frequency and priming as tools in the study of modularity

	Words may be semantically ambiguous (have more than one meaning):
		"letter"  "bug"	 ¶º¸J  ¤¤À\

	Frequency effects:  common (primary) meanings are more rapidly
	accessed than uncommon (secondary) meanings. 

	  Hogaboam and Perfetti (1975):
		Subjects given sentences ending in semantically ambiguous
		  words, e.g. "letter".
		Preceding context consistent either with the primary
		  meaning (1) or with the secondary meaning (2):

			(1)  The jealous husband read the letter.
			(2)  The antique typewriter was missing a letter.

		Task:  Is the last word ambiguous?
		Results:  Subjects are faster with (2).
			In (1), context just reinforces primary meaning;
			  accessing secondary meaning takes longer.
			In (2), context reminds subjects of the secondary 
			  meaning; primary meaning is accessed quickly.

	Modularity (?) in the processing of lexical ambiguity, using
	frequency effects & priming as research tools:

	  Do people access all meanings of a word even if the preceding
	  context makes it clear that only one meaning is relevant?
	    (NOT: "Does the sentential context have NO effect?"
	    Obviously it does eventually -- but is the effect immediate and
	    automatic or not?) 

	  Method:  The CROSS-MODAL lexical decision task (also known as
		the cross-modal priming technique):  Swinney (1979)

		Sample auditory stimulus:

		"Rumor had it that, for years, the government building 
		has been plagued with problems.  The man was not surprised
		when he found several spiders, roaches, and other BUGS in the
		corner of this room."

		Sample visual stimuli for lexical decision:

			ANT		SPY		SEW

		Coordination between the two types of stimuli:

		SEE:.......................................ANT
		HEAR:   ...spiders, roaches, and other bugs ^ in 
		the corner...

		Results:  "bugs" primes both ANT [related to primary 
		  meaning] and SPY [related to secondary meaning], 
		  regardless of preceding sentential context.

		Conclusion:  English listeners automatically access all
		  meanings of an ambiguous word, independently of 
		  sentential context.

		Other experiments by Swinney show that the effect fades several
		  words later (e.g. in above example, after a few words
		  only ANT will be primed).

	  Ahrens (1998) did a Chinese version of this experiment:

                ¦pªG¦³¥~°ê«È¤H¨Ó³X¡A§Ú³q±`¤£©Û«Ý¥L­Ì¦Y¦è¦¡¶¼­¹¡A                
                ¦Ó·|½Ð¥L­Ì¦Y¡i¤¤À\¡j¡A³o¼Ë¥L­Ì¤~·|¦³·sAªº·Pı¡A
                ¨Ã¥B¯d¤U¤ñ¸û²`¨èÃø§Ñªº¸gÅç¡C
 
                HEAR:           ... ¦Ó·|½Ð¥L­Ì¦Y¤¤À\¡....
                SEE:    ("primary" condition)     ....¤È¶º
                        ("secondary" condition)   ....ªF¤è
                        ("control" condition)     ....¤åªk
                        ("nonword" condition)     ....¬ü­Ì
 
                Task:  is the visually presented item a µü or not?
 
                Here the prime is the auditory word¡i¤¤À\¡j, and the
                targets are the various visually presented words.
 
                Results:  same as for English:  both "primary" and
                "secondary" targets are primed relative to the control.

	This effect is theoretically important, since it implies that the
	processing of words and the processing of sentences are initially
	independent of each other.  Not everybody likes this "modularity"
	hypothesis, though, and it hasn't been supported by all
	researchers (e.g.  Tabossi and Zardon 1993, who studied Italian). 

2. Morphological structure

2.1	MORPHOLOGICALLY COMPLEX words:  words containing more than one
	morpheme (also called "polymorphemic" words). 

	  Examples:  banana-s, walk-ing, re-apply, un-happy,
		     anti-dis-establish-ment-arian-ism

		     µf­X¡B¥~°ê¤H¡B¥i·R¡B¤u·~¤Æ¡B¤pªB¤Í­Ì

	  Sometimes morphemes are more difficult to determine:

		re-sult?  sn-eeze?  (cf. snot, snore, sneer)
		®à¤l¡B½¹½º¡B©_²§ªG (from English "kiwi")

2.2	One of the key questions:  Does lexical access involve 
	MORPHOLOGICAL DECOMPOSITION:  breaking such words into 
	their morphemes?

2.3	Many experiments of many kinds address morphological processing.
	We already saw one above (Sereno and Jongman 1997).  Here are some
	more (some involve language production rather than language
	perception/comprehension).

	  MacKay (1978):  morphological processing in production?

		Subjects given verbs:		govern		decide
		Must derive nouns:		government	decision

		Results:
			"government" fastest (no phonological changes)
			"decision" slowest
				(two changes:  [aj] -> [I],  [d] -> [zh])

		Interpretation:  subjects really do make those changes 
		  when producing morphologically complex words;
		  therefore such words are not stored simply as wholes.

		Criticism:  subjects are explicitly told to derive 
		  morphologically complex nouns from verbs!
		  Perhaps people do not do this normally.

	  Taft (1981):  "prefix-stripping" in word recognition?

		Subjects given written words with prefixes:        REMIND
		 or with "pseudoprefixes" (i.e. not real prefix):  RELISH

		Lexical decision task.

		Results:  RT's are faster for prefixed words.

		Interpretation:  Morphological processor automatically 
		  "strips off" anything that looks like a prefix (e.g. "RE"),
		  then searches for the base in the lexicon.

		  With words like REMIND, it will find MIND (real word),
		  but with words like RELISH, will not find *LISH, thus 
		  wasting time.

		Criticism:  although subjects are not explicitly told to strip 
		off prefixes, maybe they are implicitly told this by the kind 
		of word list they get.  Rubin, Becker and Freeman (1979) 
		found that if the word set contains 50% prefixed forms, 
		subjects use a "prefix-stripping" strategy, but if there's only 
		10% prefixed forms (more realistic), subjects do not show 
		this effect.

	  Taft and Zhu (1995):
	  	Are morphemes relevant in the recognition of written Chinese 
		words?

		Materials:  "binding characters", i.e. characters that only 
		  appear in one word (e.g ³L°C ) -- because these are two-
		  syllable monomorphemic words; and "non-binding, 
		  position-specific characters", i.e. characters that appear in 
		  more than one word, but always in first position (e.g. ®î )
		  or always in second position (e.g. «Q ).

		Task:  Subjects saw individual characters on a computer 
		  screen, and had to "name" them (i.e. read them aloud).

		Results:  Subjects were faster to name first-position 
		  binding characters (e.g. ³L ) than second-position binding 
		  characters (e.g. °C ), but non-binding characters showed no 
		  position effect.

		Interpretation:  For monomorphemic "binding" words, 
		  subjects had to access the entire word in order to find the 
		  pronunciation of each character, which implies that such 
		  words are stored as whole units.  For polymorphemic non-
		  binding words, subjects accessed each character 
		  independently.

		Question:  Is this just a matter of orthography, or would 
		  monomorphemic, non-binding words (e.g. ¨Fµo ) behave 
		  differently?  That's for future research!

	  Zhou and Marslen-Wilson (1994):
	  	Are morphemes relevant in the recognition of spoken Chinese 
		words?

		Subjects were given 4 kinds of spoken two-morpheme words,
		all of which were semantically transparent (i.e. the 
		meaning of the whole was closely related to the meaning of 
		each part):

			high frequency word, high frequency first morpheme
			high frequency word, low frequency first morpheme
			low frequency word, high frequency first morpheme
			low frequency word, low frequency first morpheme

		Task:  lexical decision (thus also equal number of nonwords)

		Results:	strong effect of word frequency;
				no effect of morpheme frequency!

		Interpretation:  In strong contrast to results on written 
		  Chinese characters, spoken Chinese words are NOT 
		  morphologically decomposed in comprehension (at least not 
		  initially -- priming studies like Zhou and
		  Marslen-Wilson (1995) imply that morphemes within
		  compounds are actively processed even in spoken word
		  recognition).

3. Models of lexical semantics

3.1 Semantic feature theory (commonly assumed by linguists):

	  Claim:  A category is defined by the sharing of features.

		For example:

		  NOUNS:

			man	[MALE][ADULT]
			husband	[MALE][ADULT][MARRIED]
			bachelor [MALE][ADULT][NOT[MARRIED]]

			horse	[CONCRETE]  [ANIMAL] [MAMMAL] ...
			robin	[CONCRETE]  [ANIMAL]  [BIRD]  ...
			tree	[CONCRETE]  [PLANT]  ...
			dream	[ABSTRACT]  ...

		  VERBS:

			die	[BECOME] [[NOT] [ALIVE]]

			kill x
				[CAUSE] [[BECOME] [[NOT] [ALIVE (x)]]]

			murder x

				[CAUSE] [[BECOME] [[NOT] [ALIVE (x)]]]
				AND
				[PERSON (x)]

		Problem:  Not all psychologically real categories can be 
		  accurately defined with features!

			"fruit" = "edible reproductive body of a seed plant"
			  (thus includes "apple", "banana", "coconut",
			  "tomato", "green pepper", "chilli pepper", etc)

			But Americans do not consider "tomato", "green 
			  pepper", and "chilli pepper" to be fruit!  (instead,
			  the first two are "vegetables" and the third is
			  a "spice")

			Note cultural factor:  Chinese do consider "tomato"
			  to be a fruit! 

		Also, there are "good examples" of fruit, and "bad examples."
		  Thus "coconut" is technically a fruit, but it's not a good
		  example. 

3.2 Prototype theory (e.g. Rosch 1973):

	  Claim:  A category is defined by the sharing of "family 
	    resemblances", i.e. similarity to some other members in 
	    the category.  There need not be a single feature or set of 
	    features shared by ALL the members.

	  PROTOTYPE:  the "best" member of the category, as 
	    determined by people's judgments; it is also typically 
	    named by the highest frequency word in its category.

		Example:

		  Category "bird"
			in the USA, "robin" is prototypical.
			"eagle" is pretty good
			  (shares many similarities with "robin")
			"penguin" and "ostrich" are pretty bad
			"bat" is terrible (it's not even in the
			  category)

		  What is the prototypical bird for Taiwanese???

	  A typical prototype effect in a SEMANTIC VERIFICATION task:

		Subjects are presented with sentences like:

			(1)	"A robin is a bird."
			(2)	"A penguin is a bird."
			(3)	"A bat is a bird."
			(4)	"A horse is a bird."

		Task:  True or false?
		Results:  (1) is responded to more quickly than (2), 
		  because "robin" is a more prototypical exemplar of the 
		  "bird" category.

		Prototype effects on Chinese classifiers:  (Ahrens 1994)

			±i  : classifier for flat things
				prototypical example:  ¯È

			Subjects must name objects shown in pictures;
			  record choice of classifier, e.g. ±i¡B±ø¡B­Ó¡B°¦

			Results:  "correct" classifier is used less frequently
			  for less prototypical examples.

				Percent usage of ­Ó or °¦ instead of ±i

					¯È	§É	®à¤l	¨Fµo
					0%	30%	45%	65%

		Prototype effects can be modeled in connectionism by 
		  assuming there is a strong connection between a prototype 
		  node and a category node; e.g. ±i   is strongly linked to
		  ¯È , but only weakly linked to ¨Fµo .

		Logical problem with prototype theory:  This theory tries to 
		  respond to the problems with the "feature" approach to 
		  lexical semantics.  And yet prototype theory uses the 
		  concept of "similarity."  How is "similarity" defined,
		  except by using features????

REFERENCES

Ahrens, K. (1994).  Classifier production in normals and aphasics. 
	Journal of Chinese Linguistics, 22, 203-247.
Ahrens, K. (1998).  Lexical ambiguity resolution:  languages, tasks,
	and timing.  Syntax and Semantics, 31, 11-31.
Caplan, D. (1993).  Language:  Structure, processing and disorders.  
	Cambridge, MA: MIT Press.
Forster, K. I. (1976). Accessing the mental lexicon."  In R. J. Wales and
	E. Walker (Eds.)  New approaches to language mechanisms
	(pp. 257-287).  North-Holland.
Gaygen, Daniel E., & Luce, Paul A. (1998).  Effects of modality on
	subjective frequency estimates and processing of spoken and printed
	words.  Perception and Psychophysics, 60(3), 465-483.
Hogaboam, T. W., & Perfetti, C. A. (1975).  Lexical ambiguity and 
	sentence comprehension.  Journal of Verbal Learning and Verbal 
	Behavior, 14, 265-274.
Hung, D. L., Tzeng, O. J. L., & Ho, C.-Y. (1999). Word superiority effect
	in the visual processing of Chinese. In O. J. L. Tzeng (Ed.) Journal of
	Chinese Linguistics Monograph Series No. 13: The biological bases of
	language, 61-95.
MacKay, D. G. (1978).  Derivational rules and the internal lexicon.
	Journal of Verbal Learning and Verbal Behavior, 17, 61-71.
Marslen-Wilson, W. (1987).  Functional parallelism in spoken word
	recognition.  Cognition, 25, 71-102.
Paap, K. R., McDonald, J. E., Schvaneveldt, R. W., & Noel, R. W. (1987).  
	Frequency and pronounceability in visual presented naming and 
	lexical-decision tasks.  In M. Coltheart (Ed.) The psychology of 
	reading, vol. 12 of Attention and Performance.  Erlbaum.
Rosch, E. H. (1973).  On the internal structure of perceptual and semantic
	categories.  In T. E. Moore (Ed.) Cognitive development and the
	acquisition of language (pp. 111-144).  Academic Press.
Rubenstein, H., & Pollack, I. (1963). Word predictability and
	intelligibility.  Journal of Verbal Learning and Verbal Behavior, 2,
	147-158.
Rubin, G. S., Becker, C. A., & Freeman, R. H. (1979).  Morphological 
	structure and its effect on visual word recognition.  Journal of
	Verbal Learning and Verbal Behavior, 18, 757-767.
Sereno, J. A., & Jongman, A. (1997).  Processing of English inflectional 
	morphology.  Memory and Cognition, 25(4), 425-437.
Swinney, D. A. (1979). Lexical access during sentence comprehension:  
	(re)consideration of context effects.  Journal of Verbal Learning
	and Verbal Behavior, 18, 645-659.
Tabossi, P., & Zardon, F. (1993).  Processing ambiguous words in 
	context.  Journal of Memory and Language, 32, 359-372.
Taft, M. (1981).  Prefix stripping revisited.  Journal of Verbal Learning
	and Verbal Behavior, 20, 289-297.
Taft, M., & Zhu X. (1995).  The representation of bound morphemes in the
	lexicon:  a Chinese study.  In L. B. Feldman (Ed.) Morphological
	aspects of language processing (pp. 293-316). Lawrence Erlbaum.
Zhou X., & Marslen-Wilson, W. (1994).  Words, morphemes and syllables in
	the Chinese mental lexicon."  Language and Cognitive Processes,
	9(3), 393-422.
Zhou, X., & Marslen-Wilson, W. (1995). Morphological structure in the
	Chinese mental lexicon. Language and Cognitive Processes, 10 (6),
	545-600.