Down with Morphemes: The Pitfalls of Concatenative!
Morphology!
Autore: David J. Peterson
!
Data della SM: 10-01-2009!
!
Data FL: 03-01-2014!
!
Numero FL: Florida-00001E-00!
!
Citazione: Peterson, David J. 2014. “Down with Morphemes:
The Pitfalls of Concatenative Morphology.”
Florida-00001E-00, Fiat Lingua, !
!
!
Diritto d'autore: © 2014 David J. Peterson. This work is
licensed under a Creative Commons Attribution-
Non commerciale, senza derivati 3.0 Unported License.!
!
!
http://creativecommons.org/licenses/by-nc-nd/3.0/
!
Fiat Lingua è prodotto e gestito dalla Language Creation Society (LCS). Per maggiori informazioni
riguardo alla LCS, visita http://www.conlang.org/
Down with Morphemes:
The Pitfalls of Concatenative Morphology
David J. Peterson
October 1, 2009
Abstract1
This paper explores the relationship between an adherence to the main
tenets of concatenative theories of morphology and the creation of less
than realistic languages.
1
introduzione: Morphology 101
In any introductory linguistics class, students will be introduced to the concept of
morphology. The first lesson usually goes something like this:
There is a
In English, we have a word like “cat” and a word like “cats”.
systematic relation between the two in both form and meaning, in that “cat”
refers to one feline entity and has no “-s” suffix, e “cats” refers to more than
Perciò, we can say that the
one feline entity and has an “-s” suffix.
phonological sequence [kʰæt]—or “cat”—refers to a particular feline entity, e
the suffix “-s” means “plural”.
Modern day linguists have taken this rather basic notion of morphology quite a bit
further in recent years2, but that research, by and large, hasn’t informed the practices
and habits of modern day conlangers.
In this paper, I shall examine the notion of the morpheme and discuss its impact
on modern conlanging. I will then propose an alternative to traditional morpheme-
1 This paper is based on a talk I gave at the First Language Creation Conference in 2006 (“Down with
Morphemes!: What Word and Paradigm Morphology Can Teach Us about Language Creation”). While
that talk focused a good deal on Bochner’s version of Word and Paradigm Morphology, this paper will
instead take a closer look at the notion of the morpheme and its role in modern day conlanging.
2 For two rather sophisticated (though divergent) theoretical approaches, see Ackerman, et al (2009) e
Halle and Marantz (1993).
1
based approaches to language creation, and finish with some sample derivations and
evolutions.
1.1 A Note to Non-Artlangers
It’s true that, for the most part, the target audience of this paper is those who aim to
create languages that are more or less naturalistic (and human). Tuttavia, those
who are interested in creating auxlangs or non-human languages may profit by reading
this paper. Per esempio, many alien languages focus on semantics and/or grammatical
categories while still employing grammatical strategies that are far from alien. Here’s an
artificial example:
(1) ðəeoikpɬɛʃtʉm-yɱfstɨk ɣʟhʏəjrvzd ɯʉɨiɨl ʍɵvbʏχq-ɑwɢɪtʰajoki.
/title-DMLR.AND. ɣʟhʏəjrvzd ɯʉɨiɨl
separate.from.two.boxes-NON-PAST.NON-FUT.NON-PUNC./
“ðəeoikpɬɛʃtʉmyɱfstɨk ɣʟhʏəjrvzd is separating an ɯʉɨiɨl from two boxes.”
This is a fake alien language, and it looks pretty strange. It’s nearly
unpronounceable, it distinguishes ninety different titles (this one is for a Dimlarian
android), ɣʟhʏəjrvzd is a pretty crazy name, it has affixes which only tell you what
something isn’t (non-past, non-future, non-punctual), and one can’t even approximate
in English what an ɯʉɨiɨl is. Then you have the verb ʍɵvbʏχq, which is a basic verb (COME
basic as “to eat” in English) which means “to separate something from two boxes”. Can’t
get more alien than that, right?
BENE, perhaps one can. The sample above, in fact, is all but equivalent structurally
to the following sentence of Turkish:
(2) Bay-an Gül mektup yaz-ıyor.
/title-FEM3 Gül letter write-PRES.CONT/
“Miss Gül is writing a letter.”
The number and strangeness of the titles in this alien language, focusing on one
example, doesn’t make the structure “title base + modifying suffix” any less ordinary.
Allo stesso modo, natural languages of all stripes have “verb + tense/aspect suffix”. Increasing
the number of suffixes and varying what they encode doesn’t make a language any less
human: it just makes encountering its system in the wild increasingly unlikely, e
makes the language look, for lack of a better word, artificial.
3 I make no theoretical claims about the status of the -an suffix in Turkish. Suffice it to say that Bay is the
equivalent of English “Mr.” and Bayan is the equivalent of English “Miss”.
2
When creating an alien language, bearing in mind how human languages work
can help to create something that’s truly non-human. Additionally, if one is trying to
create an ideal language, one needs to know what to avoid, and what actually isn’t too
much trouble for language users.
2 The Morpheme
The notion of a morpheme can be traced back to Panini, but a good definition for how it
is understood today can be found in any elementary linguistics textbook. Qui, per
example, is how morphemes are introduced in the fourth edition of Contemporary
Linguistics:
The most important component of word structure is the morpheme—the smallest
unit of language that carries information about meaning or function.
Così, the meaning of a word like “bats” can be cut up as follows:
Orthographic Form
Phonetic Form
Meaning
bat
-s
[bæt]
[s]
flying mammal
plural number
Figure 1: Morphemic analysis of English “bats”.
Crucial here is the relationship between the singular and plural forms for the
flying mammal “bat”. Specifically, in a morpheme-based theory, we know that the [s],
for example, carries information about the word’s meaning because it is the sole
difference between the singular and plural form of “bat”. No such systematicity can be
seen with, for example, IL [æ] In “bat”. Exchanging that sound for [ɛ] will, Infatti,
produce a new word, but as there is no identifiable systematic relationship between
gambling and flying mammals, there is no evidence for [æ] being a morpheme in “bat”,
just as there is no evidence for [ɛ] being a morpheme in “bet”4.
This type of analysis (whether of necessity or not is a matter worthy of
discussion) had a profound effect on the understanding of meaning. As morphemes are
4 An interesting chicken and egg argument arises just from examining this account of the relationship
between words. Specifically, in order for a difference between two words to be morphological (and in
order to determine whether or not a given bit of phonological material is a morpheme), a categorical or
paradigmatic distinction must already exist—even though it’s the morphemes themselves which
presumably create these categories. Cioè, a singular and plural paradigmatic distinction in nouns must
exist in order to prove that the “-s” suffix is a plural morpheme.
3
the storeholders of meaning, larger meaning is combined in precisely the same way that
morphemes themselves are combined. Così, a word like “bats” has a kind of
mathematical spellout as shown below:
bat
[bæt]
flying mammal
+
+
+
-s
[s]
plural
=
=
=
bats
[bæts]
many flying mammals
Figure 2: The concatenation of “bat” e “-s”.
The linear order of [bæt] e [s] In “bats” can be verified rather simply. The linear
order of the meanings of “flying mammal” e “plural” are a different matter, but we’ll
leave that aside for the moment. Let’s call this State Zero for morpheme-based analysis.
With this understanding in mind, I’ll go on to examine various problems encountered
by morpheme-based analyses, and what was done to resolve them.
3
Problems with Morpheme-Based Analyses
What follows is by no means either a chronological or an exhaustive account of the
various problems morpheme-based analyses have faced and resolved. Invece, what I
hope to do is illustrate the types of problems morpheme-based approaches face, and the
types of solutions proposed. After discussing several examples in this section, I’ll
discuss what is gained and what lost by the various resolutions that have been utilized
to rescue morpheme-based approaches to grammatical analysis in section 4.
3.1 Allomorphy
To get us started, I’ll examine a widespread and fairly recognizable phenomenon:
allomorphy. Continuing with our “bat~bats” example, the English plural is complicated
slightly when other nouns are added to the list:
(3)
UN.
b.
[bæt] “bat (sg.)” ~ [bæts] “bats (pl.)”
[dɔɹ] “porta (sg.)” ~ [dɔɹz] “doors (pl.)”
4
c.
[wɪʃ] “wish (sg.)” ~ [wɪʃɨz] “wishes (pl.)”5
No longer can one say that [s] is the English plural suffix, as now we see both [z.z]
e [ɨz] also acting as plural suffixes.
Though the suffixes differ, they seem to be related, COME [s] e [z.z] differ only in
voicing. Further analysis reveals that [s] occurs only after voiceless non-sibilant
consonants, con [z.z] occurring elsewhere (with a little something extra for words
ending in [s], [z.z], [ʃ] o [ʒ]). And so, borrowing some terminology from phonology, [s],
[z.z] e [ɨz] are said to be allomorphs of an underlying morpheme we’ll call /z/6.
With allomorphy, the linguist’s task is clear: to locate and define all the
morphemes present in a given language, and then to describe what rules govern the
distribution of their surface forms.
3.2 Categorical Alternations
Immediate (but by no means insurmountable) obstacles to morpheme-based analyses of
allomorphy arise when one ventures out into the wilds of natural language data. With
the regular English plural suffix, one might make an argument that its allomorphs are
the simple result of garden variety phonological alternations. Such can’t be said of some
of the irregular plural suffixes that we see elsewhere. Consider the following:
(4)
UN. ox ~ oxen
b. child ~ children
c. schema ~ schemata
d. datum ~ data
e. corpus ~ corpora
syllabus ~ syllabi
f.
Taking (4a) as an example, there is no formal way to predict that the plural of
“ox” should be “oxen”, given regular alternations like “fox~foxes” e “box~boxes”.
Piuttosto, it appears that the plural of “ox” is something that must be learned by rote and
memorized thereafter.
5 A quick note on transcription. I’m operating under the assumption that English has two reduced vowels:
a low and a high one. Others may transcribe reduced vowels differently—and, Infatti, may transcribe
many other items in this paper differently—but the choices in phonetic transcription don’t have any
bearing on the thrust of this paper. Suffice it to say that those items which must be distinguished shall be.
6 Calling the underlying English plural morpheme /z/ is not uncontroversial, but the controversy has no
direct bearing on the matter at hand, so I’ll leave it aside.
5
Traditional morpheme-based theories have a way to account for such
alternations. Focusing on “oxen”, one way to account for the data would be to posit two
classes of nouns in English: Class A nouns (those that take some variant of the regular “-
s” plural suffix) and Class B nouns (those that form their plural with the “-en” suffix).
Once the possibility of classes such as the above is admitted, the problem of what to do
with the data in (4) is solved: Each word belongs to its own special class. One might
summarize the alternations shown in (4) below:
(5)
UN. Class B: -ø (sg.)/-en (plu.)
b. Class C: -ø (sg.)/-ren (plu.)
c. Class D: -ø (sg.)/-ta (plu.)
eccetera.
On one level, this is a satisfactory explanation, in that it accounts for the data,
and it further codifies the notion that these forms are, Infatti, irregular, and they must
be learned. I’ll put off discussion of further issues arising from an analysis like this until
later. For the time being, I’ll continue with more English data.
3.3 Word-Internal Alternations
Already, cracks have begun to emerge in the explanation evinced in 3.2. Consider the
formulation of Class C in (5) I offered. It fairly characterizes the orthographic differences
between “child” e “children”, but it’s missing a crucial distinction all English speakers
make. Consider the added phonetic transcriptions of “child” e “children” below, along
with a few other relevant singular/plural pairs:
(6)
[tʃajɫd] “child (sg.)” ~ [tʃɪɫdɹɨn] “children (pl.)”
[ɡus] “goose (sg.)” ~ [ɡis] “geese (pl.)”
[maws] “mouse (sg.)” ~ [majs] “mice (pl.)”
UN.
b.
c.
d. [lif] “leaf (sg.)” ~ [livz] “leaves (pl.)”
e.
[wʊmɨn] “woman (sg.)” ~ [wɪmɪn] “women (pl.)”
If one ignores the theoretical notion of “morpheme”, the apparatus developed in
3.2 would have no problem accounting for the data in (6). Specifically, Class C would be
modified to account for the vowel change, and new classes would be developed from
(6b-e). Doing so, Tuttavia, fails to take seriously the theoretical notion of “morpheme”.
To reconcile the English data with a morphemic analysis, one has to expand one’s
understanding of what a morpheme is, and how it works. Take (6a), “child~children”.
6
It’s evident that the plural suffix is “-ren”, and that this is where the notion of plurality is
housed—that is, “-ren” is the plural morpheme. In addition to being atomic units,
Anche se, some have argued that morphemes are agents, in a sense7. In questo caso, adding
the plural “-ren” morpheme to “child” causes the vowel to change from [aj] A [ɪ].
Another way to rescue a morphemic analysis is to claim that the [ɪ] In “children”
is actually a part of the plural morpheme. The nomenclature might be a little
cumbersome (I imagine something like “-[ɪ]-…-[ɹɨn]”), but one could, essentially,
identify a discontinuous morpheme that acts as both suffix and vowel replacer. Come
would help to explain the alternation between “woman” e “women”, where the plural
morpheme is no longer a suffix, but either an infix, or a “vowel replacer”8. And, Di
course, the same analysis could be extended to “goose~geese” e “mouse~mice”.
Datum (6d) is a bit troubling, despite the advances in our theory. We see the
usual “-s” plural suffix working as it should (consider [kʰev] “cave” ~ [kʰevz] “caves”),
but there’s a difference in the voicing of the stem-final consonant between the singular
and plural that should be surprising. We can say for certain that it’s not a regular
phonological alternation (cioè. that the /f/ isn’t voicing due to the voiced [z.z] suffix)
because we might otherwise expect a voiced labio-dental fricative in “leaf’s” (as in “IL
leaf’s ruddiness”), “leafs” (as in “He leafs through the book”) and/or “Leafs” (as in “IL
Toronto Maple Leafs”)9. Somehow, Poi, the facts will have to be accounted for
morphologically.
Sticking to our morpheme-based analysis, we can posit a different type of
otherwise regular plural suffix. It looks and acts just like the ordinary “-s” plural suffix,
but this one causes voicing in the final consonant. Such a suffix would prove quite
useful, in that it would also help to explain pairs like [najf] “knife” ~ [najvz] “knives”
and even [haws] “house” ~ [hawzɨz] “houses”. The mechanism would be roughly akin to
IL “-ren” suffix which also causes a vowel change.
If that approach is unpalatable, we can borrow the “vowel replacer” idea and
apply it to great effect in (6d). To make it work, we simply change the “vowel replacer”
into a “feature replacer”, such that we have the regular “-s” suffix accompanied by a
floating [+voice] feature which replaces the [-voice] feature of a stem-final voiceless
7 In particular, I’m thinking about a rule-based Item and Process approach, but a specific reference doesn’t
come to mind.
8 The distinction between these two theoretical notions is absolutely crucial (as will become apparent),
even though the idea of a “vowel replacer” is a bit silly.
9 The argument comes, most famously, from Pinker (1995), who claims that the compound is “Maple
Leaf”, and that it’s the compound that’s pluralized, and not the individual word “leaf”. As the compound
is treated as a regular nominal compound, it takes the regular plural allomorph, giving us “Maple Leafs”.
One has to wonder, Anche se, why one will quite commonly both hear and see in print “Toronto Maple
Leaves” (a quick google search turned up close to 200,000 hits), but one absolutely never hears
“saberteeth” as a serious plural of “sabertooth”.
7
fricative of a noun that falls into whichever class “house”, “wife”, “knife”, “leaf” e
others like them belong to. Di nuovo, the nomenclature will be a bit odd (something like
“[+voice]…-[z.z]”), but nomenclature is merely a representation, after all, of what is really
happening.
3.4 Zero Allomorphy
So far so good (more or less). We’ve accounted for all the data and maintained in tact the
notion of “morpheme” (though its status grows more and more abstract). Dealing with
plurals in English, there is one last bit of evidence which can be summarized as a null
alternation:
(7)
UN. sheep ~ sheep
b. fish ~ fish
c. deer ~ deer
d. bison ~ bison
Very quickly, we can assure ourselves that these are count nouns, and not mass
nouns, because sentences like, “I see three deer in the meadow yonder” are perfectly
grammatical.
Alternations such as those in (7) would appear to pose a serious challenge to
morpheme-based analyses. (Un?)fortunately, such alternations have been accounted for
rather simply thanks to the notion of a zero morpheme: a phonologically null suffix
which changes the meaning from singular to plural. Così, the morphological structure
Di (7a) is as follows:
sheep
[ʃip]
bleating, wooled mammal
+
+
+
-ø
[ ]
plural
=
=
=
sheep-ø
[ʃip]
many such mammals
Figure 3: The concatenation of “sheep” e “-ø”.
I should note that though this null morpheme is conventionally realized as a suffix, its
placement, as it is theoretical, is up to interpretation, and may just as well appear as a
prefix, circumfix, or anything else one likes.
A further class, Poi, can be posited whose members all take a phonologically
null plural suffix to form their plurals, and all the data in (7) have an explanation. We
8
were forced to posit a phonologically null suffix, but if one accepts the explanation in
3.3, this new development should prove neither surprising nor troubling.
3.5 Fallout
At this point, we can effectively leave off analyzing natural language data, as pretty
much any imaginable morphological phenomenon can be accounted for with the
machinery we’ve constructed. Essentially, the notion of a morpheme can be abstracted
to the point where one can take it completely out of the phonological structure of a
word and claim that it is responsible for everything going on inside the word. But we
needn’t go that far to see that the explanation provided up to this point leads to a
number of rather drastic theoretical consequences.
To begin with, let us return to (6b): “goose~geese.” We’ve already posited one of
two possible explanations: either a zero allomorph of the plural suffix which effects a
change in the vowel (turning the morpheme into a function as well as an atomic entity),
or a “vowel replacer” morpheme which changes “-oo-” A “-ee-“. There is another
possibility I didn’t mention which should help to elucidate one of the largest problems
morpheme-based accounts of language face—namely, Quello “-ee-” o “-[io]-” is an infix—an
actual infix.
If we admit, for the moment, Quello “-ee-” is an infix (a variant of the plural
allomorph), then it would mean that the stem to which it applies is “gse” (phonetically,
[ɡs]). It might seem a bit strange to native English speakers, but it would solve a few
problems. A new question arises, Poi: What is the status of “-oo-” o “-[tu]-“? Rightly or
wrongly, we are forced to accept that “-oo-” is a singular infix.
Now we have a “problem” that actually clarifies a number of the ambiguities
present in the preceding argument. Primo, we knew already that English had a “singular”
category just as it had a “plural” category. As this is the case, how does one account for
the concatenation of meaning in a word like “dogs”? Se “dog” means “noun, singular”,
Poi “dogs” must mean “dog, singular, plural”—that is, a mammalian quadruped that is
both singular in number and non-singular in number at the same time.
There is, of course, an easy fix to this problem. We’ve already posited a null
plural allomorph. What, Poi, is wrong with a null singular allomorph? We could argue
that the null singular is the regular (Class A) singular morpheme, and that others (such
COME “-oo-“) are irregular. Doing so would allow us to account for all the data in a simple
and straightforward fashion. Below is a sample of that analysis:
9
Noun Class
Sample Lexeme
Singular Affix
Plural Affix
Class A
Class B
Class C
Class D
Class E
Class F
Class G
Class H
Class I
Class J
Class K
[bæt] “bat”
[ɑks] “ox”
-[ ]
-[ ]
-[s]
-[ɨn]
[tʃɫd] “child”
-[aj]-
-[ɪ]- -[ɹɨn]
[skimə] “schema”
-[ ]
[deɾa] “data”
[ɡs] “goose”
-[m]
-[tu]-
[ms] “mouse”
-[aw]-
[kɔɹp] “corpus”
[lif] “leaf”
-[ɨs]
-[ ]
-[tə]
-[ ]
-[io]-
-[aj]-
-[ɚɹə]
-[+voice]-…-[z.z]
[wmn] “woman”
-[ʊ]-…-[ɨ]-
-[ɪ]-…-[ɪ]-
[ʃip] “sheep”
-[ ]
-[ ]
Figure 4: Morphemic analyses of some of the data presented.
I’ve made a number of theoretical decisions in the table above (e.g. analyzing
“data” as a word with a null plural suffix and a singular -[m] suffix) that are largely
unimportant, as reformulating them would produce the same result: a lexeme column, UN
singular column and a plural column each populated by a bunch of morphemes. What
they are makes little theoretical difference.
And this, of course, arises from a simple examination of English pluralization
strategies. If one takes the entirety of the English language into account—let alone the
languages of the world—the picture gets even muddier.
4
Language Creation and Linguistics
A questo punto, it’s my hope that the reader will grasp his her or head in disgust and cry
out, “How unbelievably absurd! Why have I been taking morpheme-based theories of
grammatical description seriously all these years?!”10, but I recognize that that will not
be the reaction of most. Infatti, the most powerful thing about a framework is that it
can withstand pretty much any assault—sometimes by its proponents simply ignoring
10 Or, even better, “Gee, that’s just what I’ve been thinking all these years!”
10
any and all counterarguments11. There will always be a way to rescue any framework,
including morpheme-based analyses of language.
At this point, what I want to do is demonstrate what effect morpheme-based
theories have on conlanging, as that question is of central importance to this paper.
Before leaving linguistics behind, Anche se, I have a few brief comments and observations
produrre.
4.1 What Is Lost
The state of affairs in figure 4 above may look bleak to some, but to others the analyses
may seem perfectly viable. To others still, the analyses will look like they could use a bit
of pruning and polishing—that they’re in need of just a bit of tweaking to bring them in
line with a modern morpheme-based account of English plural patterns. Purtroppo,
no matter how much these analyses are spruced up, the most troubling problems will
remain.
Consider the problem of allomorphy. In English, there’s a simple bit of
allomorphy with the indefinite article “a/an”. Fondamentalmente, you get “UN” before words that
begin with a consonant phone and “an” before words that begin with a vowel phone, COME
below:
(8)
UN. a sheep / *an sheep
b. a conlanger / *an conlanger
c. an owl / *a owl
d. an omen / *a omen
The environment is simple, as is the rule, but the form of the indefinite article is
not, necessarily. That is, though it might be perceived as being simpler to have a
consonant in between two vowels to break up hiatus, there’s no special reason that that
consonant is [N]. It could have been anything, or a different word entirely. The form,
Tuttavia, just happens to be “an”. Stating that “UN” e “an” are allomorphs of the same
morpheme captures this distinction nicely: They are two morphologically related forms
which are not related phonologically, even though the allomorphy is phonologically
conditioned.
The problem with a morpheme-based analysis is that all morphemes bear the
same relationship. Cioè, there is no reason to expect that two morphemes should look
anything alike in any circumstance.
11 Cf. Postal (2009) on Noam Chomsky, Per esempio.
11
Consider the plural morpheme in “leaves” and the plural morpheme in “greaves”.
According to the analysis presented in figure 4, it’s an accident that these morphemes
look and sound alike. As the morpheme in “leaves” participates in a different type of
alternation (associated also with a voice alternation in the stem), it’s a different
morpheme (or allomorph), e, come tale, may have taken on any form. Cioè, we might
have “greave~greaves” e “leaf~leavork”. Formally, the morpheme-based approach
fails to capture the simple generalization that the suffix is identical.
In a similar vein, the theoretical notion of a morpheme has nothing to say about
seemingly bizarre situations such as that exemplified in the data below:
“to sip”
(regular)
sip
sipped
“to hit”
(irregular)
“to shit”
(archaic)
“to shit”
(modern)
hit
hit
shit
shat
shit
shit
Present
Past
Figure 5: Present and past forms of “to sip”, “to hit” e “to shit”.
In English, certain uncommon irregular verbs have gradually fallen into disuse,
and their accompanying irregular past tense forms have been forgotten (consider how
infrequently English speakers use “throve” as the past tense of “thrive”). The archaic past
tense of “to shit” is still around, but it’s considered a bit archaic or fanciful (or comical).
The more common past tense form is identical to the present tense form.
Morphemically, one would have to analyze the above thus:
“to sip”
(regular)
sip + -ø
sip + -ed
“to hit”
(irregular)
hit + -ø
hit + -øx
“to shit”
(archaic)
shit + -ø
“to shit”
(modern)
shit + -ø
shit + -øy
shit + -øx
Present
Past
Figure 6: Morphemic analysis of the forms in Figure 5.
The analysis in Figure 6 is, at best, purely descriptive: It’s a simple statement of
facts. Tuttavia, doesn’t it seem odd that the former irregular verb “to shit” lost its
former irregular pattern and gained another irregular pattern? Why isn’t “shitted” IL
common past tense? A morphemic analysis has nothing to say about this—nor anything
to say about the fact that the past tense form “shat” is still available to English speakers.
12
Perhaps more troubling is the status of morphemes themselves. By definition, all
morphemes are equal. Morphemes are the “building blocks” of language, and though
they differ in distribution (free vs. bound), IL “atomic weight” of each morpheme, so to
speak, is equal. If one takes the notion of “morpheme” seriously, then the plural
morpheme “-s”, the third person singular agreement morpheme “-s”, and the word
“adroit” should all take up the same space in one’s head. As all are morphemes, IL
brain should make no distinction between one or the other, and they should be treated
roughly the same.
The notion seems to run counter to the basic facts of language. Language users
interact with the morphology of a language—the collection of so-called derivational and
inflectional morphemes—quite differently from the lexicon—i.e. (in this instance) IL
collection of lexical items that comprise one’s vocabulary. As a quick example, language
users will frequently forget words, yet it would be quite odd for a speaker to say
something like, “Yeah, I saw Tommy. He’s eat…eat…dang it, what’s the suffix? Oh yeah:
-ing. He’s eating in the backyard.” There may be other explanations for this
phenomenon, but the fact remains that morphology and lexical items are treated quite
differently by language users. Conflating the two seems to be setting up a kind of
psychological fantasy that treats human brains much like computers: machines that can
amass data, but can’t do anything with it unless told specifically how to evaluate it. To
the extent that one ought to be concerned about the psychological reality of a given
theory, the foundation of morpheme-based approaches should, at the very least, be
reconsidered.
4.2 The Proper Place for Linguistics
It has been discussed at length at various times on various online listservs and fora just
what the relationship between linguistics and language creation is, and what it ought to
be. Certainly, many conlangers have had disheartening experiences within academia,
and discouraging encounters with academics—both within linguistics and without.
And though there have been strong opinions about what role language creation should
play—if any—in linguistics both within the community and within academia, Ore
appears to be resolving the matter for us. I personally have had wonderful experiences
in linguistics, and received a lot of encouragement from a number of linguists—several
of whom are conlangers themselves. There may still exist some antagonism, but I
believe the instances are growing fewer in number, and that the situation will improve
in the future regardless of any action taken by any party in the future.
Much less attention has been paid to what role linguistics and linguistic
knowledge should play in language creation. The online community of conlangers
consists of a large number of young people who either are taking linguistics courses or
have in the past, along with several professional linguists or former linguists, e
13
countless self-taught amateur linguists. A kind of knowledge separate from academic
linguistics altogether has grown and expanded with the online community, and can be
considered the collective common knowledge of the online conlanging community. Questo
knowledge, shared piecemeal in forum postings, blog posts, private and public e-mail,
and even on Twitter and Facebook status updates, has helped to enrich the community
and has advanced the art of language creation considerably in the past decade or so.
The extent to which this collective knowledge overlaps with general academic
linguistic knowledge is uncertain, but the mapping surely isn’t perfect12. IL
inconsistency may be a result of the method of delivery and the consumption of the
knowledge that comes to the community from academic linguistics. As nearly as I can
tell, there are three major sources of linguistic knowledge:
1. Linguistics students and professors.
2. Published papers and books.
3. Online sources (usually second-hand, at best).
From these sources, generally conlangers make use of the following: reported
structures from natural languages; linguistic universals; methods of analysis; e
various “rules” derived from specific analyses. Here I would like to focus on the last
two.
Ignoring errors of transmission, there is a grave danger in making use of
linguistic frameworks used for analysis and the results of such analysis in constructing
a language. The job of a theoretical linguist is clear: To analyze language data. Cioè, UN
linguist starts with data, and ends up with an analysis. Many (if not most) conlangers
have implicitly defined the job of a conlanger as being the exact opposite of a linguist’s—
questo è, to begin with an analysis, and to work backwards to create data.
There are a number of very serious problems with the “Reverse Linguistics”
approach to conlanging. I’ll briefly examine three of them.
Primo, linguists make mistakes, and frameworks change—as do languages, per
that matter. Considering languages of the Philippine type (Tagalog, Chamorro, Ilokano,
Cebuano, etc.), I’ve read Relational Grammar and Minimalist Program analyses of the
same language, and the two are night and day. Often different linguists from different
traditions will disagree on the basic facts of a given language, and their frameworks
differ so greatly that they can’t even have a conversation about the data. It’s more than
likely that a paper written now on a given language will be inaccurate (or perhaps
“inaccurate”) in no more than ten years’ Ore.
12 Consider the Conlang Trigger System: An entirely novel system based on a misunderstanding of an old
analysis of languages like Tagalog. The system is one that doesn’t exist in any natural language, but it’s a
workable system that’s been put to good use by a number of conlangers (myself included).
14
Secondo, one of the driving assumptions behind much of the work of modern
linguists is that all languages are the same (this, in particular, is a problem within the
generative tradition). As such, there’s an external pressure amongst many theoretical
linguists to make languages “fit”. Challenging a generally accepted universal is a serious
thing. It’s more felicitous to discover that a construction that looks completely bizarre
and other-worldly actually works just like something else that has already been
explained in another paper written in the same framework. And, unsurprisingly, this
happens quite a bit—which is used as evidence that the approach was correct to begin
con, and the vicious circle continues. The fault, Anche se, lies not with linguists, by any
means. The problem is that many of the frameworks developed by linguists are
indestructible. A simple Item-and-Arrangement morphemic framework can be used to
analyze any conceivable language; there is literally no phenomenon it can’t handle.
Whether these analyses are worthwhile is usually a question left for later.
Finalmente, there is a pressure amongst many phonologists, morphologists and
syntacticians to ignore the history of a language. The reasoning is sound: Children are
born into this world without any knowledge of the history of any language on Earth—
and yet, they all learn to speak. This means that they work only with the data presented
them—the synchronic state of the language. As such, the history of a given language
must be ignored by linguists trying to analyze the synchronic phenomena present in that
language. Whether one agrees or disagrees with this theoretical assumption as it applies
to linguistics, I would argue that one cannot ignore the history of a constructed
language. Linguists try to come up with an analysis for how a child will make sense of a
given system; a conlanger is trying to create that system—and, Forse, the series of
linguistic events that gave birth to that system. The two are entirely different
enterprises, and require different methodologies.
Ultimately, I would argue that within the realm of language construction,
linguistic knowledge should be subordinate to one’s own sense of how to construct a
language. As it is today, often linguistic knowledge is taken as a kind of law to be
obeyed. Realistically, the accumulation of linguistic knowledge is simply not
sophisticated enough to act as a measure of any kind for a constructed language. IL
reason is simple: Linguists and conlangers are working at cross purposes. It seems
unlikely that there will ever be a time that linguistic knowledge can be used in the way
conlangers presently would like to use it simply because the answers linguists seek are
not of immediate use to a conlanger seeking to create a naturalistic language. In fact,
utilizzando “Reverse Linguistics” will likely lead one to creating a rather clunky and
unnatural conlang, as we will see in the next section.
15
5
Language Evolution
The problem with linguistic universals is that they’re notoriously malleable. They
prescribe precisely what natural languages can and can’t do—until a natural language
does something universals say it oughtn’t, in which case the list of universals is
emended (or the language data is reanalyzed). In other words, linguistic universals
allow linguists to predict what natural languages can and can’t do—except for the stuff
they can’t predict. For one attempting to construct a naturalistic language, this
information is not terribly useful.
All natural languages do have one thing in common, Anche se: They were all
created by humans. That fact is by no means trivial. The human brain has been pretty
much the same for centuries. Ma ancora più importante, the way in which a human community
will interact with and manipulate data has also remained static. Examining language
data, Poi, is fine, but what we see are the results of humans’ interaction with and
manipulation of language. By simply examining the results, a conlanger can only
produce variations thereon. What the naturalistic conlanger needs to do is emulate the
method whereby human communities evolve language.
In this section, I’ll illustrate how Reverse Linguistics can lead to unnatural
results, and discuss language evolution.
5.1 Problems with Reverse Linguistics
Below are some sample nominative and accusative forms in four different languages:
Nominative
Accusative
Latin
eqvvs
eqvom
Russian
Turkish
Spanish
kniga
knigu
kitap
kitabı
gato
gato
Figure 7: The nominative and accusative forms of Latin “horse”, Russian “book”, Turkish “book” e
Spanish “cat”.
A strict morphemic analysis would render each nominative and each accusative
form identical: STEM + -NOM.SG for the nominatives and STEM + -ACC.SG for the
accusatives. The simplest way to create a new system based on that analysis is to create
a bunch of nouns and a bunch of suffixes. And by and large, many purportedly
naturalistic conlangs end up as, essentially, long lists of stems and affixes.
The problem here isn’t necessarily with the analysis: The problem is with how
that analysis is used. A simple examination of the languages in question will reveal that
16
the systems are far, far more complex than the analysis above would lead one to believe.
In Latin, there are different declension classes and countless exceptions. In Russian, IL
genitive plural of kniga is knig. In Turkish, the accusative is used only with definite
direct objects. In Spanish, certain direct objects actually take a kind of special case
marker depending on whether the object is animate and specific. These details aren’t
revealed with a simple morphemic analysis, but when conlangers start with that kind of
an analysis in mind, they often are unable to effectively (or realistically) reproduce what
is left out by the analysis. In effect, many conlangs are actually phonological realizations
of morphemic linguistic analyses.
5.2 Starting Over
To begin a naturalistic conlang is to face a great mystery—one that may never be fully
understood: The origin of language. As far back as we can go, there has always been
language—and the further back we go, the spottier the records get. No one knows how
language emerged from nothing, or what it looked like at its earliest stages. Even our
best opportunities to study this process—modern pidginization and the creation of
signed languages—have been lost, as linguists are usually alerted to the presence of a
new pidgin or sign languages some fifty or so years after the fact—if they’re lucky13.
As such, a naturalistic conlanger has to make an executive decision: Where to
start, and what to start with. It would be mere speculation to even guess what a brand
new language will look like ten years after its inception—especially a language created
ex nihilo. It’s up to the conlanger to decide where the language will begin, what system it
will have, what lexemes, e così via, and it’s at this point that the creator will have to
put their foot down and say, “That’s moment zero.” That is, before that point, IL
language has no history; its lexemes are unanalyzable; its systems came from nowhere.
As unsatisfying as that may be, there’s simply no other alternative (aside from those
supplied by imagination14).
Once a conlanger determines the initial state of the language, its evolution can
begin.
13 It’s worth noting, I feel, that this is, in large part, due to prejudice. Most pidgin languages are regarded
by both their speakers and their overhearers as ungrammatical, unimportant—even shameful—and not
“real” languages. The same can be said of many sign languages—even well-established ones. We can only
hope that humanity has learned its lesson with respect to these wonders of innovation, and that the next
time a language is born, linguists will be there at ground zero as events unfold.
14 For example, a good number of artlangs have imagined a kind of being or society that created a
language (perhaps a “logical” o “ideal” language), and that this construction became the proto-form of
the modern language.
17
5.3 Some Examples
The languages we have today are the result of the tension between two opposing forces.
Whatever they’re called—innovation and conservatism; dynamism and stasis;
faithfulness and markedness—their opposition fuels language growth and language
change. In order to create a naturalistic conlang, this tension must be replicated—or
fabricated, almeno.
Take English irregular plurals as an example. By looking at the endpoint, UN
conlanger trying to replicate what they see in English might create something like the
following:
(9)
UN. mate “bear” ~ matek “bears”
b. virt “sword” ~ virtak “swords”
c. borl “mouse” ~ berl “mice”
d. wug “pond” ~ wugop “ponds”
e. romp “otter” ~ ramp “otters”
toli “card” ~ tuliar “cards”
f.
And a nice description might accompany this, such as, “The regular plural suffix
È [-k] after vowels and [-ak] after consonants. There are a number of irregular classes,
pure, which you just have to memorize.” That latter bit may be true of an English speaker,
to some extent, but from the point of view of the language, the irregular classes we have
are not random: They were arrived at systematically. Taking two lexemes completely
out of context may make the system look strange (Dire “house~houses” vs.
“mouse~mice”), but looking at the system as a whole, one can see, even without
knowing the history of the language, that the irregularity is principled. Infatti, broadly
speaking, principled irregularity is the only kind of irregularity there is, and it can’t be
faked.
Consider how we arrived at “house~houses” e “mouse~mice”. There was a
stage of English where the alternation in “mouse” was regular. Through a series of
regular sound changes, shown below, it came to be irregular in the modern form of the
language (note: GVS below stands for “Great Vowel Shift”):
(10)
Initial State: muːs ~ muːsiz
Loss of [-z]: muːs ~ muːsi
muːs ~ myːsi
Umlaut:
Loss of [-i]: muːs ~ myːs
Unrounding: muːs ~ miːs
18
GVS:
maws ~ majs
Though the type of alternation in the “mouse” paradigm is rare (seen elsewhere
only in “louse~lice”), the sounds themselves are familiar enough; English is chock-full of
“-ouse” e “-ice” words. Trying to engineer something similar by going backwards, one
is much more likely to end up with random results (like that dreadful [-op] suffix in 9a).
Being able to manipulate sound changes is just part of the problem when it
comes to naturalism, Anche se. One might wonder why the series of changes that gave us
“mouse~mice” didn’t apply more generally, and why, Per esempio, the plural of “house”
isn’t “hice”. In fact, there’s a famous poem by an anonymous author that bemoans the
state of English plurals, a section of which I’ll reproduce below:
If I speak of a foot, and you show me two feet,
And I give you a book, would a pair be a beek?
I’ve reproduced this section in particular because the sound changes above did
apply to “book”, giving us a plural that would now be spelled “beek”. Why do we have
“Libri” now? An accident of history, really. Sometimes irregularities simply don’t take.
With verbs in English, by and large, it’s the most common irregular verbs that stay
irregular, the less common eventually becoming regular, but it’s hard to argue that
“book” is less common than “mouse”—or “louse” o “ox”, for that matter15. More to the
point, changes—be they phonological, semantic or pragmatic—are often capricious in
their application. It’s the conlanger’s job to figure out not only what changes applied
and how they operated, but when they applied, what portion of the vocabulary was
affected, what portion of the vocabulary didn’t yet exist, and in which words the
irregularities failed to survive.
A similar problem arises from what I call morphemic pressure. In a standard verb
paradigm (let’s say present tense indicative, two numbers, three persons), morpheme-
based thinking tends to lead to a pressure to produce either six affixes—one for each cell
—or four affixes—one for each person, and one for plural. Consider, Anche se, a modern
French paradigm (phonetic forms, not orthographic):
15 Well, nowadays, at least. Perhaps the answer lies in books not being as common with the majority of
those who used English back when the irregular patterns were forming…
19
Person/Number
Singular
First Person
Second Person
Third Person
desid
desid
desid
Plural
desidõ
deside
desid
Figure 8: The phonetic forms of the present tense conjugation paradigm for the French verb décider16.
Synchronically, and without the rest of the language in mind, the paradigm looks
exceedingly strange. Yet it is this type of strangeness that is natural. Replicating this,
Anche se, should not entail, Dire, taking four out of six cells of a paradigm like this and
deciding that the forms will be identical, giving a unique suffix to the remaining two
forms. The key lies in the development of the forms and the system.
5.4 Instituting Change
The first step towards producing a naturalistic language on par with the natural
languages we see in the world is the complete and total abandonment of the morpheme.
It is not useful, e, in fact, does more harm than good. Do not misunderstand me to
say that this means one should abandon affixes; by no means. All I mean is don’t confuse
the fact that an [-s] appears with a large number of plural nouns to mean that the [-s]
means “plural” in any real or useful sense.
This first step is important, because it will help to simplify the process of
evolution to follow. Consider a language with no verbal morphology (presumably a
very old language) to which one hopes to add inflections for person and number. Where
are they going to come from? If they’re going to be suffixes, are the suffixes going to
appear out of nowhere?
As it turns out, a number of languages develop person morphology from
personal pronouns (sometimes possessive, sometimes not), and at the initial stages, they
simply are pronouns—and are often identical in form to, for example, possessive
pronouns. In Middle Egyptian, for example, the exact same suffix for first person
possession ([-i] o )17 was used for a first person subject on verbs. At that stage (or for
16 The orthographic forms (going from first person to third and from singular to plural) are as follows:
décide, décides, décide, décidons, décidez, décident.
17 There is a standard convention in Egyptology which I have not followed here. The transcription of
glyphs is often given in a transcription system unique to Egyptology which employs letters and letter-like
symbols rather than IPA. The standard transcription of the double reed is a “j” with a c over it instead of a
dot.
20
some time during the history of Middle Egyptian) the two were identical, because they
were, in fact, the same element.
The key point here is that the forms aren’t fixed: neither the meanings nor the
phonology nor the parts of speech. All of these facets of a lexeme are open to
interpretation by speakers. Sometimes this can result in a change in usage, or a change
in meaning, a change in pronunciation—and sometimes a lexeme can lose its identity
altogether. A linguistic system is susceptible to constant reevaluation by its speakers,
and speakers of a given language, like it or not, simply don’t recognize morphemes.
Consider the wild reanalysis by English speakers shown below:
(11)
laughing > laughin’
UN.
b. running > runnin’
c. skipping > skippin’
jumping > jumpin’
d.
thumping > thumpin’
e.
This is a fairly well-documented and well-understood change that happened
probably over a century ago in English. If the reader is unfamiliar with the orthographic
form, the change here is from a velar nasal [ŋ] to an alveolar nasal [N], and the
perception is that the latter is somehow “less formal” than the former. This change,
Anche se, went beyond its original domain of application into other realms:
(12)
UN. nothing > nothin’
b. something > somethin’
Despite the fact that “nothing” e “something” don’t actually have the “-ing”
suffix, the rule still applies quite comfortably. It’s not a rule that applies generally (e.g.
one can’t say “thin'” per “thing” o “flin'” per “fling”—I doubt even if one can even say
“weddin'” as a noun [almeno, not in all dialects]), e, furthermore, doesn’t even apply
to all compounds with “thing”:
(13)
UN. everything > *everythin’
b. anything > *anythin’
One might try to capture the generalization above by stating that only disyllabic
compounds with “thing” can be reduced, but then there are certain dialects that allow
both “everythin'” e “anythin'”. Why should it be acceptable in one and not in another?
21
What rule can account for such a thing? And what does one call the suffix/non-suffix
whose final [ŋ] can change to [N]? It’s certainly not a morpheme—in (12) e (13), it’s
not even a suffix18.
The nice thing, from the point of view of a conlanger, is one doesn’t have to call it
anything at all—and, in fact, one is a better off not doing so. A native speaker is often
blind to grammatical descriptions, e, come tale, will find patterns where there are none
—or, perhaps ought not to be. And this is how we get reduced forms like “nothin'” e
“somethin'” in English.
5.5 Expansion and Evaluation
Evolving a language is a kind of cyclical two step process that takes place on many
linguistic planes at the same time (phonological, morphological, syntactic, semantic,
pragmatic…). The two steps one can call expansion and evaluation.
The definitions of “expansion” e “evaluation” are a bit nebulous, because, as of
yet, the specifics of linguistic evolution are poorly understood, at best. We know,
Anche se, that these stages have to exist based on what we have seen and what
documentation we have of older stages of the world’s languages.
What I’m calling “expansion” is a period of change in the state of a language.
How long this period is, and what it encompasses, I’m not sure anyone can say. Ma, per
example, we know that at one point in time, English didn’t use “go” to form any kind of
future tense. Then at some point in time, “I go to eat” was reinterpreted as “I will eat”,
and soon it became “I’m going to eat”, and then “I’m gonna eat”, and then “I’m’a eat”,
and then “I’m’a go eat”. I’d say an English speaker currently still has access to the
original meaning of the phrase, Anche se, so we’re not yet at the point of no return—i.e.
we still have access to the literal meaning of “I’m going to eat”. Expansion can affect the
phonological system in the form of sound changes, Quale, in turn, can affect the
morphological system. It can affect the syntax (consider that Latin was SOV, but that
Spanish is now SVO, e, in fact, moving towards VSO), the semantics (the original
meaning of “silly” was “holy”, etc.), and the various periods of expansion can happen
simultaneously or independently of each other.
The other side of the coin is what I call “evaluation”. At some point in time, UN
language needs to settle down and allow its speakers to take stock of just what is a part
of their linguistic system, and what isn’t. Per esempio, at one point in time English had
a fairly stable intervocalic voicing rule. If we borrow a new word into the language that
ends with [f], Anche se, Quello [f] isn’t going to become [v] if we add a V-initial suffix
anymore. The period of expansion that gave rise to intervocalic voicing came to an end,
18 Though it might be amusing to imagine it is. If I’m everything right now, then yesterday I everythed,
and my friend habitually everyths through my things.
22
and the period of stability that followed was left behind by new periods of expansion.
It’s that in between period of evaluation that defines the language with respect to a
certain feature for a period of time. At some point a given expansion is accepted into the
system and the system remains stable until a new period of expansion comes along and
the older expansion is fossilized.
A good way to characterize it might be to consider the rings on a tree. When one
examines a tree stump, one can see concentric circle in the tree radiating out from its
center. The rings are visible precisely because of these periods of expansion separated
by periods of stability (or vice versa, if you prefer). Inoltre, if one is standing next
to a tree at any given time, it’s nearly impossible to tell if it’s growing or not. It looks
stable enough, after all. The same can be said of language. Considering one’s own
language, it’s nearly impossible to tell if it’s changing—or how, or where it’s headed.
Languages outlive us, and grow more slowly than we do. The best we can do is
appreciate the changes that we’ve been able to document, and try to simulate (at high
velocità) the process ourselves by conlanging.
6 Conclusion
The approach I’m advocating is not new. Infatti, the first modern artlanger, Tolkien,
approached conlanging from a diachronic perspective. Oggi, Anche se, we enjoy an
advantage Tolkien didn’t have—one he remarked upon himself, in fact19. Specifically,
we have each other; we have a community: We have an audience. And if we are to
engage in conlanging as an art, then we have to move it forward.
It is on us to both define what rigor in naturalistic conlanging is, and to apply it,
if not to each other’s work, then at least to our own creations. In the past, it was such a
tremendous feat to admit to another that one engaged in language creation that the
notion of criticism was anathema. Since conlangers came together and began to form
the rudiments of what now is a robust online community in 1991, we’ve come a long
modo, and I think it’s time to move beyond our initial state of trepidation. The vice is
secret no longer.
This paper is the first step towards a definition of naturalism in conlanging20. È
an attempt to take stock of what we’ve learned as a community over the years and to
give us the language we need to articulate what it is that we do and why it’s special to
the uninitiated. If we don’t take our work seriously, no one else will, and if we can’t
19 “You must remember that these things were constructed deliberately to be personal, and give private
satisfaction—not for scientific experiment, nor yet in expectation of any audience. A consequent weakness
is therefore their tendency, too free as they were from cold exterior criticism, to be ‘over-pretty’, to be
phonetically and semantically sentimental—while their bare meaning is probably trivial, not full of red
blood or the heat of the world such as critics demand.” (From “A Secret Vice.”)
20 Note the article: a definition of naturalism in conlanging.
23
adequately explain why the elements of one conlang are superior to another, Poi, nel
eyes of the uninitiated who are just now beginning to look in on the world of language
creation, all conlangs will be equal—or will be defined by characteristics which they
themselves supply. In altre parole, if we don’t have a way to define rigor in conlanging,
very soon a definition will be thrust upon us, whether we like it or not.
7 References
Ackerman, Farrell, Gregory T. Stump, and Jim Blevins (editors). 2009. Paradigms and
Pariphrasis. Stanford: Center for the Study of Language and Information.
Bell, Sarah J. 1983. “Advancements and Ascensions in Cebuano.” Studies in Relational
Grammar, ed. David M. Perlmutter. Chicago: University of Chicago Press:
143-218.
Bochner, Harry. 1993. Simplicity in Generative Morphology. New York: Mouton de
Gruyter.
Bybee, Joan, Revere Perkins and William Pagliuca. 1994. The Evolution of Grammar: Tense,
Aspect, and Modality in the Languages of the World. Chicago: University of Chicago
Press.
Campbell, Lyle. 1998. Historical Linguistics: Un introduzione. Cambridge, MA: MIT Press.
Halle, Morris and Alec Marantz. 1993. “Distributed Morphology and the Pieces of
Inflection.” The View from Building 20, ed. Kenneth Hale and S. Jay Keyser.
Cambridge, MA: MIT Press: 111-176.
Lakoff, George and Mark Johnson. 1980. Metaphors We Live By. Chicago: University of
Chicago Press.
O’Grady, William, John Archibald, Mark Aronoff, and Janie Rees-Miller. 2001.
Contemporary Linguistics. 4th edition. Boston: Bedford/St. Martin’s.
Perlmutter, David M. and Paul M. Postal. 1983. “Towards a Universal Characterization
of Passivization.” Studies in Relational Grammar, ed. David M. Perlmutter.
Chicago: Stampa dell'Università di Chicago: 3-29.
Pinker, Steven. 1995. The Language Instinct. New York: HarperPerennial.
Postal, Paul M. 2009. “The Incoherence of Chomsky’s ‘Biolinguistic’ Ontology.”
Biolinguistics 3.1 (url http://www.biolinguistics.eu/): 104-123.
Tolkien, J.R.R. 1983. The Monsters and the Critics and Other Essays. New York:
HarperCollins.
24