In Elfling message 34423
(<http://groups.yahoo.com/group/elfling/message/34423>),
in reference to the newly published _Words, Phrases and Passages in
Various Tongues in The Lord of the Rings_ by J.R.R. Tolkien
(published in _Parma Eldlamberon_ 17, ed. Christopher Gilson
<http://www.eldalamberon.com/parma17.html>)
David Salo writes, :

> Having had a chance to glance at a copy of this text, I must say
> that my first reaction is not to note all the places where past
> analysts (including me) got things wrong -- as such "errors"
> were inevitable and, given the state of the data, irremediable
> at the time. It is rather to note where the analysts got things
> right, working only with published material. Such analyses, being
> often not the subject of controversy (though in all justice they
> might have been), often fall into "received wisdom" and so
> receive no notice. However, I have personal knowledge, in some
> cases, from what slender chains of evidence some of the
> conclusions were derived, and it is merciful to note when they
> are vindicated.

David is of course correct to highlight the happy instances where
previous guesses about the grammar of Tolkien's Elvish and other
invented languages, as exemplified in _The Lord of the Rings_, have
turned out to be correct (in the sense that they agree with Tolkien's
own explanations as he developed them in the 1960s -- or, in some
cases, agree with at least one of a set of competing explanations
Tolkien developed then). This is a testimony both to the prevailing
consistency of Tolkien's linguistic creations -- i.e. to Tolkien's
own adherence to the abstract systems of phonological and grammatical
rules he selected for his languages at any given time -- as well as
to the power of the application of the tools of historical and
comparative linguistics to Tolkien's languages. (I would however note
that in the case of conclusions derived from "slender chains of
evidence", turning out to have been "right" is more likely due to
mere luck, the more slender are the chains!)

The power of these tools, however, is not absolute, particularly when
they are applied to sets of inconsistent linguistic evidence (i.e.,
to evidence selected from multiple, inconsistent linguistic systems),
and when ultimately statistical inferences drawn from sets of data
far too limited to support them are relied upon to justify claims of
oddness (or even error). It is therefore unsurprising that those
analysts that did so produced what David alludes to at the start of
his statement: many guesses that turn out _not_ to accord with
Tolkien's explanations at all. And it is surely at least as important
to draw lessons from these errors as it is to find affirmation in
successes; for in fact, most of the errors David alludes to were
neither "inevitable" nor "irremediable" before the publication of
Tolkien's own explanations, but were the result of unsound methodology.

-----

To pick just one of the most prominent errors to exemplify the
matter, consider the past tense verb of Sindarin. What has come to be
regarded by many as the "standard view" (what David rightly now calls
the "received wisdom") of the formation of the Sindarin past tense --
as most prominently presented by Helge Fauskanger on his
_Ardalambion_ web site (from whence the "standard view" has been
received by the many) and later by David Salo in his book _A Gateway
to Sindarin_ -- holds that:

I) The "regular" past tense stem of most basic verbs is formed by the
infixion of _-n-_ (with subsequent phonological development) to the
verb stem.

II) The "regular" past tense of some basic verbs -- sc., those whose
stems end with a nasal or sonant (e.g. l, m, n, r) -- is formed by
the addition of a suffix _-n_ to the verb stem (with subsequent
phonological development).

III) An "irregular" past tense stem of some basic verbs is formed by
the infixion of _-a_, or by lengthening of the stem vowel (with
subsequent phonological development).

IV) The "regular" past tense stem of derived verbs is formed by the
addition of a suffix _-nt_ to the verb stem.

What is usually not understood by the many, however, is that the
formulation of this "standard view" relied on a set of assumptions
and exclusions that, far from being "inevitable" or "irremediable"
before the publication of _Parma_ 17, were already dubious, and even
demonstrably wrong, when _Ardalambion_ and _Gateway to Sindarin_ were
first published. For if it were true that only the publication of
_Parma_ 17 falsified the "standard view", then it would follow that
the "standard view" _did_ accurately and completely describe both
Sindarin and Noldorin as they were evidenced before the publication
of _Parma_ 17; but in fact the "standard view" failed to accurately
or completely describe either Sindarin or Noldorin, even just as they
were evidenced before the publication of the "Addenda and Corrigenda
to _The Etymologies_" in 2002; and it failed for three reasons:

1) The data counted as evidence in the formulation of the "standard
view" are drawn not just from Sindarin proper, but also, and
overwhelmingly, from Noldorin of _The Etymologies_. This was done
despite the clear knowledge that Noldorin and Sindarin are not
identical in all details of phonology, morphology, or semantics,
despite the clear knowledge that Tolkien (by his own account)
ceaselessly altered his descriptions and conceptions of his
languages, and despite the clear knowledge that Noldorin of _The
Etymologies_ preceded Sindarin in Tolkien's conceptual development by
at least 15 years. And so it was already known that the "standard
view" did not accurately describe either Sindarin or Noldorin, as
they were then evidenced, but rather described an artificial amalgam
of the two languages.

2) The data counted as evidence in the formulation of the "standard
view" also included purely hypothetical forms and formations that
were (and still are) actually nowhere attested. These supposed past-
tense forms are derived from past participles, on the assumption that
past participles are formed from "the" past-tense stem: e.g., from S.
_tirnen_ is derived the hypothetical form *_tirn_ (this was and is
the sole basis of formation II of the "standard view" above, the
"regular" past tense of basic verbs whose stems end with a nasal or
sonant: e.g. l, m, n, r.) But it was not then, nor yet, actually
demonstrated that past participles actually are formed from the past-
tense stem for all verb classes. And so it was already known that the
"standard view" did not accurately describe either Noldorin or
Sindarin as they were then evidenced (nor as now evidenced), since it
prescribed a formation not actually attested in either language.

3) Not all of the data either for Sindarin _or_ for Noldorin of _The
Etymologies_ that were then available were included; rather, certain
forms were discounted as evidence in the formulation of the "standard
view":

- One Noldorin past-tense form, _mudas_ *'laboured, toiled' (stem
*_muda-_ < *_môtâ- < MÔ-), which is given in _The Etymologies_ as
published in 1987, was excluded.

- One of the (only) four past-tense forms then actually attested for
Sindarin proper, _agor_ 'made, did' (stem _car_- < KAR-), which was
published in the essay "Quendi and Eldar" in 1994 (XI:415), was
excluded.

And so it was already known that the "standard view" did not
accurately describe either Sindarin or Noldorin as they were then
evidenced, since it excluded certain formations that actually were
evidenced for those languages.

It is important to understand just why two forms that exemplified
what in fact proved to be significant formations in Noldorin and
Sindarin, respectively, were excluded from the formulation of the
"standard view". In both cases, the chief explanation given was that
the form is "odd" (and in the case of _mudas_ as a mistake by J.R.R.
or Christopher Tolkien, and as in any event an impossible form
because it would clash with nouns in _-s_ < *_-sse_), since it didn't
accord with any other attested formation. Now, this argument is
ultimately statistical, and relies on statistical significance for
validity. A formation attested in only one form among hundreds or
even among dozens of other verbs might justifiably count as "odd".
But when a formation is attested in one of only about a dozen forms
(as _mudas_ was among derived verbs before the "A&C" was published),
or worse, in one of only _three_ forms (as _agor_ was among past
tenses of basic verbs in Sindarin, proper, when the "standard view"
was formulated; and one of only four past tense verbs of any
structure), it is surely dubious _at best_ to draw any statistical
inference, given the tiny sample space (a point that Thorsten Renk
made long ago).

Furthermore, the dismissal of _agor_ as "odd" is readily shown to
have been completely unjustified _at any time_, since Tolkien cited
it in "Quendi and Eldar" specifically as an example of _an entire
formation class_, and one which moreover he called the _usual_ past-
tense formation of basic verbs in Sindarin, characterized by "the
augment, or reduplicated base-vowel, and the long stem-vowel", i.e.,
_agor_ < *_a-kâr-_ < KAR-. Given this statement, and given the
invented nature of Sindarin, it was already known in 1994 that _agor_
stood for what must have been a large number of forms in this augment
+ long stem-vowel class in Sindarin proper. Had Tolkien's statement
simply been accepted at face value, this important formation class
would have been part of the "standard view" from long ago.

Unfortunately, still in 2004 (and beyond) _agor_, and thus this
formation class, was being discounted from the "standard view" by its
chief proponents, justified with an appeal to statistics. David Salo,
for example, in _A Gateway to Sindarin_, writes (p. 118) that:

"This formation was said to be 'usual in Sindarin "strong" or primary
verbs" ..., but in fact examples are much rarer than those of the
nasal past. One might expect such formations as *_udul_ 'he/she/it
came', *_idir_ 'he/she/it watched', *_egin_ 'he/she/it saw', etc.,
but these are not in fact found".

We see here, still in 2004, the failure to accept Tolkien's statement
about his own language at face value, and the unjustified appeal to
statistics that wrongly counts _agor_ as just one isolated form
instead of as an exemplar of an entire formation class -- both of
which errors were entirely avoidable. (Moreover, while it is true
that *_udul_, *_idir_, and *_egin_ are unattested, neither are the
forms prescribed by the "standard view", including by _Gateway_
itself -- sc., *_toll_, *_tirn_, *_cen(n)_ -- "in fact found" either,
though David gives both "_toll_" and "_narn_" in _Gateway_ without
any indication that they are in fact entirely unattested. Again, we
see an entirely avoidable, and frankly unjustified, inconsistency of
method and reasoning.)

In a similar vein, Helge Fauskanger writes (in his article
"Reconstructing the Sindarin Verb System",
<http://www.uib.no/people/hnohf/sverb-rec.htm>)
that: "Agor as the past tense of car- is of course a rather surprising
form. ...this way of forming preterites would seem to contradict
other sources" (But those "other sources" are, in fact, just one:
_The Etymologies_, and so this Sindarin formation only "contradicts"
Noldorin evidence, not Sindarin evidence, and so is no contradiction
at all!) Helge then writes: "Where would this leave the [Sindarin]
past passive participles that are apparently formed by adding -en
to the past tense? Does not [t]irnen as the word for "watched,
guarded" ... presuppose *tirn as the past tense of tir-?".

Now, _tirnen_ does _not_ in fact "presuppose *tirn as the past tense
of tir-", since this presupposition itself derives from an unproven
assumption that the past participle is always formed from "the" past
tense stem of the corresponding verb, an assumption that has yet to
be proven by any actually attested past tense forms of verbs of this
class. There is a basic rule of scholarly and scientific reasoning
that states that if evidence contradicts premises, it is the premises
that must be altered or discarded, _not the evidence_. In other
words, one does not get to exclude evidence merely because it
violates ones assumptions. Again, this was an entirely avoidable error.

-----

So, what can we learn from all this? As it happens, nothing much that
wasn't already known, and said, before, since it is mostly just good,
basic scholarly practice:

o Lesson 1: Keep your datasets coherent, consistent, and separate.

The "standard view" of the Sindarin past tense failed because it
included Noldorin data as Sindarin data, and because one (actually)
Sindarin form (of only four then attested) was discounted as "odd"
because it was inconsistent with the Noldorin data. But in fact,
Noldorin is Noldorin, and Sindarin is Sindarin: their data are
separated chronologically by at least 15 years, and are inconsistent
phonologically, morphologically, and semantically. Thus the data of
the former is not the data of the latter. Therefore, discrepancies in
the data between the two languages are not "contradictions" (as they
ought not be _assumed_ to be consistent in the first place) and
cannot be used as justification for excluding data from either: they
are (simply) differences between two distinct linguistic systems.

o Lesson 2: Don't count assumptions as facts against data.

The "standard view" of the Sindarin past tense failed because it
assumed that a) the past participle is formed from the past-tense
stem, and that b) there is only one "correct" past-tense stem for any
given verb: e.g., that the past tense of _tir-_ must be *_tirn_ (and
only *_tirn_) because of the attested past participle _tirnen_.
Since _agor_ did not conform to the "standard view", which prescribes
that the past tense of _car-_ is *_carn_, it was further discounted
as an "odd" form.

o Corollary: When data contradicts an assumption, the assumption, not
the data, must be discounted.

The form _agor_ proves that *_carn_ is not the (one and only)
"correct" past-tense form of _car-_, contrary to the "standard view";
and since no such formation as *_carn_, *_tirn_, etc. is actually
attested, the "standard view" should have been modified to account
for the data; instead, the data was discounted because it did not
conform to the "standard view". This is the opposite of the scholarly
method.

o Lesson 3: Statistical arguments are valid only when statistical
significance can be established.

The "standard view" of the Sindarin past tense failed because it
assumed that the fact that a formation is attested in only one form
out of a total corpus of just a handful of forms is significant. This
was an unjustified assumption.

There is one additional lesson, specific to Tolkienian linguistics;
indeed, to the study of any language that is solely the invention of
one man:

o Lesson 4: Tolkienian linguistics is largely a _metalinguistics_.

The "standard view" failed because it did not take Tolkien's own
statements about his own languages at face value, and so failed to
accept that _agor_ is not an isolated form, but rather an exemplar of
an entire formation class. This was because those formulating the
"standard view" believed that Tolkienian linguistics neither is nor
should be conducted differently from the linguistics of "real"
languages.

In the linguistics of "real" languages, the linguist collects all the
forms he can find, compares them with each other to find patterns of
correspondence, organizes them into classes according to these
perceived patterns of phonological, morphological, and semantic
correspondence and development, and then writes descriptions of these
classes of patterns, correspondences, and developments that account
for all the collected forms. The problem with trying to apply this
method to Tolkien's languages, however, is that Tolkien did not
create his languages by simply writing out texts in his languages, or
even by simply creating dictionaries: rather, he created his
languages (overwhelmingly) by providing _linguistic descriptions of
his languages_. That is, he created his languages by writing what in
the case of "real" languages would be the _end result_ of linguistic
study, _not_ by providing the masses of "raw data" that were the
_input_ to that linguistic study. In other words, Tolkien's writings
about his languages are designed to image the _results_ of the work
of numerous linguists that, within the fiction, _have already done_
the linguistic collection, classification, and description. In other
words still, Tolkien's languages exist not (just, or even primarily)
as masses of raw language data, but rather as _descriptions_ of the
phonological, morphological, and semantic classes and features of his
languages, most of which are only _exemplified_ by a handful of
forms: we are _not_ given the mass of data that within the fiction
would have been used to establish these descriptions.

Because of this, we in fact can know many things about Tolkien's
languages that cannot be established based solely on an inspection
and count of the forms he left behind, in isolation from Tolkien's
linguistic descriptions. For example, we know -- and have known since
1994, despite its absence from the "standard view" -- that the
_usual_ past tense of basic verbs in Sindarin proper is formed by the
augment with lengthening of the _sundóma_, and that this large
formation class is _exemplified_ but not exhaustively _evidenced_ by
Tolkien with the form _agor_. And we knew this because Tolkien
provided a linguistic statement describing this formation _class_,
_not_ because of the number of forms attested from this class.

The ongoing publication of Tolkien's linguistic descriptions thus
makes, and will continue to make, this source of information about
Tolkien's languages ever more important and preponderant, while
diminishing the reliance of our understanding on the collection and
linguistic examination of "raw data". The chief task of _Tolkienian_
linguistics, proper, will increasingly become a sort of
_metalinguistics_: that is, a study of Tolkien's _linguistic
descriptions_ and their implications, rather than of the exemplifying
(_not_ exhaustive) data embedded in those descriptions.

-----

I have detailed all this at such length because I believe that there
are in fact very important lessons to be learned from "not[ing] all
the places where past analysts got things wrong", and determining
which of those errors were not in fact "inevitable" and
"irremediable", but instead resulted from avoidable failures of
methodology. My convictions are that:

1) The methodological failures I've identified and given specific
examples of above are detectable throughout what has for some years
now been regarded as the chief "standard" presentations of "Quenya"
and (most especially) of "Sindarin";

2) These methodological failures were all along eminently avoidable
simply by adhering to the standards of scholarly practice and
reasoning, and by recognizing and acknowledging the true nature of
Tolkien's languages, of his creation of those languages, and of the
(increasingly) "metalinguistic" nature of Tolkienian linguistics; and

3) The sooner these methodological failures are recognized and
abandonded in favor of sound, scholarly methodologies, the better the
subsequent study and understanding of Tolkien's languages, and of
Tolkienian linguistics, will be.

Carl