Proceedings of the 8th International Workshop on Tree Adjoining Grammar and Related Formalisms, pages 159–164,
Sydney, July 2006. c©2006 Association for Computational Linguistics
Reconsidering Raising and Experiencers in English
Dennis Ryan Storoshenko
Department of Linguistics
Simon Fraser University
Burnaby, B.C., Canada
dstorosh@sfu.ca
Abstract
In this paper, structures involving the rais-
ing verb seem, are examined. Specifically,
it is shown that previously-proposed ele-
mentary trees for seem with an experiencer
argument are inadequate, based upon syn-
tactic testing. In Storoshenko (2006), new
articulated structures for the seem pred-
icate are proposed, modelled upon the
treatment of ditransitive verbs. This pa-
per recapitulates and further motivates the
ditransitive-style analysis, while illustrat-
ing its potential value in issues surround-
ing extraction and the raising construction
in TAG.
1 Introduction
The raising predicate seem is often cited as one of
the core examples in discussions of TAG’s appli-
cation to natural language syntax. Under a gener-
ative/minimalist account, a sentence such as (1a)
will have the underlying structure in (1b):
(1) a. John seems to like coffee.
b. Johna0 seems a1 ta0 to like coffeea2 .
In TAG, the subject John remains local to the ele-
mentary tree headed by like, the elementary tree in
which its theta role is assigned. The observed dis-
placement effect is a result of the extension of the
like-headed tree after the adjunction of an auxil-
iary tree headed by seem (Kroch and Joshi, 1985).
In the more recent analysis of Frank (2002), a sen-
tence such as (1a) is derived through the composi-
tion of the elementary trees of Figure 1 to derive
the final tree in Figure 2.
TP
DP
John
Ta3
T
to
VP
V
like
DP
coffee
Ta3
T VP
V
seems
Ta3a5a4
Figure 1: Elementary trees to derive John seems to
like coffee.
TP
DP
John
Ta3
T VP
V
seems
Ta3
T
to
VP
V
like
DP
coffee
Figure 2: Derived tree for John seems to like cof-
fee.
159
1.1 Defining the Problem
At issue in this paper will be the structure of sen-
tences such as those in (2):
(2) a. John seems to me to like coffee.
b. John seems to like coffee to me.
Here, a prepositional phrase to me now appears
in the clause; as illustrated, its position is vari-
able. The individual introduced in this preposi-
tional phrase is interpreted as being an experiencer
of the verb seem, in no way dependent upon the
embedded like predicate. As such, according to
the Fundamental TAG Hypothesis (Frank, 2002),
this experiencer must be composed as a part of the
seem auxiliary tree. For discursive ease, the case
in (2a) will be termed a medial experiencer, and
the (2b) case will be a final experiencer. What is
now required is an auxiliary tree for seem which
retains the desired recursivity, and supports this
experiencer in either possible position. Further
syntactic diagnostics will be used to determine the
necessary shape of such an auxiliary tree.
1.2 An Existing Account
In Frank (2002), a structure is given for this type
of raising verb with an experiencer, as in Figure 3.
Ta3
T VP
V
seems
PP
P
to
DP
Ta3 a4
Figure 3: Auxiliary tree for seem with an experi-
encer (Frank, 2000)
This tree would adjoin into the Ta3 node of an in-
finitival clause tree, as in Figure 1, yielding the
correct string order (after substitution of the fron-
tier DP-experiencer), for a raising sentence with a
medial experiencer (2a). Frank’s discussion of this
ternary structure is essentially limited to the well-
formedness of its functional architecture, and the
fact that a stipulation will need to be put in place to
obviate the satisfaction of the T head’s EPP feature
by the experiencer. While a valid point, there are
still two key unanswered questions with regards to
this structure: first of all, are the complements of
the verb straightforwardly interchangeable (to ac-
count for the variable position of the experiencer),
and is there any evidence for or against the ternary
branching structure? These questions emerge to be
inter-related, and in exploring the consequences of
the ternary structure, it will be shown that simple
transposition of the verb’s complements is not an
option within a flat ternary structure.
2 Establishing Argumenthood
Before embarking upon a discussion of the con-
sequences of Frank’s ternary branching structure,
a more straightforward solution must be consid-
ered. Instead of treating it as a part of the seem-
headed tree, one could attempt to formulate an
argument that the prepositional phrase bearing
the experiencer is introduced as a syntactic ad-
junct. This could be conceivably be accomplished
through the use of one of the two trees of Figure
4. These are adjunct auxiliary trees, recursive on
VP, which would introduce an experiencer prepo-
sitional phrase at either the left or right periphery
of the VP, respectively.
VP
PP
P
to
DP
VPa4
VP
VPa4 PP
P
to
DP
Figure 4: Possible adjunction structures for an ex-
periencer prepositional phrase
While an anonymous reviewer points out that
considering the experiencer to be an argument of
seem is quite uncontroversial, there does appear
to be some evidence that a prepositional phrase of
this form, serving to introduce something akin to
an experiencer, can exist independent of the pred-
icate seem:
(3) a. ? John to me likes coffee.
b. John likes coffee to me.
160
While the first example here sounds quite marginal
to the ears of most native speakers, the second sen-
tence is perfectly acceptable, and is a likely para-
phrase of a sentence such as John seems/appears
to like coffee to me. This suggests at least the pos-
sibility that the prepositional phrase bearing the
experiencer might be considered an adjunct1.
However, in the case of a sentence such as (2a),
it can be easily demonstrated that adjunction of
the prepositional phrase as an independent auxil-
iary tree is not an option. Adjunction of the right-
recursive VP tree of Figure 4 into the VP node of
either tree of Figure 1 would, after all the trees
were composed, yield one of the following string
orders:
(4) a. * John seems to to me like coffee.
b. * John to me seems to like coffee.
As shown, there is no way to derive the me-
dial experiencer string-order using a simple VP-
adjunction tree. This provides clear evidence that
the mechanics of TAG derivation force an analysis
where at least the medial experiencer must enter
the derivation as a part of the seem auxiliary, giv-
ing further thrust to the contention that the experi-
encer here is indeed an argument of seem.
In turning to the experiencer in final position,
matters are less clear-cut, as there is a viable struc-
ture in which the prepositional phrase can adjoin
to the seem auxiliary and appear at the end of
the sentence, using the left-recursive tree of Fig-
ure 4. Recalling the examples of (3), it is pos-
sibly even more important to establish the argu-
menthood of this position, as there are strikingly
similar sentences in which the equivalent prepo-
sitional phrase appears to be a bona fide adjunct.
For the final experiencers of seem, evidence can
be provided to show that the prepositional phrase
is not opaque to extraction, and therefore not an
adjunct:
(5) a. The woman whoma0 John seemed to
like coffee to ta0 kept refilling his cup.
b. John seems to like coffee to the wait-
ress. Her boss, too.
1The possibility that sentences such as those in (3) are de-
rived from a raising structure from which the raising predicate
seem was subsequently elided can be easily dismissed. Aside
from employing a host of tests to identify elision phenomena,
one must simply observe that the verb like appears with finite
tense, a distinct anomaly if one were to treat it as having been
part of a raising structure.
c. Whoa0 is it that you saw the woman
who seemed to like coffee to hima0 ?
In (5a), it is quite clear that the experiencer can be
relativised out of the final position with no diffi-
culty at all. Similarly, the stripping case in (5b),
where it also seems to Mary’s boss that John likes
coffee, indicates that the experiencer her boss can
be extraposed from the sentence final position, and
the rest of the sentence stripped away. Finally, the
use of a resumptive pronoun to repair the com-
plex noun phrase constraint violation in (5c) pro-
vides further proof that the final-position prepo-
sitional phrase is not opaque to extraction. This
is thus an argument position, part of the seem-
headed auxiliary. As such, the question left at the
end of Section 1 must now be answered: can the
ternary-branching auxiliary tree account for inde-
pendent syntactic observations related to this par-
ticular structure?
3 An Alternative View
At first glance, Frank’s ternary branching struc-
ture is reminiscent of early accounts of ditransi-
tive verbs. Such structures were famously argued
against in Larson (1988), and subsequently re-
examined in Harley (2002). In these treatments, a
ternary structure is replaced with a VP-shell struc-
ture, as schematised in Figure 5.
VPa0
DPa1a3a2a5a4a7a6a9a8 Va0 a3
Va0 VPa10
DPa11a13a12a15a14a5a16 Va10 a3
ta0 DPa17a19a18 a4a21a20a22a4
Figure 5: Schematic tree for a ditransitive verb
phrase
In the lower VP, the goal and theme of a di-
transitive verb are projected as the specifier and
complement, respectively. The verb itself then
raises to an upper VP, which supports the agent
of the ditransitive predicate. The motivation for
adopting this structure lay in the observation of c-
command phenomena between the goal and theme
161
positions. In a flat ternary structure, mutual c-
command between these two positions would be
expected, however Larson gives considerable data
to argue that mutual c-command does not exist be-
tween these two positions.
In looking at the tree from Figure 3, it is clear
that straightforward considerations of mutual c-
command will not be informative, as one of the
ternary branches of the seem-headed tree will con-
tain the remainder of the embedded clause ma-
terial which exists below the Ta3 adjunction site.
However, what can be observed is whether or not
a c-command relation exists between the experi-
encer of seem and the embedded clause theme.
This will speak to the matter of the possible trans-
position of the VP complements: if they do indeed
exist in a flat structure, then the experiencer should
c-command the embedded clause theme from both
the medial and final positions2.
In Storoshenko (2006), it is argued that a seem
auxiliary with an experiencer should be analysed
with a similar VP-shell analysis. Among the ev-
idence provided, three of Larson’s c-command
tests are employed to illustrate that the experiencer
of seem does c-command the embedded clause ob-
ject when in the medial position:
(6) a. John seems to nobody to like any-
thing. (NPI Licensing)
b. John seems to every boya0 to like hima0 .
(Bound Variable)
c. * Whata0 does John seem to whom to
like ta0 ? (Superiority)
For negative polarity licensing and bound vari-
able readings to obtain in these cases, the expe-
riencer must c-command the direct object. Sim-
ilarly, the fact that extraction of the embedded
clause theme (which would not in itself be the
product of an ill-formed elementary tree), is un-
grammatical here. This is a straightforward supe-
riority violation, again illustrating that the experi-
encer c-commands the embedded theme.
The opposite is demonstrated to be the case
where the experiencer is in the final position:
(7) a. * John seems to like anything to no-
body.
b. John seems to like him
a4
a0 to every boya0 .
2The observed ability of an argument DP to c-command
out of its PP in this type of structure is noted in Jackend-
off (1990)
c. Whata0 does John seem to like ta0 to
whom?
Here, the negative polarity item is not licensed,
and a bound variable reading does not obtain.
However, the embedded theme can be extracted
in the case where the experiencer is in the final
position. These results demonstrate that in the fi-
nal position, the experiencer does not c-command
the embedded object, contrary to what would be
expected of a flat ternary structure like that of Fig-
ure 4. The experiencer must not be in a position
where it c-commands the embedded clause mate-
rial beneath Ta3 . The elementary trees for seem
with an experiencer in medial and final position,
respectively, are given in Figure 6.
Ta3
T VPa0
Va0
seemsa0
VPa10
PP
P
to
DP
Va10 a3
Va10
ta0
Ta3 a4
Ta3
T VPa0
Va0
seemsa0
VPa10
Ta3 a4 Va10 a3
Va10
ta0
PP
P
to
DP
Figure 6: Two seem-headed trees with experi-
encers (Storoshenko 2006)
As in the case of the ditransitive structure of
Figure 5, there is verb movement here. The lower
VP supports the experiencer and the Ta3 foot node,
essential if recursivity is to be maintained, while
seem itself raises to an upper VP projection. Un-
like the ditransitive case, seem projects no position
for an agent argument, which retains Frank’s argu-
mentation for having an elementary tree rooted in
162
Ta3 . Crucially, this movement is licensed within
TAG, as it remains local to this one elementary
tree, and has no impact upon the recursive nature
of the tree.
In terms of the relationship between the two
experiencer positions, there are two possibilities,
both of which have been explored in the paral-
lel literature on ditransitives. In the pattern of
Larson (1988), the two trees of Figure 6 would
be derivationally related, one having been derived
from the other. Countering this is the approach
of Harley (2002), in which similar alternations are
argued to be the result of lexically distinct (yet
phonetically indistinguishable) predicates project-
ing different syntactic structures. The second ar-
gument is taken in Storoshenko (2006): there is
no derivational relationship between the two trees
Figure 6. Each is headed by a seem predicate
which specifies whether the experiencer appears
in the medial or final position.
Beyond c-command facts, there is additional
evidence that such an articulated structure for
seem may be required. An anonymous reviewer
comments that the opening of potential adjunction
sites is a common motivation for binarism over
ternary structures in TAG-based syntax. In this
case, neither the seem-headed tree of Figure 1 or
3 will account for the position of a VP-adjoined
manner or temporal adjunct modifying the raising
predicate:
(8) a. John seems for all intents and pur-
poses to be a professor to me.
b. John seemed for as long as we knew
him to like coffee.
Assuming these adjuncts to be introduced through
elementary trees recursive on VP, only the pres-
ence of the lower VP node in the shell structure
allows for an adjunction into the seem auxiliary
which yields the correct string order. Indeed, (8b)
may indicate that the shell structure is required
even in cases where there is no experiencer.
4 Extending the Analysis
Thus far, this discussion has been limited to cases
in which seem is adjoined into an infinitival clause.
There are at least two other types of structure on
which this analysis needs to be tested: those where
seem adjoins into a small clause, and those where
seem takes a finite clause complement:
(9) a. John seems happy.
b. It seems that John likes coffee.
In exploring these cases, a further challenge to the
ditransitive-style analysis arises. While the expe-
riencer is licit in both positions where the seem-
headed tree is adjoined into an infinitival clause,
apparent asymmetries can be noted in these other
constructions, calling into question the broader ap-
plicability of the structures in Figure 6. Where the
seem auxiliary has adjoined into a small clause,
the experiencer is degraded in the position imme-
diately following seem, and is more acceptable in
the sentence-final position, as in (10). Conversely,
in the finite complement case, the experiencer is
marginal at best in the sentence-final position, il-
lustrated in (11).
(10) a. ? John seems to me happy.
b. John seems happy to me.
(11) a. It seems to me that John likes coffee.
b. ? It seems that John likes coffee to
me.
However, it has been pointed out (Tatjana Schef-
fler, p.c.) that considerations of phonetic weight
may be at work in these cases. For the small clause
cases, replacing the simple adjective with a more
complex element yields a more comfortable sen-
tence with the medial experiencer, and the experi-
encer in final position now seems more awkward:
(12) a. John seems to me competent enough
to finish the task at hand.
b. John seems competent enough to fin-
ish the task at hand to me.
The same reversal can be observed with the finite
clause cases where a heavier experiencer appears
alongside the complement clause. The sentence
final experiencer is made to seem much more nat-
ural than in the simpler case above:
(13) a. It seems to all of the cafe’s customers
that John likes coffee.
b. It seems that John likes coffee to all of
the cafe’s customers.
Taking this into consideration, these apparent
variations are nothing more than red herrings, with
the relative positioning of experiencer and embed-
ded material demonstrating sensitivity to consid-
erations of phonetic weight. Such considerations
may determine which seem-headed auxiliary is the
better choice for native speakers in a given context.
163
Furthermore, difficulties in the case of (11b)
may be a function of ambiguity. An alternative
derivation does exist in which the PP to me is not
an argument of seem. Recalling the cases where
a “pseudo-”experiencer appeared without an ac-
companying raising predicate, it is possible that
the to me of (11b) and to all the cafe’s customers
of (13b) are adjuncts to the embedded clause VP,
in the same pattern as (3b). Extraction tests along
the lines of those employed earlier can be used to
show that the experiencer can be an argument, but
this still will not negate the fact that a derivation
exists wherein it may simply be an adjunct.
5 Conclusion and Implications
With the elimination of challenges to this new
analysis of seem, the conclusion is that the struc-
tures in Figure 6 are justified, and generalisable to
many uses of the verb. Potential counterexamples
are either functions of weight considerations, or
interference from ambiguous analyses.
Having used extraction-based tests to reach this
conclusion, it is worth noting that accounting for
extraction from the seem auxiliary tree remains a
problem for TAG (Frank, 2002). A Wh-question
formed through the extraction of the experiencer
argument would necessarily be extended all the
way to CP, thus sacrificing recursivity. While this
problem has not been solved here, the refinements
to the structure of seem will contribute to future
accounts. Specifically, any account of extraction
which is sensitive to issues such as superiority or
crossover will benefit from this analysis. Consider
the sentences in (14):
(14) a. Bill seems to Johna0 to like hima0 .
b. Bill seems to like him
a4
a0 to Johna0 .
c. To whoma0 does Bill seem to like hima0 ?
In theory, either of (14a) or (14b) could repre-
sent the underlying structure of (14c). Binding,
as shown in (14c), is possible for this question,
though only the (14a) sentence shows equivalent
binding. Extraction of the experiencer in the
(14b) case would result in a weak-crossover vio-
lation, should the extracted experiencer bind the
embedded object. This asymmetry between (14a)
and (14b) would not be predicted by a ternary-
branching analysis, but is captured by the struc-
tures in Figure 6. These sorts of alternations, and
their implications, will need to be kept in mind as
further work on extraction from raising predicates
progresses.

References
Robert Frank. 2002. Phrase Structure Composition
and Syntactic Dependencies. Cambridge, MA: MIT
Press.
Heidi Harley. 2002. Possession and the double object
construction. Linguistic Variation Yearbook, 2:29–
68.
Ray Jackendoff. 1990. On Larson’s treatment of
the double object construction. Linguistic Inquiry,
21(3):427–465.
Anthony Kroch and Aravind Joshi. 1985. The linguis-
tic relevance of Tree Adjoining Grammar. Technical
Report MS-CS-85-16, Department of Computer and
Information Sciences, University of Pennsylvania.
Richard Larson. 1988. On the double object construc-
tion. Linguistic Inquiry, 19(3):335–391.
Dennis Ryan Storoshenko. 2006. Seems like a dou-
ble object. In Proceedings of the 22a0 a1 NorthWest
Linguistics Conference.
