“ID Tracking Model” Replaces “Billard Ball Model” of Conceptualization
Kow KURODA and Hitoshi ISAHARA
National Institute of Information and Communications Technology (NICT), Japan
1 Overview
This work proposes “ID Tracking Model” (IDTM for short) of conceptualization, which is developed by Kuroda [5] and Kuromiya [6] to replace the Billiard Ball Model (BBM) of conceptualization proposed by Croft [1] and Langacker [7, 8]) independently. Main motivation for this is to avoid the arbitrariness in the diagramming convention of Cognitive Grammar (CG).
1.1 Basic Assumptions of IDTM
Unlike BBM, IDTM assumes the following, among others:
(1) Conceptualization is state-based rather than motion- or force-based.
(2) a. Lexical realization is based on profiling: no lexical item can be realized without getting profiled; and no strong profile cannot es- cape from lexical realization. So, profiling is strictlylexically-basedin that it needs to be constrained to avoid arbitrary specifica- tion of semantic structure.
b. Profiling has degrees (e.g., none, weak, strong), and conceptual entities with strong profiled get lexically realized.
(3) Profiling doesn’t determine perspective. In- stead, perspective constrains profiling.
(1) means thatstates are primitives of conceptu- alization, which a conceptualizer is always keeping track of under the same ID (hence the name ID track- ing). This implies that the idea of “energy flow”
is inadequate to characterize basic properties of hu- man conceptualization.
(2) means that relational terms like verbs and prepositions don’t profile their argument nominals strongly enough for them to be lexically realized.
More specifically,saw, as an inflected verb, andAl- ice saw Bill, as a tensed clause, have different pro- file configurations: in saw, the subject and object nominals are not profiled enough, whereas inAlice saw Bill, they are all profiled. A clause “inherits”
profiles from argument nominals, unlike the pro- posed diagramming convention in CG framework.
In IDTM, relationals are not powerful enough to de- termine the profile configuration of a given clause.
In a sense, this is one way the idea of “parallel dis- tributed semantics” is implemented. I will return to this issue in§1.2.
(3) means that the profiling is not enough to spec- ify perspective, and it needs to be “selected” for in- dependent reason.
1.2 Pr´ecis de Parallel Distributed Semantics
B B
A
B
A
A
Figure 1: EncodingAandBwith the “trace” effect
To make the point in (2b) clearer, let me take for an example, decomposition of a complex unitCinto unitsAandB, on the one hand, and composition, or
“integration,” ofAandBintoC, on the other. Let
“C⇒ {A,B}” denote decomposition process, and
“C⇐ {A,B}” denote composition process. If “C= {A,B}” is used, the difference is neutralized.1)
IDTM adds one important assumption on mental representation of A and B to make them realistic:
1)Use of bracketing like [. . . /C] = [ . . . [ . . . /A] . . . [. . . /B] . . . ] does no justice here, since there is no inherent temporal order betweenAandB. IDTM is careful not to make a “secret interpre- tation” of the spatial arrangement of the elements at the semantic pole into the linear order of them at the phonological pole. This is one of the tricks that makes the CG framework ineffective.
1
M(t) M(t´)
T T0
t t´
z(t) p
q
r
z(t´) s1
s2
s3
p´
q´
r´
x GIVE1 y TO z (primary profile on the {x, y} plane (carnation); secondary profile (orchid) on the {y, z} plane) PROFILED RELATIONS:
{u, p, p´} corresponds to: x(t) GIVE1 y(t) (primary)
{p} corresponds to: x(t) HAVE y(t) {p´} corresponds to: NOT x(t´) HAVE y(t´) {s1} corresponds to: x(t) R (primary) {s2} corresponds to: y(t) MOVE (primary}
{w*} corresponds to: y(t) TO z(t) (secondary)
UNPROFILED RELATIONS:
{v} corresponds to: x(t) CAUSE z(t) {w} corresponds to: z(t) RECEIVE y(t) {u*} corresponds to: y(t) FROM x(t) {s3} corresponds to: z(t) R w
w*
u v
u*
v*
x(t)
y(t)
x(t´)
y(t´)
Figure 2:IDTM characterization of (5)
their representations have “traces” of their original environment, i.e., C, in which they occurred. This means that units like Aand B “remember” where they were. This is the trace effect, applying an analogy with Hintzman’s “multiple-trace memory”
model [4].2)
To denote the trace effect in the composition ofC out ofAandB, digrammed in Fig. 1, IDTM assumes the following notation:
(4) C={ {A,B},{A, B}}
WhileAandBare fully profiled units, A and B are not; they are “underprofiled.”
Note that A and B, which are underprofiled, en- code “elaboration sites” for A and B, respectively.
Thus, the notation in (4) largely dispenses with the
“correpondence” links that play a crucial role in the CG-style diagrams.
Consider the representation ofsawagain. Its rep- resentation is [SsawO], which has two underpro- filed elaboration sites,SandO. This is distinct from Alice saw Bill, whose representation is [Alice saw Bill], where elaboration sites are linked to two pro- filed lexical items,AliceandBill. [AliceV] and [S VBill].
2 IDTM Analysis of GIVE
For illustration, IDTM characterizes the semantics of (5) and (6) as in the diagrams in Fig. 2 and Fig. 3, respectively. Note: colors in the diagram have no significance.
(5) xGIVEyTOz
(e.g., [xAndy] gave [yhis bicycle] to [zCarol])
2)Another source of this conception is the idea of “wickel- phones” discussed in Rumelhart and McClelland [9].
(6) xGIVEz y
(e.g., [xAndy] gave [zCarol] [yhis bycycle])
2.1 Interaction among states
M(t), M(t0)are “stative models” that represent the state configurations of the worldWat timingstand t0. Each model is populated by entities like{x(t), y(t),z(t), . . .}, wherex(t)denotes the state ofx at timet.
In IDTM, states interact rather than not entities.
State interactions like{p,q, . . . ,u,v, . . .}are candi- dates for lexical realization via profiling.
2.2 Different perspectives result in dif- ferent argument structures
Two “planes,” one for primary and another for sec- ondary, are selected out of three possible planes.
This choice determines perspective. This accounts for the semantic difference between (5) and (6).
Perspective is used here roughly in the same sense as Fillmore’s [2]. He characterizes the “commer- cial transaction” event in this way. Different per- spectives are cast on the one and single configura- tion of semantic roles such as hbuyeri, hselleri, hcosti,hgoodsi. While relying on the idea of se- mantic frame, this is a different approach to argu- ment structures than Goldberg’s [3] more mapping- oriented model.
3 Concluding Remarks
This paper showed that ID tracking model (IDTM) of conceptualization is a viable alternative to Billiard Ball Model of it. IDTM is also a promising model in which to specify “constraints” on profiling in some
2
M(t) M(t´)
T0 T
t t´
y(t) p
q
r*
y(t´) s1
s2
s3
p´
q´
r*´
x GIVE2 z y (primary profile (carnation) on the {x, z} plane); secondary profile (orchid) on the {x, z} plane
PROFILED RELATIONS:
{v, q, q´} corresponds to: x(t) GIVE2 z(t) (primary)
{q}, {q´} correpond to: ??
{s1} corresponds to: x(t) R (primary) {s3} corresponds to: z(t) R (primary) {w} corresponds to: z(t) RECEIVE y(t) (secondary)
{r*} corresponds to: NOT z(t´) HAVE y(t´) (secondary)
{r*´} corresponds to: z(t´) HAVE y(t´) (secondary)
UNPROFILED RELATIONS:
{p} corresponds to: x(t) HAVE y(t) {p´} corresponds to: NOT x(t´) HAVE y(t´) {u} corresponds to: x(t) LOSE y(t)?
{w*} corresponds to: y(t) TO z(t) {s2} corresponds to: y(t) MOVE {u*} corresponds to: y(t´) FROM x(t) (ternary)
w w*
u v
u*
v*
x(t)
z(t)
x(t´)
z(t´)
Figure 3:IDTM characterization of (5)
natural way. This is desirable because the effects of profiling don’t seem to be adequately constrained in the Cognitive Grammar framework despite its huge theoretical significance. Also, IDTM links conceptu- alization to semantic frames, and is hoped to inte- grate Cognitive Grammar and Frame Semantics in a natural and effective way.
References
[1] Croft, W. (1991).Syntactic Categories and Grammatical Relations. University of Chicago Press.
[2] Fillmore, C. J. (2003). Topics in lexical semantics. In C. J. Fillmore,Form and Meaning in Language, 201–60.
CSLI Publications.
[3] Goldberg, A. (1995). Constructions. University of Chicago Press.
[4] Hintzman, D. (1986). “Schema abstraction” in a multiple-trace memory model. Psychological Review, 93(4), 411–28.
[5] Kuroda, K. (2004). Proposing the “ID tracking Model”
of conceptualization.Proceedings of the Japanese Cogni- tive Linguistics Association,5, 1–11.
[6] Kuromiya, K. (2004). Effectiveness of the ID Tracking Model.Proceedings of the Japanese Cognitive Linguistics Association,5, 12–22.
[7] R. W. Langacker (1987, 1991).Foundations of Cognitive Grammar, Vols. 1 and 2. Stanford University Press.
[8] Langacker, R. W. (2000).Grammar and Conceptualiza- tion. Mouton de Gruyter.
[9] Rumelhart, D. E., and J. L. McClelland (1987). Learn- ing the past tenses of English verbs: Implicit rules or parallel distributed processing? In B. J. MacWhin- ney (Ed.),Mechanisms of Language Acquisition, 195–248.
Hillsdale, NJ: Lawrence Erlbaum Associates.
3
