THEME SESSION
International Cognitive Linguistics Conference 2003, La Rioja, Spain
keywords: Embodied construction grammar, construction grammar, embodiment, computational models
Benjamin Bergen (contact person)
Department of Linguistics
University of Hawaii
George Lakoff
Department of Linguistics
University of California, Berkeley
Embodied Construction Grammar (ECG) is a recent paradigm formulated within the Neural Theory of Language. It combines insights from the two major grammatical theories used by the cognitive linguistics community - Cognitive Grammar and Construction Grammar – and integrates them, along with other cognitive mechanisms such as image-schemas, frames, metaphor, mental spaces, and blending, in a neurally plausible computational formalism.
Like cognitive grammar, ECG views linguistic knowledge as composed of sound-meaning pairings that operate at varying levels of formal and semantic specificity. Like standard Construction Grammar, ECG is expressed in an explicit formalism usable in Natural Language Processing. Like Cognitive Grammar, ECG makes use of notions of profiling, semantic and formal connections between constituents, and dynamic mental simulation.
ECG differs from these models in several ways, all of which pertain to its "embodiment".
First, its formalism has been constructed alongside a detailed mapping from that formalism to neural computational structures. Embodiment is carried out via such neural grounding. Second, ECG has been designed to form the core of computational language understanding and production systems, and as such must be explicit enough to serve purposes of language processing rather than simple linguistic knowledge. Third and finally, ECG incorporates cognitive linguistic mechanisms like image-schemas, frames, metaphor, mental spaces, and blending into its grammatical structures, allowing the different mechanisms to interface naturally.
The aims are this theme session are twofold: to unify the theoretical and formal mechanisms of ECG, and to make public research that has already been done in this framework, including studies of morphology, reference, binding, language acquisition, force-dynamics, and others.
We propose to hold the session over the course of a full day.
|
9:00-9:30AM |
Introduction to Embodied Construction Grammar |
Benjamin Bergen |
|
9:30-10:00AM |
There we go again: Embodied Construction Grammar Meets the Deictic Construction |
Nancy Chang |
|
10:00-10:30AM |
An Embodied Construction Grammar account of English reflexive pronouns |
Keith Sanders |
|
10:30-10:45AM |
Break |
|
|
10:45-11:15AM |
Exiting Events in English and Spanish: Some Empirical Evidence from Second Language Acquisition. |
Carmen Bretones and Teresa Cadierno |
|
11:15-11:45AM |
Force-dynamics: an Embodied Construction Grammar account |
Ellen
Dodge |
|
11:45-1:00PM |
Lunch |
|
|
1:00-1:30PM |
From Set Theory to X-Schemas: Why Aspectual Coercion is Aspectual [Powerpoint presentation] |
Laura Michaelis |
|
1:30-2:00PM |
An Embodied Construction Grammar account for Georgian Verbal Morphology |
Olya Gurevich |
|
2:00-2:30PM |
The Semantics of Verbal Reduplication in Embodied Construction Grammar: A Cognitive Explanation |
Jenny Lederer |
|
2:30-2:45PM |
Break |
|
|
2:45-3:15PM |
Applying Embodied Construction Grammar: a description of some Afrikaans morphological constructions [Powerpoint presentation] |
Gerhard van Huyssteen |
|
3:15-3:45PM |
Constructional Analysis with Embodied Construction Grammar |
John Bryant |
|
3:45-4:15 |
Construal Resolution with Embodied Construction Grammar |
Robert Porzel and John Bryant |
|
4:15-4:30 |
Break |
|
|
4:30-5:15 |
Comments by discussants |
George Lakoff, Hans Boas, and Adele Goldberg |
Discussants’ topics:
Abstracts are presented in alphabetical order by first author below
Benjamin Bergen
Embodied Construction Grammar (ECG) is a recent paradigm that combines insights from the two major grammatical theories used by the cognitive linguistics community – Cognitive Grammar and Construction Grammar, and integrates them, along with other cognitive mechanisms such as metaphor, mental spaces, and blending, in a neurally plausible computational formalism. This talk presents an outline of the basic concepts and mechanisms of ECG, as well as its formalism. The goal of this talk is be to provide a common basis for the rest of the talks in this theme session.
Like cognitive grammar, ECG views linguistic knowledge as composed of sound-meaning pairings that operate at varying levels of formal and semantic specificity. Like standard Construction Grammar, ECG is expressed in an explicit formalism usable in Natural Language Processing. Like Cognitive Grammar, ECG makes use of notions of profiling, semantic and formal connections between constituents, and dynamic mental simulation.
ECG differs from these models in several ways, all of which pertain to its "embodiment".
First, its formalism has been constructed alongside a detailed mapping from that formalism to neural computational structures. Embodiment is carried out via such neural grounding. Second, ECG has been designed to form the core of computational language understanding and production systems, and as such must be explicit enough to serve purposes of language processing rather than simple linguistic knowledge. Third and finally, ECG incorporates cognitive linguistic mechanisms like image-schemas, frames, metaphor, mental spaces, and blending into its grammatical structures, allowing the different mechanisms to interface naturally.
The ECG formalism differs radically from those of Construction Grammar and Cognitive Grammar, and this presentation will provide a general introduction to it.
Exiting Events in English and Spanish:
Some Empirical Evidence from Second Language Acquisition.
Carmen Bretones and Teresa Cadierno
Language comprehension and language generation are possible thanks to a set of constructions or form-meaning pairs (Goldberg 1995, Fillmore et al.1988, Fillmore and Kay 2000, Kay and Fillmore 1999, Lakoff 1987; Bretones and Robles, forthcoming). According to Embodied Construction Grammar (Bergen et al. 2001, Bretones et al. 2001) those constructions are elaborated by means of the interaction of semantic schemas and simulation-based inference based on bodily grounded structures: image schemas (Johnson 1987, Lakoff and Johnson 1999) and executing schemas (Narayanan 1997) or structures that carry out dynamic processes. Since the main goal of this study is to analyse the mental images that speakers construe when it comes to understand a construction, at first we will focus basically on the specifications that refer to the constituents of this construction and its semantic constraints. In a contrastive analysis between English and Spanish we will analyse "exiting event constructions" such as "out-of" and "salir-de". Both the English and the Spanish constructions involve two I-schemas; Trajector-Landmark (TL) schema and Source-Path-Goal (SPG) schemas. However, they differ with respect to their third schema; whereas in English we find a container schema, in Spanish we find a boundary schema (Bergen et al 2001; Bretones et al. 2001:12). The Spanish construction contains the following semantic constrains: (1). the source of the SPG-schema is bound to the landmark; (2).the landmark is bound to the region A of the B-schema; (3).the goal of the SPG schema is bound to region B of the B-schema (Bretones et al. 2001), and the crucial point in the construction is the semantics of the landmark given by the different types that can occur. This allows a broader range of landmark instanciation in Spanish than in English. (Bretones et al.2001:6).
The
contrastive analysis of how exiting events are conceptualised in English and
Spanish has clear implications for the study of adult Second Language
Acquisition (SLA) (Cadierno and Lund, in press; Cadierno forthcoming). More
specifically, this analysis allows for positing theoretically-motivated
hypotheses with respect to how English learners of Spanish and Spanish learners
of English will acquire the corresponding exiting construction in their foreign
language. Given the crucial differences in the image-and executing schemas
involved in the two languages, it is hypothesized that the English native
speakers will initially tend to judge as grammatically correct Spanish
sentences where the landmark of the exiting event only consists of a container;
on the other hand, Spanish native speakers will initially tend to judge as
grammatically correct English sentences where the landmark of the exiting event
consists of either a container, a region or a portal. The same tendencies are
also hypothesized to be operant when learners are required to produce L2
sentences where an exiting event is involved. In other words, English native
speakers will initially tend to judge and produce a more restricted set of
exiting constructions in Spanish than the native speakers of this language.
Spanish native speakers, on the other hand, will initially tend to judge and
produce a less restricted set of exiting constructions in English than the
native speakers of this language. In order to test these hypotheses, a
bi-directional empirical investigation will be reported where beginner English
learners of Spanish and beginner Spanish learners of English will be given two
tasks: (1) a grammaticality-judgement task where they will have to judge the
extent to exiting constructions in the L2 are grammatical; and (a) a production
task where they will have to describe in the L2 a series of pictures depicting
exiting events. The results obtained can validate the theorethical analysis of
the different mental images involved in the two constructions and can further
provide support for the importance of investigating SLA from a cognitive
perspective.
Bergen,
Benjamin K., Chang, Nancy C. & Paskin, Mark A. 2001.
“Simulation-Based Language Understanding in Embodied Construction
Grammar”. http://www.icsi.berkeley.edu/techreports. In press; J.
Östman (ed.). Construction Grammar(s): Cognitive and Cross-language
dimensions.
Bretones, C. ;
Cristóbal, M. and Ibarretxe,
I. 2001. “The Construction Salir-de in Spanish. How Spanish Speakers Conceptualize Exiting Events.”
Berkeley: International Computer Science Institute Technical Reports 2001:
tr-09-01. http://www.icsi.berkeley.edu.
In press; in Hans Boas Jr. and Mirjam Fries (eds). Construction Grammar:
back to the roots. John
Benjamins.
Bretones, C. and Robles, Adela. Forthcoming. “Échale Guindas a Pavo: Two ways of constructing the destination in Spanish.” International Conference on Adpositions of Movement Procedings. Catholic University of Leuven.
Cadierno, T. and Lund, K. In press. "Cognitive linguistics and second language acquisition: Motion events in a typological framework." In B. VanPatten, J. Williams and S. Rott (Eds.), Form-meaning connections in second language acquisition: Selected proceedings. Philadelphia: Lawrence Erlbaum.
Cadierno, T.
Forthcoming "Expressing motion events in a second language: A cognitive
typological perspective." In M. Achard & S. Neimeier (Eds.), Cognitive
linguistics and second language acquisition. Berlin: Mouton de Gruyter.
Fauconnier,
G. and Turner, M. 2002. The Way We Think.
Cambridge: CUP.
Fillmore, C. J. 1982. “Frame Semantics”. In Linguistic Society of Korea (ed). Linguistics in the Morning Calm. Seoul: Hanshin, pp. 111-138.
Fillmore, Charles J. 1985. “Frames and the Semantics of Understanding.” Quaderni di Semantica 6: 2(1985). Pp. 222-254.
Fillmore, C. J. and Kay, P. 1995. Construction Grammar. Stanford: CSLI (forthcoming).
Fillmore, C.J; Kay, P. & O’Connor, M.C.
1988. Regularity and Idiomaticity in Grammatical Constructions: The Case of Let
Alone. Language 64(3):
501-538.
Goldberg, A. 1995. Constructions. A Construction Grammar Approach to Argument Structure. Chicago: University of Chicago Press.
Johnson, Mark. 1987. The Body in the Mind. The Bodily Basis of Meaning, Reason, and Thought. Chicago: University of Chicago Press.
Lakoff, George. 1987. Women, Fire and Dangerous Things: What Categories Reveal about the Mind. Chicago: University of Chicago Press.
Lakoff, George & Mark Johnson. 1999. Philosophy in the Flesh. The Embodied Mind and its Challenge to Western Thought. New York: Basic Books.
Langacker, Ronald W. 1987. Foundations of Cognitive Grammar: Theoretical Prerequisites. Vol.1. Stanford: Stanford University Press.
Narayanan, Srini. 1997. Knowledge-based Action Representations for Metaphor and Aspect (KARMA). Doctoral dissertation. University of California at Berkeley. http://www.icsi.berkeley.edu/NTL/publications.html.
Talmy, Leonard. 2000. Towards a Cognitive Semantics. Massachusetts: MIT Press.
John Bryant
Embodied Construction Grammar (ECG) (Bergen and Chang, 2002) supports a new computational approach to cognitively motivated semantic analysis. Historically, there has been a gap between grammar formalisms precise enough to be used for parsing and those used in cognitive approaches to language. ECG bridges this gap by providing a formalism that marries the precision of unification grammar with the expressiveness of cognitive semantics. While there are many linguistic issues involved, this paper focuses on computational concerns, particularly those that arise in constructional analysis.
The
basic linguistic unit of ECG is the construction, a pairing of form and
meaning. On the form side, ECG allows for more expressive constraints, dropping
the Phrase Structure Grammar requirements for strict constituent ordering. On the meaning side, ECG uses
cognitively motivated schemas, including image schemas and executing schemas.
These schemas are represented as feature structures with meaning constraints
modeled as coindexation constraints.
ECG
also extends standard unification-based formalisms with several novel operators
that provide additional semantic
expressiveness. The evokes operator makes
related meanings accessible, thus allowing background frames and preconditions
to be bound to the meaning pole.
The self operator allows
constructions to be self-referential (prohibited in traditional unification
grammar) and the ::
operator allows for conditional or dynamic unification that facilitates dealing
with events.
Since
ECG is more expressive than traditional grammar formalisms, more flexible
approaches to parsing must be developed. Chart parsing is still useful, but
instead of a single monolithic parser keeping track of the state of every rule,
each grammar rule is compiled into a chunk of active knowledge - the recognizer
for that construction. When
a recognizer checks the form constraints, it also checks the meaning pole,
requiring that the meanings associated with each constituent be consistent, not
only with each other, but also with the semantic constraints imposed by the
construction.
The
recognizers are arranged into levels (Abney 1996). Each level combines the
results of the previous levels, putting smaller analyses together, thus producing
a set of spanning constructional analyses. The set of
interconnected schema instances that comprise the shared meaning pole of
such an analysis is known as the semantic specification
(semspec).
ECG's
expressiveness incurs an increased computational burden on the analyzer, but
the rich semantic representation can be exploited to offset this cost. Specifically, the early pairing of form
and meaning allows the analyzer to score the partial analyses in the chart
using a heuristic we call semantic density. Semantic density scores constructional
analyses based upon how well the recognized constructions bind up the roles in
the semspec. This heuristic
leverages the increased semantics for greater accuracy and faster performance,
and provides a means for dealing with analyses in which no single construct
spans the input.
The
implemented analyzer was tested on a subset of the CHILDES corpus (MacWhinney
1998). While every utterance
licenses some set of constructional analyses, using the semantic density metric provides encouraging
results for winnowing out less complete analyses, thus showing that when aided
by cognitive semantics, a language analyzer can converge more quickly onto
better analyses.
References:
Abney, Steven. Partial
Parsing via Finite-State Cascades. In Proceedings
of the ESSLLI '96 Robust Parsing Workshop. 1996.
Bergen, Ben and
Chang, Nancy. Embodied Construction Grammar in Simulation-Based
Language Understanding. Technical Report TR-02-004, International Computer
Science Institute. 2002.
MacWhinney, Brian, The
CHILDES system, in Handbook of child language acquisition, edited by
W. Ritchie and T. Bhatia (New York: Academic Press, 1998), 457-94.
There we go again: Embodied Construction
Grammar Meets the Deictic Construction
Nancy Chang
Lakoff's [1]
discussion of the Deictic (or `there'-) Construction remains a landmark
demonstration of the extent to which semantic and pragmatic factors can
motivate relations among constructions. The relevant class of expressions is
analyzed in terms of a family of related constructions, whose central case --
exemplifed by "There goes Harry" -- is characterized as pairing a
particular sequence of constituents with a discourse scenario in which the
speaker points out to the addressee the existence or motion of some third
(visually) perceptible entity. Diverse specializations of the construction --
e.g., "There's Harry with his red hat on", "There goes the
bell!", "Here's your pizza" -- involve a wide array of seemingly
arbitrary constraints on both form and meaning. Motivations for these constraints include the underlying
speech act; metaphorical mappings from physical motion to domains like
perception, discourse, and existence; and more specific conditions on the pragmatic
situation and narrative genre.
From
a theoretical standpoint, construction grammar is an ideal framework for
describing the various form and meaning correlations involved in Deictic
Construction(s). In practice, however, construction grammarians have not
converged on a consistent set of formal tools expressive enough to capture the
requisite range of complex interactions among diverse conceptual
apparatus. Recent efforts
addressing such representational issues have led to the development of Embodied
Construction Grammar (ECG) [2,3], a computationally explicit formalism for
describing the conceptual underpinnings of analyses in the construction grammar
and cognitive linguistics literature [4,5,6,7]. ECG provides a common representational language for
conceptual structures (or schemas), constructions, metaphorical/metonymic maps
and mental spaces, as well as flexible means of expressing relations among
these structures. Moreover, the formalism is situated within a broader effort
addressing the embodied basis of language acquisition and use, in which
utterances are understood in terms of active, context-sensitive simulations of
embodied schemas. It thus includes precise ways of expressing how schemas and
constructions can interact with situational context and ontological knowledge
and what dynamic constraints hold at different event stages.
This
paper treats the Deictic Construction and its variants as a challenge to the
representational adequacy of ECG. The ECG version of the Central Deictic
Construction draws on key features of the formalism: constituency is decoupled
from word order relations (which here depend on whether a pronoun is used to
refer to the focused entity); dependencies across constituents are allowed
(here, to enforce subject-verb agreement and consistency in deixis markers);
and contextual conditions (on, e.g., the attentional states and visual fields
of both speaker and addressee) are cleanly integrated into the
interpretation. The special cases
impose even more demanding constraints that push the limits of the formalism,
e.g. constraints on stress, elongation, intonational contours, aspectual phase,
and perspective (see also [8]). The techniques used in this case study should
not only facilitate the analysis of other complex families of constructions,
but also demonstrate how the sharper representational tools provided by ECG
allow more precise questions on the nature of constructional relations to be
asked (and answered).
[1]
George Lakoff. 1987. Women, Fire, and Dangerous Things: What
Categories Reveal about the Mind.
University of Chicago Press.
[2] Benjamin K. Bergen and Nancy C. Chang. Forthcoming. Simulation-based language
understanding in Embodied Construction Grammar. In Jan-Ola Ostman (editor), Construction Grammar(s):
Cognitive and Cross-language dimensions.
John Benjamins.
[3] Nancy Chang, Jerome Feldman, Robert Porzel,
and Keith Sanders. 2002. Scaling
cognitive linguistics: Formalisms for language understanding. Presented at the First International
Workshop on Scalable Natural Language Understanding.
[4] Adele E. Goldberg. 1995.
Constructions: A Construction Grammar Approach to Argument
Structure. University of Chicago
Press.
[5] Charles Fillmore. 1988. The
mechanisms of construction grammar.
In Berkeley Linguistics Society, volume 14, pages 35--55.
[6] Ronald W. Langacker. 1987.
Foundations of Cognitive Grammar, Vol. 1. Stanford University Press.
[7] Gilles Fauconnier. 1985. Mental
Spaces: Aspects of Meaning Construction in Natural Language. MIT Press/Bradford, Cambridge, Mass.
and London.
[8] Nancy Chang, Srini Narayanan and Miriam R.L.
Petruck. 2002. Putting Frames in
Perspective. In Proc. 19th
International Conference on Computational Linguistics.
Force-Dynamics; an Embodied Construction
Grammar Approach
Ellen Dodge
Talmy (1988, 2000) has convincingly shown that force-dynamics is fundamental to human conceptual systems. However, his representation of force-dynamics is largely diagrammatic, and is designed to capture general force-dynamic patterns rather than specific lexical meanings. Embodied Construction Grammar (ECG) allows us to remedy these problems and to provide a more explicit analysis of force-dynamics. This paper includes precise representations of force-dynamic scenarios, incorporating other non-force-dynamic concepts. Both lexical and grammatical constructions are represented. Additionally, it shows how mental spaces (Fauconnier, 1997) can be used to characterize Talmy’s force-dynamic “tendencies”, as well as characterizing crucial aspects of cause-effect relations.
Meaning in ECG presupposes a dynamic simulation semantics. Input to this is in the form of static schemas representing semantic structure. Schemas can have internal aspectual structure, e.g. Initial, Pre-Central, Central, Post-Central, and Final actions and states. Schemas can also characterize mental space structure. Force-dynamic schemas contain roles for at least two entities: one entity (the Force-Source) is the source of an active force and the other (the Force-Target) provides resistance. The Force-Source has, in Talmy’s terms, a tendency toward action and the Force-Target has a tendency toward rest. Tendencies can be represented in ECG using default expectation spaces and counterfactual minimal-difference spaces in which an entity with a tendency to move does move, and an entity with a tendency to rest does rest. At the heart of every force-dynamic interaction, in the Central part of the schema’s aspectual structure, is an Impingement relation between the Force-Source and the Force-Target. In the Post-Central part, each entity undergoes the effects of impingement, e.g. the Force-bearer undergoes resistance to motion and the Force-Target undergoes pressure or strain at the contact area. Effects may follow from these experiences, such as motion initiation or directional change. Ultimate effects are represented in the Final state.
In (1) and (2), impingement causes the undergoer of the force to be affected.
(1) He bent the pole.
(2) The armor deflected the arrow.
These examples can be represented using a Cause-Effect schema containing two spaces. In the Reality space, the force-interaction schema describes the actual scenario. In (1), the pole undergoes pressure, and its shape changes. In (2), the arrow undergoes resistance to motion, and its trajectory direction changes. In the Counterfactual Minimal-Difference Space, Impingement is not present and the effects don’t occur. Instead, the tendencies are manifested, e.g. the pole retains its original shape, and the arrow continues along its original trajectory.
Because this formalism is not specific to force-dynamics, schemas from other conceptual domains can be represented as well. For example, motion schemas can specify manner or path of motion, and can represent an entity’s motion tendency or motion caused by force-dynamic interaction. Additionally, parameterization of forces allows for representation of distinctions between open-class lexical items specifying forces of different directions (e.g. push, pull, turn), or magnitude (e.g. nudge, shove). Force-dynamic interactions resulting in object deformation can include specifications related to force direction (as in squeeze or twist) and resultant deformation (e.g. crush, flatten).
In this way, ECG allows us to integrate a more complete and precise description of force-dynamics into cognitive linguistic descriptions as a whole.
References:
Faucconier, Gilles. 1997. Mappings in Thought and Language. New York: Cambridge University Press.
Talmy, Leonard. 2000. Toward a Cognitive Semantics. Massachusetts: MIT Press.
Talmy, Leonard. 1988. Force Dynamics in language and thought. In Cognitive Science: 12, 49-100.
Olya Gurevich
Embodied Construction Grammar (ECG) is a framework built to be compatible with human cognitive and neural mechanisms, as well as child language acquisition principles. As such, it should be able to describe the full range of data for a given language. This includes form, as well as meaning, relationships in a language. Paradigmatic morphology is a linguistic phenomenon where formal relationships between words may not correspond to relationships in meaning. I will present an analysis of Georgian verbal morphology within the ECG framework, demonstrating how a usage-based approach can resolve many of the traditional problems in the analysis of Georgian.
Georgian verbal morphology is notoriously complicated, as each verb can potentially agree with the subject, the object, the indirect object, and the beneficiary of an action. Each verb can appear in a variety of tenses, which fall into one of three possible valence patterns (called series): nominative-accusative (1), ergative-nominative (2), and dative (subj)-nominative (obj) (3). The subject and object agreement markers appear in the post-stem and pre-stem positions, respectively, for the first two valence patterns. The tenses in 3rd series are most often associated with an evidential meaning. The agent appears in the dative and the patient – in the nominative. However, the verbal agreement is based on case of the nouns rather than grammatical relations, i.e. the same markers indicate the subject in 1st and 2nd series, but the object in 3rd series.
In all cases, verbal agreement markers participate in complicated slot competition. In the pre-stem position, a subject marker appears when a form is unmarked for object agreement (3p object forms). In post-stem position, a single suffix -t appears to mark plurality of either the subject or the object, but only when no other parts of the word (i.e. prefixes) signal the difference between singular and plural forms and a more specific plural suffix isn’t available.
The agreement patterns in Georgian morphology pose significant problems for traditional item-and-arrangement approaches, as it is almost impossible to assign unique meaning to each agreement marker. The apparent generalizations are much easier to state if one takes a language learner’s perspective, looking at relationships between forms within a paradigm rather than at individual forms.
Within ECG, such an approach makes use of mappings to represent
relationships between forms, which are organized into networks. Individual exponents mark contrasts
between forms rather than bearing their own meaning. In a network representation of a paradigm, different forms mutually
inhibit each other. Thus it is possible to express the generalization that a
given form means what it means because it stands in contrast to all other
form/meaning pairs within the paradigm, instead of having to sum up the
meanings of the affixes. Moreover,
an ECG-style approach allows representation of local generalizations that only
apply to sub-paradigms. This view
is supported by evidence from child language acquisition: children start out by
using some verbs only in the ergative tenses and others – only in the
nominative-accusative tenses, and generalize to entire paradigms later.
(1) tkven
čven gv-k'lav-t
youPL.NOM we-sg.DAT/ACC 1plObj-kill.PRES-pl
'You
(pl) kill us'
(2) k'ac-ma čven mo-gv-k'la
man.ERG us.NOM preverb-1plObj-kill.AOR
'The
man killed him'
(3) čven k’ac-i mo-gv-i-k’lav-s
we.DAT man.NOM preverb-1plSubj-kill.1RESULT
'We
have apparently killed the man'
References:
Bergen, Benjamin K. and Nancy C. Chang. 2002. “Embodied Construction Grammar in Simulation-Based Language Understanding.” Technical Report TR-02-004, International Computer Science Institute. <www.icsi.berkeley.edu>
Hewitt, B.G. 1995. Georgian: A Structural Reference
Grammar. Amsterdam/Philadelphia: John
Benjamins
Applying Embodied Construction Grammar: a
description of some Afrikaans morphological constructions
Gerhard van Huyssteen
In the development of human language
technologies, linguists often have to communicate the complexities of language
structure to other co-workers with little (or none) linguistic knowledge. For
this declarative purpose, as well as for more procedural purposes (like parsing
and generation), formal rule systems (i.e. formal grammars or formalisms) can
be employed to “treat the empirical contents of grammatical analysis and
the functioning of communication as neutrally as possible” (Hausser,
1999: 125). Embodied Construction Grammar (ECG – see Bergen & Chang,
forthcoming) offers linguists a formalism that is compatible with the
theoretical presuppositions of Cognitive Linguistics.
Since the formalism was originally
designed for purposes of computational natural language understanding,
“principles of semantics and pragmatics” form the backbone of ECG
(Chang, Feldman, Porzel & Sanders, 2002). Most of the applications in the
ECG literature therefore also deal with larger utterances (i.e. syntactic
constructions – see for example Chang & Maia, 2001), but in so far as
I could determine, very little has been done to represent morphological
constructions with the ECG formalism.
In an ongoing project to develop a
spellchecker for Afrikaans based on morphological analysis, various automatic
rule-based and finite-state morphological analysis modules are being developed,
including a stemmer, and word segmenter. Since the linguists in the project
team have to present the non-linguists with a simple, declarative grammar of
Afrikaans morphology, and since all the linguists in the project team have a
good grounding in Cognitive Grammar, we have decided to explore the
possibilities that Embodied Construction Grammar offers to describe
morphological constructions.
In this paper, our first experimental experiences with the ECG formalism will
be presented, focusing on inflectional and derivational processes in Afrikaans.
In view of the fact that our aim is to
represent morphological constructions (and more specifically the
form/phonological pole), we limit our scope to the construction definition
notation of the ECG formalism. In the absence of comparable examples from other
scholars, we have soon come to realise that we need to adapt the ECG formalism
to accommodate some features that we deem necessary for our specific purposes
(for example, the computer scientists in our team require graphemic
alternations to be represented by means of regular expressions, and we
therefore have to include regular expressions in the formalism). It however
remains a question whether these adaptations are of any value to the broader
ECG “community”.
This paper aims to demonstrate the
power of ECG in mediating between linguists and computer scientists. The usefulness of the ECG formalism within
the domain of morphological constructions will also be tested, by illustrating how a few Afrikaans
inflectional and derivational constructions could possibly be represented. I
will also pay special attention to our incorporation of regular expressions in
the formalism (i.e. as form elements and constraints).
References
Bergen, B.K. & Chang, N.C.
forthcoming. Embodied Construction Grammar in Simulation-Based Language
Understanding. In: Östman, J. & Fried, M. (eds.). Construction
Grammar(s): Cognitive and Cross-Language Dimensions.
Amsterdam: John Benjamins.
Chang, N., Feldman, J., Porzel, R. &
Sanders, K. 2002. Scaling Cognitive Linguistics: Formalisms for Language
Understanding. Paper presented at the First International Workshop on
Scalable Natural Language Understanding. Villa
Bosch, Heidelberg, Germany. 23-24 May 2002.
Chang, N. & Maia, T. 2001. Learning
Grammatical Constructions. Presented at the 2001 Meeting of the Cognitive
Science Society. Edinburgh, Scotland. August 2001.
Hausser, R. 1999. Foundations of
computational linguistics. Berlin: Springer.
Jenny Lederer
Reduplication is the phenomenon in which a phonological part of a word or the whole word is repeated in sequence to alter in some way the semantic sense of the lexical item. An example of partial reduplication is the Taizang sentence:
emo e-ke- kep (Abbi 1991:47)
she/he cries cries
‘He/she is crying (continuously).’
A common use of reduplication is to express verbal aspect within the grammar of the language; these categories of aspect can include: simultaneity, non-precipitation, continuation-duration, iteration, sequentiality, inception, and the perfective of an action. Much linguistic work has undertaken the task of describing these semantic-aspectual categories of reduplication (Abbi 1991, Niepokuj 1997), yet the overall treatment of verbal reduplication has been quite superficial, often resulting simply in a list of aspectual categories and a conclusion that the semantics of reduplication are just iconic and arbitrary. I will argue, via the ECG formalism, however, that the meanings of verbal reduplication are not arbitrary, but systematic, and that a simple statement that reduplication is iconic is not a sufficient explanation.
In the ECG formalism, the progressive, iterative, perfective, etc. form of a verb in a particular language is treated as a construction, a dual instantiation of both form and meaning. A construction can derive semantic content to its semantic pole from a conceptual schema, another ECG primitive. In the case of a progressive verbal construction, an internalized X-Schema, or “action sequence”, is automatically activated to aid in the understanding of the various stages of an event, to understand how the ongoing stage of an event is marked by the progressive form of the verb. In the ECG formalism, a construction necessarily has a form, and in the case of certain languages the form activated by the aspectual construction is “reduplication”, unlike other languages where the form is “auxiliary verb”, or “verbal suffix”. In this paper, I will argue, that what separates the “reduplication” form, from the other forms is an additional activation/link to a metaphorical map (mapping), a third ECG primitive. The “reduplication” form activates specifically the LINGUITIC EXPRESSIONS ARE CONTAINERS (Reddy 1979, Lakoff & Johnson 1980) map, a “sub-case” of the CONDUIT METAPHOR. The language user is able to infer from the CONDUIT METAPHOR that if LINGUISTIC EXPRESSIONS ARE CONTAINERS, the containers have content, and if the LINGUISTIC EXPRESSION is multiplied, as in reduplication, the CONTENT is multiplied as well.
The co-activation of this map and the particular aspectual construction allows the language user to understand that a reduplicated verb is linked to the multiplication of a particular stage in an action sequence. For example, if reduplication marks the progressive aspect, then the ongoing stage of the action is multiplied, if reduplication marks iterative aspect than the entire event/action is multiplied.
This ECG analysis will also show how one language can link the same morphological form of reduplication to different aspectual constructions to express different types of aspectual semantics, i.e. two co-existing morphological categories of the iterative and the inceptive. Verbal reduplication will be shown to be a basic process of grammatical formation, whose transparency allows considerable insight into the necessary mechanisms for the semantic structure of aspectual time. The current analysis will draw from examples in different language families that exploit reduplication including Dravidian, Tibeto-Burman, Austro-Asiatic, Indo-European, American Indian languages, and signed languages.
Abbi, A. (1992). Reduplication in South Asian Languages. New Delhi, Allied Publishers Limited.
Lakoff, G. and M. Johnson (1980). Metaphors We Live By. Chicago
London, University of Chicago Press.
Niepokuj, M. (1997). The Development of Verbal Reduplication in Indo-European. Washington D.C., Institute for the Study of Man.
Reddy, M.
(1979). The Conduit
Metaphor. In Ortony, A. (ed.), Metaphor
and Thought. Cambridge,
Cambridge University Press.
284-324.
From Set Theory to
X-Schemas: Why Aspectual Coercion is Aspectual
Laura Michaelis
Speakers conceive of situations flexibly, and use
a variety of morphosyntactic devices to alter the situation-type
representations projected by verb-argument combinations. While such alterations
may be effected by dedicated type-shifting constructions like the Progressive,
they can also be effected via coercion, the triggering of semantic conflict between an aspectual operator
and its situation-type argument (Jackendoff 1997, Moens & Steedman 1988).
Such conflicts require the creation of an enriched representation, as when the
stative predication She like- the new receptionist is combined with the frame adverbial in an
instant, yielding a reading in
which the onset of liking occurred within the denoted frame. Coercion has
generally been modeled by functors which map between sets of situation types.
These models are problematic because they obscure both the internal structures
of the situation types being mapped and the isomorphic properties of the
mappings. Since such mappings ‘swap out’ whatever properties of the
input type conflict with those of the output type, they discard aspects of verb
meaning which would otherwise play a role in semantic composition. One such
mapping, iteration (ITER), derives a state that “describes an unbounded
number of eventualities of the type described by the predicate” (De Swart
1998:383). In what respect does an iterated accomplishment, e.g., She greet-
the customers, qualify as a
state? In Aktionsart classification, a situation consisting of a series of
type-identical subevents, e.g., bouncing a ball, qualifies as a dynamic
situation—an activity. Why then should ITER stativize, if in fact it
does? The same question is raised by the habitual operator HAB, which maps
event descriptions onto state descriptions (ibid: 383). This model of habituality
begs two basic questions. First, why should habitual situations be stative? They do not
qualify as such by their internal composition, which is analogous to that of
iterated events. Second, what aspectual operators trigger the stative
type-shifts which lead, either directly or indirectly, to habitual readings? In
the case of habitual sentences like My officemate smokes and My neighbor went to that gym the only temporal operator is tense, which
apparently is neither an aspectual-class selector nor, concomitantly, an
aspectual-coercion trigger.
To address the foregoing problems, I propose that aspectual type-shifting involves the unification of aspectual x-schemas (Narayanan 1997). Such schemas will be taken to consist of states, transitions and event chains. Unification is constrained by Aktionsart Preservation: input and output schemas must permit overlay with no loss of information. Transparent overlay constrains all operations upon Aktionsart structure, including permutation and concatenation. Permutation adds or selects a single component of the input x-schema. Concatenation outputs event-chain representations. Because a transition is defined relative to a prior or subsequent state, all intervals which adjoin a transition, including those which precede onset transitions or follow offset transitions, are states, which I will refer to as rests. Rests can be selected by permutation, resulting in a stative output schema. Because selection can target any rest—initial, final or intermediate—it finds states within Aktionsart representations where none have been presumed to exist. For this reason, I will show, selection provides a transparent account of a wide variety of stative type shifts, including progressive and habitual/generic construals. By treating aspectual type-shifts as x-schema-based operations, I will argue, we can explain: (1) the relationship between input and output types, (2) constraints upon possible aspectual transitions, (3) the means by which tenses permute Aktionsart representations, and (4) the fact that coerced representations mimic verbal Aktionsart representation, as progressive-form statives (e.g., She’s living in Berkeley) denote episodes of stasis like those denoted by homogeneous activity verbs, e.g., sit, stand, lie.
Construal Resolution
with Embodied Construction Grammar
Robert Porzel and John Bryant
Construal, referring to
processes that resolve "unconventional" meaning, constitutes an
important subject for cognitive linguistics. We claim that embodied construction grammar (ECG) [1]
provides a formalism that is sufficiently precise and well-suited for tackling
this problem formally and computationally. We will show that, ECG provides a
knowledge representation language and mechanism adequate for ontological
construal resolution. Other types of construal depending on discourse-, user-
and situational context are not considered here. Using the examples below, we
describe the implemented construal engine.
In ECG predicates (and type
constraints) can be associated with roles in both constructions and
schemas. A conflicting predicate
or constraint on a role entails that the proposed filler for that role does not
satisfy the predicate or constraint.
Assuming a meaningful utterance, some unconventional usage must therefore
be causing the failure, thereby triggering the construal resolution
process. Consider the following
examples:
(1) "Goethe
often visited the historical museum".
(2) "The
Palatine museum was moved to a new location in 1951".
(3) "The
apothecary museum was renovated in 1983".
(4) "In 1994
the museum bought a new Picasso".
With respect to the word museum, we consider
example (1) to be construal-free (conventional), (2) and (3) to exhibit
polysemy and (4) to exemplify a metonymic construal. We propose to treat
concepts, such as museum, as radial categories, i.e. a
category that evokes more than one conceptual frame [2]. In our model museum, a subtype of public_place, evokes both a building and a collection_of_artifacts frame. The evokes operator
allows us to state that building is an important
concept associated with museums, but that museums are more than just buildings
or collections of artifacts.
This distinction between
inherited frames and evoked frames provides an elegant mechanism for
differentiating the polysemous readings found in (2) and (3) from the metonymic
one in (4) and the conventional reading of (1). In ECG the lexical
constructions associated with verbs introduce specific constraints on their
associated frame elements. The predicate associated with moved requires the moved
entity to be movable, renovated requires the renovated entity to be a building
and bought requires the buyer to be an agent. In each of these cases, the
associated constraints will not be satisfied because they are not licensed by
museum or its supertype(s).
The task of construal
resolution, then, is to find a particular sense for the concept in
question. In examples (2) and (3),
the evoked frames (collection_of_artifacts and building) satisfy the aforementioned constraints respectively. E.g. collection_of_artefacts is +movable and therefore
licenses the reading found in (2) while the building-sense is –movable and thereby
inhibited. Example (4) constitutes another ontologically resolvable construal.
Since the buying frame demands an agent, that is not licensed by the general
museum frame, construal resolution is initiated, finding the corresponding
agentive institution role associated with public_place.
Thus, the ECG formalism not
only facilitates a clear definition of construal, but also supplies the means
for resolving such “unconventional” language use. Nevertheless, as construal does not
always involve ontological type conflicts, many cases thereof can only resolved
taking broader context into account [3].
[1] Bergen, Ben and
Nancy Chang. Embodied Construction Grammar in Simulation-Based Language
Understanding. Technical Report TR-02-004, International Computer Science Institute,
2002.
[2] Lakoff, George. Women, Fire and Dangerous Things. 1987. Chicago: University of Chicago
Press.
[3] Porzel, Robert and Iryna Gurevych. Towards Context-adaptive
Utterance Interpretation. In Proceedings of the 3rd SIGdial Workshop on
Discourse and Dialogue, Philadelphia, 2002.
An ECG account of English reflexive pronouns.
Keith Sanders
Linguists have long sought a unified principle to account for all the uses of English reflexive pronouns, also known as anaphors or self-pronouns. Numerous problematic cases, like those exemplified in (1)-(4) below, have led to difficulty in this search.
In response, some linguists have posited additional formal constraints (such as Chomsky's Specified Subject Condition), or proposed generalizing the semantic constraints on self-pronoun use (Kuno 1987, van Hoek 1997). Others posit two or more disjoint principles, e.g. HPSG binding theories (Pollard & Sag 1992, Golde 1999) in which syntactic constraints determine the antecedents of 'locally o-commanded' self-pronouns, while discourse principles apply to 'locally o-free' ones as in (1).
In this paper, I instead propose a set of four principles which interact to motivate all uses of self-pronouns in English, including adnominal and adverbial uses as in (2)-(3).
Use of a self-pronoun to
denote a referent R is more acceptable to the extent that:
a. Energy-Transfer relation: R can be construed as the Energy-Sink in a reflexive Energy-Transfer relation. (Langacker 1991)
b. Trajector-Landmark relation: R can be construed as the Landmark in a reflexive Trajector-Landmark relation. (Talmy 2000)
c. Viewpoint source: R can be construed as a `viewpoint source'. (Kuno 1987)
d. Self-Other contrast: R can be construed as the Self (or center) in a 'Self-Other deictic field'. (Zubin et al. 1990)
These constraints are independently motivated and straightforwardly stated in the framework of Embodied Construction Grammar (ECG). Together, they co-define a 'Prototypical Reflexive Scene': Prototypical uses (John hit himself) satisfy all four constraints, while less central cases (found a picture of herself, did it herself, ...) satisfy just some of them. The various functions of self-pronouns thus form a radial category (Lakoff 1987), as expected in cognitive-semantic theories.
These constraints interact with the semantics of grammatical constructions to account for the central cases usually covered by 'purely syntactic' principles. Their interaction also accounts for the picture NP effects in (1), the varying acceptability of self-pronouns as prepositional objects in (3), and attested 'long-distance' binding effects as in (4).
(1)a. Maryi found a picture of herselfi in the newspaper.
b. Maryi was ecstatic. Those pictures of herselfi in the Times would help her campaign.
(2)a. The President himself will be speaking here tonight.
b. I made it myself.
(3) a. John spilled soda on himself / *him.
b. John wrapped the blanket around himself / ? him.
c. John hid the dent behind him / *himself.
(4) "The buyeri causes the goods to come to himselfi."
REFERENCES.
Golde, Karin. 1999. Binding theory and beyond. Dissertation, Ohio State University.
Kuno, Susumu. 1987. Functional syntax. University of Chicago Press.
Lakoff, George. 1987. Women, fire, and dangerous things. University of Chicago Press.
Langacker, Ronald W. 1991. Concept, image, and symbol. Mouton.
Pollard, Carl, and Ivan Sag. 1992. Anaphors in English and the scope of binding theory. Linguistic Inquiry 23: 261-303.
Talmy, Leonard. 2000. Toward a cognitive semantics. MIT Press.
van Hoek, Karen. 1997. Anaphora and conceptual structure. University of Chicago Press.
Zubin, David, Soon Ae Chun, & Naicong Li. 1990. Misbehaving reflexives in Korean and Mandarin. In BLS 16: 338-352.