Computational Cognitive Morphosemanticsnschneid/bls36-slides.pdf · Construction Grammar • In the family of cognitive theories known as Construction Grammar, there is no separation

Computational Cognitive Morphosemantics

Modeling morphological compositionality in Hebrew verbs with

Embodied Construction Grammar

1

Nathan Schneider ~ BLS 36 ~ 7 February 2010

http://www.cs.cmu.edu/~nschneid/bls36-slides.pdf



Overview

• An analysis of Hebrew verbs linking cognitive semantics to complex morphological constructions

‣ Must account for compositionality as well as idiosyncrasy

• Cast within the ECG formalism to facilitate computational processing

‣ Previously, ECG was only used for syntax

2

3

meaning

3

בונה בונה

meaning

morphologically complex forms(e.g. Hebrew verbs)

bonɛ bona

3

בונה בונה

meaning


LANGUAGE USE IN

PROGRESSconstruction

grammar

bonɛ bona

3

בונה בונה

meaning


LANGUAGE USE IN


grammar

bonɛ bona

3

בונה בונה

meaning

formal representation


LANGUAGE USE IN


grammar

bonɛ bona

3

בונה בונה

meaning

formal representation


LANGUAGE USE IN


grammar

automatic analysis procedure

bonɛ bona

Construction Grammar for Hebrew Verb Morphology

• Challenges:‣ Nonconcatenative morphology

‣ Semantics of roots, paradigms, and verbs (whether compositional or idiosyncratic)

• I will use Embodied Construction Grammar, a formalism designed to support computational analysis and simulation of sentences

4

Construction Grammar

• In the family of cognitive theories known as Construction Grammar, there is no separation between lexicon and grammar

• Words, lexical categories, multiword expressions, syntactic phrases, idioms all form-meaning pairs: constructions, albeit with different levels of generality

• Usage-based theories of grammar: constructions may be stored redundantly in memory (“constructicon”); sensitive to factors such as frequency

5

e.g.: [Fillmore et al. 1988] [Kay & Fillmore 1999] [Goldberg 1995, 2006] [Langacker 1990] [Croft 2001] [Tomasello 2003]

Other Related Work

Formal Approaches to Semitic/Nonconcatenative Morphology[McCarthy 1979] proposed an autosegmental analysis for the root-pattern morphology of Arabic. [Finkel & Stump 2002] used inheritance in the KATR formalism to describe Hebrew verb forms. For other approaches to nonconcatenative morphology, see [Orgun 1996] [Rubba 2001] [Roark & Sproat 2007].

Morphology in Construction GrammarPrevious work has described composition of morphological constructions [Riehemann 1998] [Booij 2005, 2007] [Gurevich 2006]. Several mechanisms for adding morphology to ECG were entertained in [Bergen 2003], but none were implemented. [Rubba 1993] (synopsis in [Rubba

2001]) takes a Cognitive Grammar approach to nonconcatenative morphology, situating words in a network (cf. [Bybee 1985, 2001]). Two other relevant approaches to phonology are found in [Inkelas 2008] and [Nathan 2007]. [Mandelblit 1997] offers an extensive semantic account of Hebrew verb paradigms.

6

There has been a great deal of formal work on Semitic morphology using a variety of approaches, including rules, autosegmental phonology, and unification grammars. There has also been some work on morphology in Construction Grammar and related theories. To my knowledge, this is the first work to explicitly combine detailed semantic representations from cognitive linguistics in a formal description of morphological constructions.

7e.g. [Berman 1978]

Key claim: The grammar has morphological constructions at multiple levels, all of which contribute to the form and meaning of the composite word.

ROOT

7

g n b

e.g. [Berman 1978]


7

g n b

e.g. [Berman 1978]


STEM hi i

7

g n b

e.g. [Berman 1978]


hi i

7

g n b

e.g. [Berman 1978]


INFLECTION u

hi i

7

g n b

e.g. [Berman 1978]


uhi i

7

g n b

e.g. [Berman 1978]


8

Root /g/▫/n/▫/b/ ~ ‘steal’/g/▫/n/▫/b/ ~ ‘steal’ Paradigm Hif’il (P5) Meaning ‘smuggle in’

Tense/Num Past m Past f Present m Present f Future m Future f

1.sg hignavtihignavti

magniv magniva

ʔagnivʔagniv

2.sg hignavta hignavt magniv magniva tagniv tagnivi

3.sg higniv hignivamagniv magniva

yagniv tagniv

1.pl hignavnuhignavnu

magnivim magnivot

nagnivnagniv

2.pl hignavtem hignavten magnivim magnivot tagnivutagnivu

3.pl hignivuhignivu

magnivim magnivot

yagnivuyagnivu

8

• A single stem for each root/paradigm/tense: here past /hignib/*, present /magnib/, future /agnib/ The root fits into a pattern: /hi◦◦i◦/, /ma◦◦i◦/, /a◦◦i◦/

• Affixes specifying person, gender, and number—not sensitive to paradigm: /-ti/, /-im/, /t- -u/, etc.




magniv magniva

ʔagnivʔagniv



yagniv tagniv


magnivim magnivot

nagnivnagniv


3.pl hignivuhignivu

magnivim magnivot

yagnivuyagnivu

8

• A single stem for each root/paradigm/tense: here past /hignib/*, present /magnib/, future /agnib/ The root fits into a pattern: /hi◦◦i◦/, /ma◦◦i◦/, /a◦◦i◦/

• Affixes specifying person, gender, and number—not sensitive to paradigm: /-ti/, /-im/, /t- -u/, etc.




magniv magniva

ʔagnivʔagniv



yagniv tagniv


magnivim magnivot

nagnivnagniv


3.pl hignivuhignivu

magnivim magnivot

yagnivuyagnivu

• For brevity, assume some phonological details are handled elsewhere: Consonant allophony: /b/ is sometimes realized as [v], /k/ as [x], and /p/ as [f]

Certain root consonants (e.g. /ʔ/, /w/, /h/) will affect the pattern in systematic ways

Stress-sensitive vowel reduction and deletion

* The last vowel in this paradigm’s past tense stem undergoes the phonological change /i/→[a] in 1st & 2nd person

9

Morphological Generalizations: Stored or Inferred?

• I will present general morphological constructions as if they are stored in the lexicon along with all other constructions.

• However, some approaches to morphology claim that no constructions below the word level are stored in memory; rather, an online process of distributed analogy is hypothesized to account for morphological productivity. [Gurevich 2006]

‣ For those taking this view, the generalizations presented here can be interpreted as formalizing an online analogical process.

10

• The root /g/▫/n/▫/b/ is one which manifests itself in all seven paradigms, though its P3 and P4 verbs are limited to literary usage.

P TraditionalCharacterization1

Transitivity: always (often)2

/g/▫/n/▫/b/ Verbs3/g/▫/n/▫/b/ Verbs3P TraditionalCharacterization1

Transitivity: always (often)2 Hebrew Gloss

1 “Simple” (Transitive) ganav ‘steal’

2 “Refl., passive” Intrans. (Passive) nignav ‘be stolen’

3 “Intensive” (Transitive) ginev ‘steal repeatedly’ (lit.)

4 “Intensive Passive” Passive gunav ‘be stolen/taken stealthily’ (lit.)

5 “Causative” (Transitive) higniv ‘smuggle in, insert stealthily’

6 “Causative Passive” Passive hugnav ‘be smuggled in/inserted stealthily’

7 “Reflexive-passive” Intrans. (Passive) hitganev ‘sneak (in, out, or away)’1. [Halkin 1970] 2. [Arad 2005] 3. [Bolozky 1996]

Paradigms (Binyanim)

Paradigm Semantics

• There is a lot of idiosyncrasy in the meanings of verbs within the various binyanim. That is, the verb’s meaning is often not completely predictable from the root and paradigm.

• Mandelblit [1997] attacks this problem under the rubric of grammatical blending [Fauconnier & Turner 1996]

‣ She concludes that the different paradigms arise from a construed causal relationship, which explains the prototypical semantics

11

Paradigms: An English Analogy

adapted from [Mandelblit 1997, p. 36]

Susan trotted the horse into

the stable. (causative)

protagonist

predicate

protagonist

predicate

direction

CAUSEsubject

verb root

object

oblique

The horse trotted into the

stable. (basic)

protagonist

predicate

direction

subject

verb root

oblique

Susan

trot

the horse

into the stable

horse

trot

into the stable

Paradigm Semantics

Mandelblit argues that the root contributes the “content” of the verb, and the paradigm picks out part of a causal sequence. For example: “The causative hif’il verbal pattern is used to mark a single sub-event (the effected sub-event) within a conceived causal sequence of events. Marking other sub-events entails the usage of other binyanim.”

13

Paradigm Semantics: Hif’il

These verbs have the same root, /r/▫/w/▫/ʦ/ ‘run’. In the causative hif’il sentence (b), heriʦ ‘cause to run’ indicates that the root refers to the effected event—that is, what the soldiers are made to do. The causing event, i.e. how the commander makes them run, is unspecified.

14

sentence, though we understand that it is the motion of the napkin. Mandelblit explains:

Speakers automatically complete the sequence of events with additional information from back-

ground knowledge of mental models and past conceived scenarios stored in memory [. . . ]. For

example, a specific manner of motion is almost automatically inferred in each of the Caused-Motion

examples [. . . ] though nothing in the linguistic expression (neither the lexical items, nor the argu-

ment structure) explicitly provides this information. (Mandelblit, 1997, p. 42)

The horse and napkin examples demonstrate the flexibility of the English Caused-Motion construction: either

the causing or the effected event may be highlighted as the verb in the construction, according to the needs of

the speaker, without any special marking. This is an instance of constrained best-fit: the speaker’s expression of

the situation is constrained by the chosen construction, and she must choose which predicate is more relevant

to the intended meaning. Likewise, the listener must consider both highlighting alternatives and choose the

interpretation that better fits the context and background knowledge.

Whereas the choice of highlighted event in English isn’t marked in the sentence—it is up to the listener to

determine, based on context and world knowledge, whether the highlighted event is the causing or effected

event—Mandelblit’s thesis is that in Hebrew, this choice of highlighting is marked morphologically by the choice

of binyan; or, to put it in another way, a particular construal and highlighting of the event sequence is the

semantic contribution of each of the different binyanim.

2.2.2 Blending analysis of hif’il and pi’el

This section will recapitulate the key points of Mandelblit’s (1997) analysis of Hebrew verbal morphology which

will be relevant to the ECG analysis presented below. We will start by looking at each binyan in turn, and how

they contrast with each other.

As is evidenced in Arad’s (2005) and Bolozky’s (1999) observations, binyan hif’il is prototypically used for

causatives—which Bolozky defines to include verbs of “causing the patient to do something,” “causing an entity

to be (or become) something,” and “causing an entity to begin a new state” (pp. 53–54). Examples (10)–(12),

from Mandelblit (1997, ch. 4), illustrate the “causing the patient to do something” variety when contrasted with

the pa’al verb with the same root:

(10) a. ha-xayal raţ misaviv la-migraS.the-soldier run.PA’AL.PAST.3.M.SG around to.the-courtyard

‘The soldier ran around the courtyard.’

b. ha-m@faked heriţ Pet ha-xayal misaviv la-migraS.the-commander run.HIF’IL.PAST.3.M.SG ACC the-soldier around to.the-courtyard

‘The commander made the soldier run around the courtyard.’

(11) a. ruti yarSa Pet ha-bayit.Ruti inherit.PA’AL.PAST.3.F.SG ACC the-house

‘Ruti inherited the house.’

b. dani horish Pet ha-bayit [email protected] inherit.HIF’IL.PAST.3.M.SG ACC the-house to-Ruti

‘Danny bequeathed the house to Ruti.’

(12) a. ha-tinok Paxal Pet ha-daysa.

the-baby eat.PA’AL.PAST.3.M.SG ACC the-cereal

‘The baby ate the cereal’

9

Paradigms Semantics: Hif’il

[Mandelblit 1997, ch. 4]

The commander ran the soldiers.

(causative)

protagonist

predicate

protagonist

predicate

CAUSEsubject

verb root

commander

run

HIF’IL

The soldiers ran. (basic)

protagonist

predicate

subject

verb root

soldiers

run

PA’AL

The space on the right (Input 1) illustrates the conceived causal relationship bound to specific participants and actions: she is understood to have taken some unspecified action—the causing event—which resulted in the horse trotting into the stable, trotting being the effected event. The space on the left (Input 2) shows how the Caused-Motion construction orders certain types of participants and predicates in an event sequence, associating them with syntactic categories. The Caused-Motion construction is said to be an integrating syntactic construction because it frames the sentence as a single event, even though the sentence has unintegrated semantics with two events in a causal relationship (depicted in Input 1). The blending operation results in the space at the bottom, with lexical items denoting some of the participants and predicates from Input 1 bound to syntactic positions from Input 2. Those participants and predicates which are realized in the blend, with increased cognitive salience and overt representation in the sentence, are said to be profiled or highlighted.

Inherited Generalizations

PAST

PRESENT

FUTURE

hi ig n bCAUS.+STEAL

a ig n bCAUS.+STEAL

ma ig n bCAUS.+STEAL

16

HIF’IL STEMS


PAST

PRESENT

FUTURE

hi ig n bCAUS.+STEAL

a ig n bCAUS.+STEAL

ma ig n bCAUS.+STEAL

hi iCAUS.

a iCAUS.

ma iCAUS.

16

HIF’IL STEMS


hi iCAUS.

a iCAUS.

ma iCAUS.

16

HIF’IL STEMS


hi iCAUS.

a iCAUS.

ma iCAUS. CAUS.

PARADIGM

16

HIF’IL STEMS

Paradigm Semantics: Pi’el

Prototypical transitive pi’el verbs have a root denoting a causing event. Thus, they contrast with hif’il verbs much like sneeze in Rachel sneezed the napkin off the table contrasts with trot in She trotted the horse into the stable:

17





Prototypical transitive pi’el verbs, she writes, have a root denoting a causing event. Thus, they contrast with hif’ilverbs much like sneeze in Rachel sneezed the napkin off the table contrasts with trot in She trotted the horse intothe stable (see the previous section). As an example, Mandelblit gives:

(14) ha-maQasik piter Pet ha-Qoved.the-employer fire.PI’EL.PAST.3.M.SG ACC the-worker‘The employer fired the worker.’

The verb fire is the causing event in a conceived causal sequence in which an employer takes an action (thefiring) and as an effect, the employee no longer has a job. A sentence using this verb is thus said to highlight thecausing event, and it is left up to the listener to infer that the resulting event would be something like a stativeverb ‘be unemployed’.

Mandleblit indicates that her analysis doesn’t apply to certain small subcategories of pi’el verbs, such as theso-called intensives.

2.2.3 Blending analysis of huf’al and pu’al

One important function of the binyan is to indicate voice. The hif’il and pi’el examples given thus far are active.Two other binyanim, huf’al and pu’al, are their passive counterparts. Huf’al and pu’al are the most semanticallypredictable of the binyanim, in that they must have an active counterpart with the same root (in hif’il or pi’el,respectively). Aside from voice, there is no difference between the meanings of active and passive. Note, how-ever, that the lexical correspondence is asymmetric: though every passive verb requires an active counterpart,many hif’il verbs do not have huf’al counterparts, and likewise with pi’el/pu’al (Coffin and Bolozky, 2005; Arad,2005)—which I see as evidence against generative approaches which mechanically derives passive forms fromactive ones (Arad, 2005; Berman and Bolozky, 1978).

The verbs in (10) and (14), it so happens, do have passive counterparts:

(15) ha-xayal huraţ (Qal y@dei ha-m@faked).the-soldier run.HUF’AL.PAST.3.M.SG (on account.of the-commander)‘The soldier was made to run (by the commander).’

(16) ha-Qoved putar (Qal y@dei ha-maQasik).the-worker fire.PU’AL.PAST.3.M.SG (on account.of the-employer)‘The worker was fired (by the employer).’

These passive binyanim differ from their active counterparts, Mandelblit insists, only in that the passive blendingschemas highlight the theme—that is to say, the causing protagonist—as the subject of the sentence, whereasactive blending schemas highlight the agent—or affected protagonist—as the subject. (Like in English, the passiveby-phrase is optional; because it is in a position of demoted prominence, an overt agent in a passive sentence isstill said to be unhighlighted.) Huf’al, like hif’il, highlights the effected event in a conceived causal sequence;and pu’al, like pi’el, highlights the causing event. That they evoke “a similar underlying conceptual structure andblending configuration,” she says, accounts for the regularity in the meaning relationship between hif’il/huf’aland pi’el/pu’al pairs (Mandelblit, 1997, p. 132).

The blending schemas of the four binyanim discussed thus far may illustrated as follows:

(17) Illustration of blending schemas for pi’el, pu’al, hif’il, and huf’al (Mandelblit, 1997, p. 133)

11

Paradigm Semantics: Huf’al and Pu’al

18






Huf’al is the passive counterpart of hif’il, and pu’al is the passive counterpart of pi’el:













11













11

Paradigms Semantics

adapated from [Mandelblit 1997, p. 133]

Blending schemas for paradigms P1 & P3 (active) and P2 & P4 (passive). The construed causal sequence on the RHS of each paradigm contains a causing event (top box) and an effected event (bottom box), each with agent and predicate. Paradigm constructions map one of the protagonists (agents) to the subject and one of the predicates to the verb root.

P1: Pi’el

protagonist

predicate

protagonist

predicate

CAUSEsubject

verb root

P3: Hif’il

protagonist

predicate

protagonist

predicate

CAUSEsubject

verb root

P2: Pu’al

protagonist

event

protagonist

predicate

CAUSEsubject

verb root

P4: Huf’al

protagonist

predicate

protagonist

predicate

CAUSEsubject

verb root

The space on the right (Input 1) illustrates the conceived causal relationship bound to specific participants and actions: she is understood to have taken some unspecified action—the causing event—which resulted in the horse trotting into the stable, trotting being the effected event. The space on the left (Input 2) shows how the Caused-Motion construction orders certain types of participants and predicates in an event sequence, associating them with syntactic categories. The Caused-Motion construction is said to be an integrating syntactic construction because it frames the sentence as a single event, even though the sentence has unintegrated semantics with two events in a causal relationship (depicted in Input 1). The blending operation results in the space at the bottom, with lexical items denoting some of the participants and predicates from Input 1 bound to syntactic positions from Input 2. Those participants and predicates which are realized in the blend, with increased cognitive salience and overt representation in the sentence, are said to be profiled or highlighted.

Noncompositional Verbs

• The story until now assumes the paradigm semantics is fully compositional given the root: the verb cxn may be deduced online

• To handle noncompositional verbs, we introduce a verb-specific base construction which pairs a particular root with a particular paradigm, and the associated semantics

‣ Tense/other inflectional information does not alter the verb-specific meaning

20


21

hi ig n bSMUGGLE

a ig n bSMUGGLE

ma ig n bSMUGGLE

g n bSMUGGLE

BASE

STEMS

CAUS.PARADIGM


22

HIGNIBBASE

HIGNIBPASTSTEM

HIF’IL

HIGNIBFUTSTEM

HIGNIBPRESSTEM

HIF’ILPASTSTEM HIF’ILFUTSTEMHIF’ILPRESSTEM

This shows an inheritance hierarchy of constructions (à la HPSG): Hif’il is the most general and more specific cases inherit and elaborate upon its properties. HignibBase and its subtypes are idiosyncratic with respect to meaning, overriding the inherited prototypical causative meaning.


• Unification-based, semantically-rich formalism for describing lexical and syntactic—and now morphological—constructions [Bergen & Chang 2005] [Feldman 2006] [Feldman et al. 2009]

‣ Part of the Neural Theory of Language project to develop computational simulations of language understanding

‣ ECG grammars can represent embodied semantics: primitives include schemas/frames, as well as metaphors and mental spaces [Gilardi to appear]

23

• Facilitates (verifiably consistent) analyses of particular linguistic phenomena, e.g. motion-related constructions in English [Dodge 2010]

• Facilitates cognitive computational models of sentence processing [Bryant 2008] and language learning [Chang 2008] [Mok 2008]

• Tools include the probabilistic parser of [Bryant

2008] and a user-friendly interface for grammar engineering [Gilardi to appear]

24


ECG Analysis: Setup

• We want to formally specify a “constructicon” for Hebrew verbs

‣ Small but very detailed decomposition of morphology and semantics

‣ For our purposes, phonology is simplified to string concatenation

• Given this constructicon, a computer program can take an input word and list its possible analyses—including semantic frames and their bindings

25

26

Schemas and Constructionsschema Causation subcase of ComplexProcess roles causingProcess: Process effectedProcess: Process causalProtagonist: Entity affectedProtagonist: Entity

In ECG, meaning schemas are used to represent the frame semantics of a construction, and form schemas are used to decompose morphological forms. Schemas exist in an inheritance lattice and can define roles, which may be string-valued or may point to other schema instances.

26

Schemas and Constructionsschema Causation subcase of ComplexProcess roles causingProcess: Process effectedProcess: Process causalProtagonist: Entity affectedProtagonist: Entity

schema GNB subcase of Root roles r1 r2 r3 constraints r1 ← "g" r2 ← "n" r3 ← "b"

In ECG, meaning schemas are used to represent the frame semantics of a construction, and form schemas are used to decompose morphological forms. Schemas exist in an inheritance lattice and can define roles, which may be string-valued or may point to other schema instances.

construction Root_GNB subcase of VerbRoot form: GNB meaning: Steal

27

Constructional Levels: Base

Recall that paradigm hif’il highlights the effected process and the causal protagonist.

general construction Hif’il subcase of Paradigm constructional constituents root: Root form constraints root.r1 before root.r2 before root.r3 meaning: Causation roles highlightedProtagonist: Entity highlightedProcess: Process constraints highlightedProcess ↔ root.m highlightedProcess ↔ effectedProcess highlightedProtagonist ↔ causalProtagonist hif’il-specific

Hif’il

The Hif’il construction on the left specifies the compositional meaning and gives (underspecified) constraints on the form. HignibBase inherits from Hif’il for the root GNB, overriding the compositional meaning. The tense-specific stem constructions will inherit from HignibBase in turn.

27

Constructional Levels: Base

Recall that paradigm hif’il highlights the effected process and the causal protagonist.

general construction Hif’il subcase of Paradigm constructional constituents root: Root form constraints root.r1 before root.r2 before root.r3 meaning: Causation roles highlightedProtagonist: Entity highlightedProcess: Process constraints highlightedProcess ↔ root.m highlightedProcess ↔ effectedProcess highlightedProtagonist ↔ causalProtagonist

construction HignibBase subcase of Hif’il constructional constituents root: Root_GNB meaning: Smuggle

an idiosyncratic meaning (overrides

Causation)

hif’il-specific

Hif’ilHif’il

The Hif’il construction on the left specifies the compositional meaning and gives (underspecified) constraints on the form. HignibBase inherits from Hif’il for the root GNB, overriding the compositional meaning. The tense-specific stem constructions will inherit from HignibBase in turn.

Conclusion

The aforementioned approach

• brings together the theoretical framework of Construction Grammar and studies of verbs in Modern Hebrew;

• integrates the form and meaning components of morphological structures in a single analysis; and

• employs and extends the Embodied Construction Grammar formalism so as to enable cognitive computational modeling of morphology.

28

Arad, M. (2005). Roots and patterns: Hebrew morpho-syntax. Dordrecht: Springer.Bergen, B. K. (2003, March 18). Towards morphology and agreement in Embodied Construction Grammar.

Manuscript. Retrieved October 18, 2007, from http://www2.hawaii.edu/~bergen/papers/ECGmorph.pdf.Bergen, B. K., & Chang, N. (2005). Embodied Construction Grammar in simulation-based language understanding.

In Construction grammars: cognitive grounding and theoretical extensions (pp. 147–190). John Benjamins.Berman, R. A. (1978). Modern Hebrew structure. Tel-Aviv: University Publishing Projects. Bolozky, S. (1996). 501 Hebrew verbs. Hauppauge, NY: Barron’s Educational Series. Booij, G. (2005). Compounding and derivation: evidence for Construction Morphology. In W. U. Dressler, F. Rainer,

D. Kastovsky, & O. Pfeiffer (Eds.), Morphology and its demarcations (pp. 109–132). John Benjamins.Booij, G. (2007). Construction Morphology and the lexicon. In F. Montermini, G. Boyé, & N. Harbout (Eds.), Selected

proceedings of the 5th Décembrettes: Morphology in Toulouse (pp. 34–44). Somerville, MA: Cascadilla Press.Bryant, J. (2008). Best-fit constructional analysis. Ph.D. dissertation, University of California, Berkeley.Bybee, J. L. (2001). Phonology and language use. Cambridge University Press. Bybee, J. L. (1985). Morphology: a study of the relation between meaning and form. Typological studies in language.

Amsterdam: John Benjamins. Chang, N. (2008). Constructing grammar: a computational model of the emergence of early constructions. Ph.D.

dissertation, University of California, Berkeley.Croft, W. (2001). Radical Construction Grammar: syntactic theory in typological perspective. Oxford University Press.Dodge, E. (2010). Conceptual and constructional composition. Ph.D. dissertation, University of California, Berkeley.Fauconnier, G., & Turner, M. (1996). Blending as a central process of grammar. In A. E. Goldberg (Ed.), Conceptual

structure, discourse, and language (pp. 113-129). Stanford, CA: Center for the Study of Language and Information (CLSI), Cambridge University Press.

Feldman, J. A. (2006). From molecule to metaphor: a neural theory of language. MIT Press.Feldman, J. A., Dodge, E., & Bryant, J. (2009). A neural theory of language and Embodied Construction Grammar. In

B. Heine & H. Narrog (Eds.), The Oxford Handbook of Linguistic Analysis. Oxford University Press.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.Finkel, R., & Stump, G. (2002). Generating Hebrew verb morphology by default inheritance hierarchies. In

Proceedings of the ACL-02 Workshop on Computational Approaches to Semitic Languages. Philadelphia, Pennsylvania, USA: Association for Computational Linguistics.

29

Goldberg, A. E. (1995). Constructions: a construction grammar approach to argument structure. University of Chicago Press.

Goldberg, A. E. (2006). Constructions at work: the nature of generalization in language. Oxford University Press.Gurevich, O. (2006). Constructional morphology: the Georgian version. Ph.D. dissertation, University of California,

Berkeley. Halkin, A. S. (1970). 201 Hebrew verbs. Hauppauge, NY: Barron’s Educational Series.Inkelas, S. (2008, February). The morphology-phonology connection. Presented at the 34th meeting of the

Berkeley Linguistics Society, Berkeley, CA.Kay, P., & Fillmore, C. J. (1999). Grammatical constructions and linguistic generalizations: the What’s X doing Y?

construction. Language, 75(1), 1–33.Langacker, R. W. (1990). Concept, image, and symbol: the cognitive basis of grammar. Berlin: Mouton de Gruyter.Mandelblit, N. (1997). Grammatical blending: creative and schematic aspects in sentence processing and

translation. Ph.D. dissertation, University of California, San Diego.McCarthy, J. J. (1979). Formal problems in Semitic phonology and morphology. Ph.D. dissertation, MIT.Mok, E. (2008). Contextual bootstrapping for grammar learning. Ph.D. dissertation, University of California,

Berkeley.Nathan, G. (2007). Phonology. In D. Geeraerts & H. Cuyckens (Eds.), The Oxford handbook of cognitive linguistics

(pp. 611–631). Oxford University Press US.Orgun, C. O. (1996). Sign-Based Morphology and Phonology with special attention to Optimality Theory (Ph.D.

dissertation). Berkeley, CA: University of California, Berkeley.Riehemann, S. Z. (1998). Type-based derivational morphology. The Journal of Comparative Germanic Linguistics,

2(1), 49–77.Roark, B., & Sproat, R. W. (2007). Computational approaches to morphology and syntax. Oxford surveys in syntax

and morphology. Oxford: Oxford University Press. Rubba, J. (1993). Discontinuous morphology in modern Aramaic (Ph.D. dissertation). University of California, San

Diego.Rubba, J. (2001). Introflection. In M. Haspelmath, E. König, W. Oesterreicher, & W. Raible (Eds.), Language

typology and language universals: an international handbook (Vol. 1, pp. 678–694). Berlin: Walter de Gruyter.Tomasello, M. (2003). Constructing a language: a usage-based theory of language acquisition. Cambridge, MA:

Harvard University Press.

30

31

Acknowledgments

Thanks are in order...

To Jerry Feldman, Eve Sweetser, George Lakoff, John Bryant, and the rest of the Neural Theory of Language Group; Miriam Petruck; and Rutie Adler and her Hebrew Linguistics class; and Noah Smith, Lori Levin, and Scott Fahlman for their advice & feedback

And to all of you for listening!

Documents

Computational Cognitive Morphosemanticsnschneid/bls36-slides.pdf · Construction Grammar • In the family of cognitive theories known as Construction Grammar, there is no separation