1 or f es Processing Language 2005 etz Groningen al v of 23, Nerbonne Boro Challeng Natur in ersity Sept. John Univ ances Adv Computationalists Recent

Linguistic

¡ 2 Linguistics to ute alk T ib of contr to CL Structure in ing poised is ies conclusions Engineer , Linguistics Acquisition Contact CL and preliminar areas

Some Science Dialectology Diachronic Language Language Other

¡ • • • • • • • Thesis: 3 mation) or y inf , Linguistics theor time to of ute ammatical r reasons ulating g iptions or as (f stim xamples e contrib list descr utions applications! e e such ib to ies vincing contr potentially actical) con xamples of e (pr y poised a wn comprehensiv comprehensiv ongoing w xamples e at at y ignore CL b with to Preliminar ell-kno w attempt attempt plea areas concrete oid v – –

Argument A No No No

¡ • • • • • Some 4 uc- sum- ammar instr r g & , action, tools xtr e ... , m systems Shieber) ter ing! , language y mation (incl. ansducers or Ka tr IR ... , inf , oreign , f engineer aids ing), and access er parsers w Engineering & (Joshi, ans science and in classes; understanding, handicapped strcuture , us question useful ization speech & , automata y & language Science ucture aph theosaur r acter mining, & astr CL y xt xicog ammar infr r te y) le g controlled , are Char ing dictionar (theor , king, ization, softw anslation, chec mar tions tr Language – –

Science Engineer Large

¡ amiliar • • • F 5 ... contact, . language gains acquisition, adio) science (r dating) ... actical to language , pr (carbon y not techniques applied linguistics , electromagnetics , linguistics CL adiochemistr r iosity optics diachronic biochemical , cur applies applies y computational y applies b dialectology en ? = applies iv

Genetics Archaelogy Astronom ... X X

¡ • • • • • • Dr 6 k loc compact) city-b as ules r .g. e , [u]). making ([i], or f 5 1 0 0 4 i-u segments erences 1 0 0 1 0 i-e aun in erence eatures f diff Diff er t) v k) than erences (cle u 1(rounded) 6(bac 3(shor 4(high) Almeida-Br account diff into also smaller useful en Segment t) high) rounded) uch less measure tak used, m e 0(not 3(mid 3(shor 2(front) to be [e]) w system ho also system ([i], ini ws t) rounded) can ] (SPE) r Measuring sho @ i 0(not 3(shor 4(high) 2(front) erence eatures f CL diff i,e:, ˜ , [ : in itics ancement long high rounded adv

Example distance Phonetics Diacr Vieregge-Cucchiar Chomsky-Halle

¡ • • • • 7 (bilinguial . CL in another to used ing also str 1.4 0.8 0.1 0.5 one ite distance wr silence . 3 ing re to Distance I/ ing. r str to r distance , t distance ation Sum replace inser delete segment engineer distance oper l l l 3 3 are g of gIl g gIr @ edit r venshtein sequence O@ O@ O O@ s s s s compare cost to softw Le aka , ican matics minimal deletions distance or Distance df Amer = and bioinf segment tions enshtein v lift Bostonian Standard http://www.let.rug.nl/˜kleiweg/lev/ Le

L-distance Inser

¡ • • alignment), Idea: 8 ) e additiv is (distance Algorithm distances ieties ord ar w v dialectology of the ish of ) Ir s s 2 1 7 to umber ogramming sum n s 1 6 ddres Pr a the , large e 1 5 4 to in application r 4 1 3 2 adresse — equal is d 3 2 2 1 ’95 sample Dynamic CL d 2 1 1 0 2 ord EA , distance a 1 2 0 1 3 4 100-w 0 3 1 2 4 7 5 6 essler enshtein-distance( Use dialect K

v

¡ r s s a d e e • • • Le 9 arieties (Groningen) V inga Heer t Bulgarian Wilber & Sofia) Between (LML, a v Oseno Distance etja P e g with vera ation

A

¡ Collabor 10 olving v in ers’perceptions) erences diff speak hniques iation dialect ar ec v measures T (to xical vis-à-vis CL le ings iptions ... ing of , str alidity off v of , anscr tr airs measur f sets , in to off ing) eighting step phonetic ings w (consistency aired f as str , Applications ms off (cluster parsing from or f ing or f air ther ord f w statistics distance measurements ing, y Fur frequency-based responses). air analyser f edit ator , off erse) v xical ultiple air

f m Le Lemmatizing Lifting Assessing Explor (In

¡ • • • • • • 11 iance) ar v Dialectology in 15-60% inciple ≈ intuitions ( pr y ers’ aph r Steps speak geog vel organizing ergence of as Hypothesis) No div dialect ua vity tance and a ersus (Gr v contin impor . vs ergence v alidation

Areas Con V Quantify Dynamics

¡ • • • • • 12 Varna Malko Tyrnovo 1 2 3 4 5 Ruse Areas age) er v a eighted (w ing cluster

Via

¡ • 13 . iance ar erred. v inf of be 95% can or f Varna Malko Tyrnovo Burgas Shumen coordinates accounting ideal Ruse , uum? coordinates Lovech Plovdiv distances Pleven Smolyan Teteven Contin en 3-dim. giv er Razlog inf Sofia e Blagoevgrad w Scaling: distances  2 n rom

Multi-Dimensional F

¡ • • 14 Varna Malko Tyrnovo Burgas Shumen Ruse Reconciliation Lovech Plovdiv Pleven Smolyan Teteven noise Razlog Visual Sofia Blagoevgrad with ing cluster

Repeated

¡ • 15 Diffusion Venus Earth Linguistic Sun Moon of Deimos Mars Phobos Hypothesis vity

Gra

¡ 16 . ity similar vity linguistic Gra and , via 2 2 m 1 r m settlements G o accommodation = tw F Cohesion the of and linguistic them, , on masses een promotes orce f Linguistic e betw population activ speculated contact attr be the 2 distance social the on’t m w is , the

1

¡ Idea: F m r G 17 vity Gra and , ity /D 2 1 2 p 1 r p = Linguistic 2 similar 2 p /G 1 r 1 of settlements p o ∝ G tw = D produce ects the F of Eff should them masses which ity een betw Detecting action, population dissimilar attr the distance ling. ling. 2 p is is the

,

¡ 1 F D p r 18 in- in stands postulate e vity w Gra 2 p 2 which 1 2 p , p 1 r via p ity − ∝ ∝ and 2 , contact. 2 p D 1 r p dissimilar AND /G social 1 2 r of Cohesion ∝ settlements ∝ o linguistic D orce f D tw e the them activ , of measure attr een e Linguistic w the : betw distance to bene populations the distance relation linguistic 2 Notate p is the

,

¡ 1 erse v D p r 19 ) ? root x or f √ 57 . of 0 = 200000 2 r ( 150000 Zero? Function 100000 Shape? : Geographic Distance (m) root Gravity predicts a positive quadratic effect! 50000 . vs Linguistic Distance vs. Geographic

atic

0

20 15 10 5 Dialect Distance Dialect

Quadr

¡ 20 . am- per- r g alidly ieties v , ... ar ly v erence diff theoretical reliab xpression entire e e CL one iz etc. , unciation erences y from acter regular and diff ed aph pron r char iz aluation/assessment, of v e geog eighting, acter of w of unciation char distances ect measure lems pron useful be eff y be prob Dialectology frequency ma to linguistic enshtein ing, v dialectal on out of erse Le n v areas quantifying in of tur cluster , ages) led: er v measures dialect lusions enab iptions and oundation f techniques are is (sums/a ua anscr distance CL tr Conc or f . contin e other questions y regations enshtein technique mars Lemmatizing/stemming, w v –

spectiv Le Agg Dialect Ne CL Man

¡ • • • • • • 21 Eng. ; pisces Lat. enshtein v Le ‘fish’, of a (Eng. iant r r r r cognates ity? ar Correspondences habere v in /p:f/ a ¯ a i e e e Lat. .g. ’ via e regular , , h t ð t t e v ‘ha a a a i a unciations aligns global Regular f p p p pron and Eng. es identify allel to ish par correspondences w Engl. Ir Indic Greek Latin cognates ho systemiz notes alse f regular Linguistics: ) on (2002) oid puzzle: v us a ak onic ocus linguistics plen f hr ondr K issue: ical Lat. ky ic r Diac

‘full’, which Histor T Solution: Computational

¡ • • • • 22 to pro- recall corre- the order and to (in ones segmental alences global precision to equiv with regular 2003) correspondence or f data Correspondences xical alignments anslation (1999, le tr looking local , alences) Algonquian regular ords Sound from s w or equiv f e identifying Tiedemann on aliz See sound looking cognate , gener Bloomfield’ ideas Regular s to on correspondences alents). & identify aligns allel to ithm needs par equiv sentences sound Melamed’ order algor one allel , notes (in linguistics par anslation tests applies obtaining tr cases alignment ak ak (2002) of 90% aligns ak both MT ondr ondr lem

b identify near spondence MT Diachronic In K K

¡ ondr • • • • • K 23 of u- en se- v m e ution ib the per or source , contr ens using ed? its v doz popular that quantifying to impro of most ed be aliz the ords means concludes w correspondences a conditioning? gener He Linguistics . as be related elsenstein) regular F phonetic onic of y ordNet cognates? b to cognates W hr e in 2003)? with , see semantically Diac enlists PHYLIP sensitiv ( to of along significance Nature & correspondences e ak lik more ondr CL ieties? biology K ould ar , v w identified statistical Atkinson, made recognition identifying & be one be y or the f a More: related lap cognates Gr of wings er Can v measures . computational o measures (see models is alignment borro y spotting minor essler

ideas is mantic tation hundreds K In Can Can Can Wh

¡ • • • • • • 24 y” CL in theor Acquisition e g linguistic techniques of ning lem Langua lear & prob al CL machine “centr in match? interest Acquisition vious

Lg. Huge Ob

¡ • • • 25 sum: e minimiz and segmentation w allo ances), ning! le) ience utter lear ailab of v xper xicon postulated a e le a in types length Acquisition in por e types endings ord ens g w iption (cor ? ord types tok Brent of and w gi/ of of O of descr sd 1 speech Michael na Langua umber umber y entropies length n n b 2R@ & minimal s (beginnings giw 90’ O and CL sd 1 mid child-directed in gina k O phonotactics or /d w acquisition in whether use lg. ed ords

Pioneer Ask W Idea: Link

¡ • • • • • 26 from med or phonotactics ell-f of doggie!’ w ate nice studies a separ What to . (Groningen) doggie required Acquisition e Nice . g statistics ‘Doggie onstantopoulos Dutch K → in Langua , v gi/ compact, & O not sd ned? 1 CL Stoiano na lear ely occur?) 2R@ Russian, this techniques ectiv in les giw Sang, is O eff w sd OK 1 / Kim syllab S t med E —Ho gina or O

ill-f (what /d Tjong /vstr Rule-based

¡ • • • • . 27 ience xper e bias). in (in . not Interest ial ange r mater wing within es) assumptions o y Acquisition Ha Gr among & coming ight . innateness Language e ms adually (Albr r or f y g erentiating , le Human diff , ationaliz theor of lems Acquisition: possib oper e prob profile g less Models ed linguistic . error vs of ulations ’05 t le on unsolv sim par CL of A Langua the on possib ocus f & iz ’04, on on hor CL esting

T COLING Psycho-Computational Computational Interest Linguistic Huge

¡ • • • • • 28 mixing. and contact al cultur in “atlases” Contact use eling e v g le interest ... acquisition , to limited systematic of dialect due no , , uctures le Langua str , & ailab creoles perhaps v , a dialectology a second-language CL sounds , por from ect) pidgins interest, , cor ords in w of oinés (imperf k wing to in ro wing g situation: —techniques

Borro Mixing Area Linking Data

¡ • • • • • 29 and , signifi- us y measure , er v ector Lauttam v viate de into to viation test ams de r Hänninen, ants ig r of ection viation utation de Inf Opas immig m POS-tr of sources per ection” late ree viation ute “inf via of collect deg ib de or f . in of attr set, (Groningen), Syntactic to ants r speech tag w of significance reflected candidate measure immig sho be Wiersema one ical smallish techniques can child assess will a, e , of w (with with umer por n Measure a ection cor erence por o inf result speech tw diff y result: (Oulu)): (automated) cor e am from r tak both need onen v ag

cantly histog Hir Idea: T Hypothesis: Expected Preliminar Still

¡ • • • • • • 30 dis- parsing, impact (psychological theoretical opics T in attention of limited al. ut viv b center re of lier Linguistic y lished, th or w estab Other wn, perhaps discussion) processing—ear ut or ell-kno f b (question ammar—w

ambiuation), Gr Psycholingistic ??

¡ • • • 31 (to pro- specific demon- the emes x to to le (with . erent ning speech alences diff lear t of Science? linguistics equiv ms ical or f child-direct suppor to to histor anslation in tr identify mation to Illuminate or techniques inf identifying ML stemmer on k ter sufficient correspondences apply or or w P s a dialectology). Engineering sound iments in use contains eg xper lap Can e Melamed’ data er regular v o Kleiw ning & input adopts xical lear finding le that ak Aside: al of er v ate ondr lem

b detect str biases). Nerbonne K Se

¡ • • • 32 ...) , Eisner ) Linguistics Gerdemann, passim & et seq.) ...) utions et ib utions Noord 2004 ib an 2002) v Illuminates contr 1997 inga empen, ak contr K , CL Heer er LFG) CL or k ondr f ttunen, (Brent, k, or (K f or (potential) w (Kar (Croc HPSG, y tunities wn tunities (o (CG, Acquisition Contact Theor Linguistics oppor Computation oppor ical wn ammars Language Histor Language Optimality Psycholinguistics Dialectology Gr ell-kno – – – – – – –

W Emerging

¡ • • 33 1974-2000 (sg). man ds Ger Bentheim, sg in nordhorn standard Standar border . nieuw schoonebeek neuenhaus lattrop vs & hoogstede man lage s wilsum (sd) vasse uelsen der schoonebeek Dutch emlichheim langeveen Dutch-Ger itterbeck Bor radewijk coevorden bergentheim standard ergence gramsbergen ard Div sd w to 2000: Application: al. ergence et v con inga

Heer Blue

¡

¡ ¢

34