Produc'vity and schema'city of the way - construc'on in Late Modern English Florent Perek University of Birmingham 1
Diachronic construc'on grammar • Language change consists either of construc'onaliza'on or construc'onal change (TraugoB & Trousdale 2013) • Three aspects of construc'onal change: – Composi'onality: how seman'cally transparent a construc'on is – Produc'vity: the range of lexical items that may occur in it – Schema'city: the level of detail (esp. seman'c) with which the construc'on is stored; defines restric'ons on use • This talk focuses on how the last two can be characterized from corpus data 2
Produc'vity and schema'city • Produc'vity can be observed in corpus data; what about schema'city? • The two are commonly thought to be interrelated – More schema'c construc'ons have more schema'c slots: fewer constraints on the lexical items that can be used – Conversely, the occurrence of more diverse items makes a slot more schema'c 3
Produc'vity and schema'city • By no means a one-to-one rela'on – Compa'bility of an item with a schema does not mean that it will necessarily be aBested – Conversely, new coinages can happen outside of a schema, e.g., by analogical extension • Schema'city vs. produc'vity ~ licensing vs. coining • How to characterize schema'city when only aBested types are observable? 4
Case study: the way -construc'on • Verb one’s way PP, e.g., He pushed his way through the crowd • Describes mo'on of the subject referent • Three senses of the construc'on: – Path-crea'on: the verb describes what enables mo'on They hacked their way through the jungle. – Manner: the verb describes the manner of mo'on They trudged their way through the snow – Incidental-ac'on: the verb refers to some co-occurring ac'on unrelated to mo'on He whistled his way across the room 5
The way -construc'on in diachrony • Previous research mostly focused on the origins of the construc'on (Israel 1996, TraugoB & Trousdale 2003) • LiBle discussion of the recent history of the construc'on (19 th –20 th ) – Gramma'cally stable since the 19 th century – Good case for the study of syntac'c produc'vity • Excep'on: Mondorf (2011) – But her focus is on the comparison with the self -resulta've construc'on (e.g., He worked himself to exhaus<on ) – Only ten verbs, few datapoints 6
Data • All tokens of “V Poss way Prep” between 1830 and 2009 extracted from the Corpus of Historical American English (COHA, Davies 2010) • Manually filtered, annotated for construc'onal meaning: path-crea'on, manner, incidental-ac'on • Quan'ta've measures of produc'vity – Token frequency: how oeen the construc'on is used? – Type frequency: with how many different verbs? 7
Token frequency (per million words) Verb type frequency 60 path − creation path − creation ● ● ● ● Tokens per million words 150 manner ● manner 50 incidental incidental ● ● ● 40 ● ● ● ● ● ● ● ● ● ● ● ● ● 100 ● ● Types ● ● ● 30 ● ● ● ● 20 ● 50 ● ● ● ● ● ● 10 ● 0 0 1830 1860 1890 1920 1950 1980 2010 1830 1860 1890 1920 1950 1980 2010 • Token frequency is rela'vely stable • Steady increase in type frequency: the construc'on is used with more and more different verbs 8
Type frequency • Type frequency reflects the lexical range of a construc'on • But it is a purely quan'ta've measure – Only indirectly related to seman'c diversity – No account of how different items are • Ques'ons: – What kinds of verbs joined the distribu'on? – Did it become more seman'cally diverse? – Are there par'cular seman'c domains favored by the construc'on? 9
Distribu'onal seman'cs • Most studies in DiaCxG draw on seman'c intui'ons • This paper takes a different approach: using distribu'onal seman'cs to measure seman'c similarity (Perek 2014, 2016) • Words that occur in similar contexts tend to have related meanings (Miller & Charles 1991) “You shall know a word by the company it keeps.” (Firth 1957: 11) • Therefore, a way to characterize the meaning of words is through their distribu'on in large corpora • Widely used in computa'onal linguis'cs • Benefits: – Fully automa'c – Data-driven and objec've 10
Distribu'onal seman'c model • The more frequent collocates are shared by two words, the more similar they will be considered • “Bag of words” approach – Extrac'on of lexical collocates of each verb in a 5-word window from a large corpus – Each verb is assigned an array of values (= a vector) derived from co-occurrence frequencies – High correla'on between vectors is an indica'on of seman'c relatedness 11
Distribu'onal seman'c plots • Output: pairwise distances between verbs • Define a seman'c space that can be ploBed for visualiza'on – By means of t -Distributed Stochas'c Neighbor Embedding algorithm ( t -SNE) (Van der Maaten & Hinton 2008) – Places objects in a 2-dimensional space such that the between-object distances are preserved as well as possible – Superior to mul'dimensional scaling (MDS) for dense spaces with many dimensions – Distance matrix converted to coordinates for each verb • Seman'c domain of the construc'on ploBed for four 'me periods: 1830-1879, 1880-1929, 1930-1969, 1970-2000. • Three senses of the construc'on examined separately 12
The path-crea'on sense 13
1830 − 1879 1880 − 1929 scent brew taste scorch smell smell simmer drink burn burn eat eat blaze gnaw gnaw bite dig scratch hew hew burrow cut cut carve hack prick claw burrow carve slash dig kick lick saw shear wear smooth kick wear paw thrash batter press squeeze press squeeze stammer beat sing wedge wedge hug screw screw fiddle punch fumble dance poke poke shove fumble grope talk flap shove rip grope rip push flap butt gesture push tear rend think think shoulder hitch shoulder plow tear plough worry open crush smash guess crash hustle crush plough plow pierce break shoot crack read take strike jostle fret spell hit burst stuff burst pierce pick pick break spell take pilot explode steer steer drive maneuver blast melt steal fan probe pave spread guide gain gain melt teach explore spread win probe pave bore extend earn feel feel experiment win find lie trace lie trace perfect find work track pay pay make work understand sleep sleep leak make reason build dispute argue purchase dare buy enforce root plan advertise force conquer bribe battle force wrestle fight fight struggle bribe bully beg wrestle trick marshal shape plead shape beg fit forge 1930 − 1969 1970 − 2009 smell smell sniff foul drink drink wash dust mop sample burn burn taste sip puff smoke eat soak scrub nibble eat blaze cook grate chew chew gnaw chop nibble peel slice slice spit chop sting peel cut scrape dig scratch dig etch hack hack slash burrow paint saw carve stomp hew burrow bite cut scratch tickle wear kick etch paw paw saw bellow lick thrash carve pound trim claw hammer snarl hammer ruffle claw pound bawl smooth wear kick clip stroke thrash batter grind grind beat kiss wrinkle batter squeeze press press hug twitch beat shiver whip mumble mumble pry dance improvise dance kiss whisper wail squeeze drum wedge grip wedge grin chatter screw groan tap poke poke blink smile smile sing punch fumble sing skate punch vibrate chat laugh play play ram cling finger shove talk finger clutch laugh push nudge butt kid click shrug shove fumble gamble grope rip grope nod tick shoulder nudge shoulder tear joke joke rip flip talk push flap quarrel butt flap dial pull slam pull slam swing think key think tangle harvest hook tear plow hitch hitch open paste bust puzzle crush crack smash wrap plough pitch rattle wheel scribble toss crumble worry jostle pick shoot smash knock crowd take read brood write cram spell drive bat run break shatter break cram pick shoot box read take write maneuver blast sort navigate steer maneuver blast steal probe pilot probe blow pave melt explore melt wrest bore acquire feel stretch study grow bore feel lie trace lie trace win lead explainwin pay graze gain analyze spend spend find find earn pump seep figure pay sleep work settle imagine work fish dream earn reason hunt marry inflate argue contemplate make build invest swap figure fund make borrow borrow trade dream reason deal con live kill buy root marry will con argue kill entertain murder act act flatter slaughter consume purchase conspire buy compromise plot flatter aim agree negotiate charm negotiate export rape arrest focus force plan manage brave struggle sign digest bribe battle bargain seduce bribe labor force announce manage storm wrestle bully trick fight battle fight cheat wrestle cheat bully sue discipline petition forge pray coax rage beg seethe drill drill forge model Clear concrete/abstract divide in the distribu'onal seman'c plot Higher density of verbs describing forceful ac'ons ( cut , push , kick , ..) , especially in earlier periods 14
Recommend
More recommend