A"Twi&er)Based"Study"of"" - - PowerPoint PPT Presentation

A"Twi&er)Based"Study"of"" Newly"Formed"Clippings"" in"American"English"! !

Sravana!Reddy,!Joy!Zhong,!James!Stanford! Dartmouth!College!

Previous!Work!

Baclawski!(2012)! A!study!of!Fs!(‘adorbs’)!in!40!TwiLer!users!

Research!QuesNons!

• Are!these!clippings!just!cyclical!“slang”?!(Eble!1996,!

2004)! !

• Are!they!an!increasingly!producNve!process!with!new!

social!meanings?! !

• Is!this!type!of!clipping!more!producNve!than!past!

generaNons?! !

• What!is!the!role!of!the!–s!suffix!(adorbs,(awks,(totes)?!

!

• Which!speakers!use!it!the!most?!Age,!gender,!ethnicity?!

This!Study!

Hypothesis:!! Women!are!leading!in!the!usage!of! these!new!clippings,!and!it!is!more! urban/suburban"than!rural!

Labov!(1990,!2001),!Trudgill!(1972),!Coates!&!Pichler!(2011),!Holmes! &!Meyerhoff!(2003),!Wolfram!&!SchillingFEstes!(2006:155F6)!

Why!use!TwiLer!for!! American!Dialect!Research?!

Each!era!applied!contemporary!technology…!

– Kurath!(1939)! – Hanley’s!recordings!(1931F1937)!(Purnell!2012)! – Chambers!&!Trudgill!(1998)! – Labov,!Ash!&!Boberg!(2006)! – Kretzschmar!(2009)! and!many!more !!

Now:!Social!Media!analysis,! computaNonal!modeling,!! Mechanical!Turk!

TwiLer!for!SociolinguisNcs!

– Eisenstein,!O’Connor,!Smith!&!Xing!(2010)! US!regional!variaNon!in!lexical!items! – Bamman,!Eisenstein!&!Schnoebelen!(2012)! Gendered!language!and!networks! – Maybaum!(2012)!! TwiLer!terms! – Zappavigna!(2013)!! TwiLer!discourse!and!variaNon! – Doyle!(2014)! Geographic!distribuNon!of!“needs!done”!

Methodology!

• Collected!185!million!geoFtagged!tweets(
• riginaNng!in!the!US!(JulFNov!2013)!!

by!893,024!users!

• AutomaNcally!extracted!a!list!of!clippings!!
• For!each!word,!created!demographic!profile!of!

users!!

– Gender! – PopulaNon,!median!age,!and!ethnic!distribuNon!at! the!user’s!locaNon!

• Compared!demographic!features!of!clipping!

and!its!original!form!

ExtracNng!Clippings!

• Rather!than!manually!compiling!list!of!

clippings,!automaNcally!learn!from!TwiLer!data!

• A!clipping!and!its!original!form!will!be!used!in!

roughly!similar!contexts!

ExtracNng!Clippings!

• Represent!every!word!

type!as!vector!of!its!leo! and!right!context!

• Rank!every!word!pair!by!

context!vector!similarity!!

• Extract!top!ranked!pairs!

where!first!three! characters!match!

totes! leo:!am,!are,!was,!were,!is..! right:!okay,!ok,!adorbs,!fine…! ! adorable! leo:!is,!so,!these,!looks…! right:!omg,!with,!dork,…! ! totally! leo:!am,!are,!was,!were,!is…! right:!okay,!fine,!insane,!not…! ! !

• Survey!on!Mechanical!Turk!!
• Demographic!quesNons:!age,!gender,!locaNon!!
• Rate!familiarity!with!each!clipping!
• Unfamiliar!
• Familiar,!but!I!do!not!use!it!
• I!use!it!in!speech!only!
• I!use!it!in!wriNng!only!
• I!use!it!in!speech!and!wriNng!
• Same!survey!also!conducted!with!Dartmouth!

undergraduate!students!

Old!vs.!New!Clippings!

Old!vs.!New!Clippings!

• Split!survey!respondents!into!ages!18F29!and!30+!
• For!each!clipping,!compute!average!familiarity!

score!within!the!two!age!groups! !

0!=!Unfamiliar! 3!=!Familiar,!but!I!do! not!use!it! 4!=!I!use!it!in!speech!

• nly!

5!=!I!use!it!in!wriNng!

• nly!

6!=!I!use!it!in!speech! and!wriNng!

1.13 0.89 0.76 1.94 1.52 4.25 3.75 3.6 4.72 4.14 sesh

• bvi

perf probs totes Top858New8Words Avg8Familiarity8Score Over830 Under830

Old!vs.!New!Clippings!

1.13 0.89 0.76 1.94 1.52 4.25 3.75 3.6 4.72 4.14 sesh

• bvi

perf probs totes Top858New8Words Avg8Familiarity8Score Over830 Under830

• Newness!score!for!clipping!!

=!below!30!familiarity!–!above!30!familiarity! !

Old!vs.!New!Clippings!

• Newness!score!for!clipping!!

=!below!30!familiarity!–!above!30!familiarity!

• Threshold!at!1.0!newness!score!

sesh!

• 2 -1 0 1 2 3

perf! adorbs! ridic! vom! alc! vid! frat! perv! choc! cig!

New!clippings!(25)! Old!clippings!(55)!

doc!

Old!vs.!New!Clippings!

sesh! perf! adorbs! ridic! vom! alc! vid! frat! perv! choc! cig!

New!clippings!(25)! Old!clippings!(55)!

doc!

• bvi!

probs! totes! cray! choreo! presh! craycray! convo! def! hilar! gorg! guac! vocab!

• bv!

fam! adorb! calc! chem! esp! prez! merch! vacay! roomie! mins! diff! parm! prac! pedi! sched! prob! prolly! sibs! delish! defs! sus! pic! intro! comfy! prep! secs! approx! fave! milli! collab! combo! apps! perv! pregs! champ! info! preggo! breaky! liq! taL! meds! undies! defly! gents! lolly! anon! fams!

• rtho!

preg! num! McD’s! chiro! cig!

Clippings!on!TwiLer!

0! 100000! 200000! 300000! 400000! 500000! 600000! 700000! 800000! 900000! Clipping! Original!

Old! New!

0! 100000! 200000! 300000! 400000! 500000! 600000! 700000! 800000! 900000! Clipping! Original!

Number!of!users!

Demographic!Analysis!

• Gender!

(following!Bamman!et!al.)!

– Most!TwiLer!users!report!a!name!in!addiNon!to! their!pseudonym! ! ! ! – Match!first!name!against!the!Social!Security! AdministraNon!list!of!baby!names!born!in!1995! – About!2/3!of!users!have!names!in!the!SSA!list!and! are!assigned!a!gender!

Demographic!Analysis!

• LocaNon!!

(following!Eisenstein!et!al.)!

– Tweets!are!geoFtagged!with!laNtude/longitude! – Map!each!geoFcoordinate!to!one!of!33000!! Zip!Code!TabulaNon!Areas!(ZCTAs)! – Ignore!users!that!tweet!from!more!than!one!ZCTA! – Get!demographic!aLributes!of!ZCTAs!from!2010! Census:!PopulaNon,!Median!Age,!White%,!African! American%,!Asian%,!NaNve!American%,!Hispanic%! – Each!user!is!now!associated!with!a!demographic! profile!of!their!environment!

Demographic!Analysis!

• LogisNc!regression!

– Predicted!variable:!clipping!or!original?! – Features:!demographic!profile!of!users!!

• Gender!
• PopulaNon!
• Median!Age!
• Ethnicity!

New!and!Old!Clippings!

Log!odds! All!factors! shown! are!! significant! (p<0.05)!

F0.15! F0.1! F0.05! 0! 0.05! 0.1! New! Old!

Gender!Female!

0! 0.005! 0.01! 0.015! 0.02! 0.025! 0.03! 0.035! New! Old!

Log!odds! Log10!PopulaNon! Log!odds!

No!Significant!Effects!

New!and!Old!Clippings!

Log!odds!

F0.005! F0.004! F0.003! F0.002! F0.001! 0! 0.001! 0.002! 0.003! 0.004! 0.005! New! Old!

Log!odds! Median!Age! Ethnicity!

Usage!of!Fs!suffix!in!clippings!

0! 0.1! 0.2! 0.3! 0.4! 0.5! 0.6!

Gender!Female! Log!odds!

F0.016! F0.014! F0.012! F0.01! F0.008! F0.006! F0.004! F0.002! 0!

Log!odds! Median!Age! adorbs/adorb,! probs/prob,! fams/fam,! awks/awk,! pregs/preg,! defs/def! !

Conclusion!

Hypothesis(Confirmed:!! Women!are!leading!in!the!usage!of! these!new!clippings,!and!it!is!more! urban/suburban"than!rural!

Are!clippings!a!TwiLer!arNfact?!

They!abound!in!longFform!blog!posts!too…! …!and!TwiLer!users!ooen!lengthen!words!

Are!clippings!a!TwiLer!arNfact?!

Baldwin!et!al.!(2013)!measure!average!word!lengths! in!TwiLer!and!different!corpora!

Are!clippings!a!TwiLer!arNfact?!

Eisenstein!et!al.!(2013)!find!shortened!forms!! are!mainly!used!in!tweets!of!length!much!less!than!! 140!characters.!Shortening!is!not!used!in!order!to!! fit!length!constraints!!

Are!clippings!a!TwiLer!arNfact?!

Our!experiment:!avg!lengths!of!tweets!containing!! clippings!compared!to!tweets!with!original!forms!

77.01! 81.64! 0! 20! 40! 60! 80! 100! 120! 140! Clipping! Original!

Old! New!

65.08! 73.74! 0! 20! 40! 60! 80! 100! 120! 140! Clipping! Original!

Future!Work!

• Track!spread!of!clippings!in!TwiLer!over!Nme!

– Will!these!clippings!spread!throughout!the! populaNon?! – Geographic/demographic!dimensions!of!spread?! – When!did!these!clippings!originate?!! !

• MorphoFphonological!study!of!clippings!

!