C, 342 331 bytes

char*x[]={"teen","one","two","three","four","five","six","seven","eight","nine","ten","eleven","twelve","thir","four","fif","twenty","thirty","fourty","fifty"};void main(int z,char**i){for(z=1;z<=atoi(i[3]);z++)printf("%s%s%s\n",z<16?x[z]:z<20?z^18?x[z-10]:"eigh":x[z/10+14],z>20&&z%10?"-":z>12&&z<20?*x:"",z>20&&z%10?x[z%10]:"");}

Try it online!

Python3, 276 271 269 243 237 235 232 217 bytes

Taking a cue from the @smls perl submission...

from unicodedata import*
o=[name(chr(k)).split(' ',2)[-1]for j in['①⑴','㉑㉠','㊱㋀']for k in range(ord(j[0]),ord(j[1]))]
i=input()
w=i in o
for i in range(w and o.index(i)+1or int(i)):print(w and i+1or o[i])

I suspect it might be golfed a little bit further.

It makes use of the system library unicodedata to look up names for numbers. It requires upper case number names (separated by space: FORTY TWO) or decimal integers as input.

(This is my first code golf submission.)

(I also just noticed I was miscounting the length (encoding), so it's a few bytes less than previously thought. I've only updated the most recent byte count, though. Oops.)

C++11, 484 480 477 bytes

#import<iostream>
#import<cstdlib>
#import<vector>
using namespace std;f(){int j,i=2;string s="teen";vector<string>v={"","one","two","three","four","five","six","seven","eight","nine","ten","eleven","twelve"};for(;i++<9;)v.push_back(v[i]+s);v[13]="thir"+s;v[15]="fif"+s;v[18]="eigh"+s;for(i=19;i++<50;){string n[4]={"twenty","thirty","forty","fifty"};v.push_back(n[i/10-2]+v[i%10]);}cin>>s;if(i=atoi(s.c_str()))for(j=0;j++<i;)cout<<v[j]<<" ";else while(v[i++]!=s)cout<<i<<" ";}

Text input in lower-case without hyphens.

SAS, 179

%macro c(n);%let f=words.;%if%length(&n)>2%then%do;%do c=1%to 50;%if%qsysfunc(putn(&c,&f))=&n%then%let n=&c;%end;%let f=2.;%end;%do i=1%to &n;%put%sysfunc(putn(&i,&f));%end;%mend;

Output is written to the log separated by newlines. SAS has a built-in format for converting digits to words, which is a major advantage for this challenge, but annoyingly it lacks an informat for doing the reverse.

Python 2, 432 422 416 403 bytes

I'm sure this can be improved on. At the very least if I can get away with hardcoding the value to be worked on and not needing a function I can save 20. It needs a space to separate words in text input. Saved 6 bytes thanks to JonathanAllan's comment on ElPedro's answer, 4 for rearranging maths.

def z(f):
 a,b,i,d="one two three four five six seven eight nine ten eleven twelve thir#four#fif#six#seven#eigh#nine#".replace("#","teen ").split()+[""],"twenty thirty forty fifty".split(),1,f>50
 if d:f=f.split();f=a.index(f[-1])+21+b.index(f[-2])*10 if len(f)>1 else b.index(f[-1])*10+20 if f[-1]in b else a.index(f[-1])+1
 while i<=f:s=i if d else a[i-1]if i<20 else b[i//10-2]+a[i%10-1];print s;i+=1

(NB: The actual version of this uses tabs to indent instead of spaces. QPaysTaxes added a single space because that wasn't rendering properly, to ensure that the given code compiles. It shouldn't change the byte count.)

R, 452 430 424 bytes

o=c("","one","two","three","four","five","six","seven","eight","nine") 
t=gsub(0,"teen",c("ten","eleven","twelve","thir0","four0","fif0","six0","seven0","eigh0","nine0"))
s=c("twenty","thirty","forty") 
p=""
for(i in s){for(j in o){p=paste0(p,i,j," ")}}
as.data.frame(t(d<-1:50))
names(d)=c(o[-1],t,as.vector(strsplit(p," ")[[1]]),"fifty")
f=function(x){if(is.numeric(x)){names(d)[1:x]}else{matrix(d[1:d[x]],dimnames=NULL)}}

#> f(5)
#[1] "one"   "two"   "three" "four"  "five" 

#> f('five')
#     [,1]
#[1,]    1
#[2,]    2
#[3,]    3
#[4,]    4
#[5,]    5

Places the numbers in a data.frame with written-out numbers as column names, making the translation between the two (and subsequent printing) pretty easy.

Main attempt at golfing was in creating the written-out numbers for 20-49, probably much more to golf here.

I made an attempt with as.matrix to print the data.frame with just the numbers, but am still left with a matrix header. Hopefully that's ok.

Ungolfed:

ones <- c("","one","two","three","four","five","six","seven","eight","nine") 
teens <- c("ten","eleven","twelve","thirteen","fourteen","fifteen","sixteen","seventeen","eighteen","nineteen")
tens <- c("twenty","thirty","forty") 

p=""
for(i in tens){
  for(j in ones){
    p=paste0(p, i, j," ")
  }
}

nums <- 1:50
as.data.frame(t(nums))
names(nums) <- c(ones[-1], teens, as.vector(strsplit(p, " ")[[1]]), "fifty")
f <- function(x){
  if(is.numeric(x)){
    names(nums)[1:x]
  } else {
    matrix(nums[1:nums[x]], dimnames = NULL)
  }
}

Python 3, 305 303 bytes

Converted to Python 3 after advice from @nedla2004. Now also has no space between written numbers on input or output e.g. enter twentytwo

l='one two three four five six seven eight nine ten eleven twelve thir#four#fif#six#seven#eigh#nine#'.replace('#','teen ').split()
m='twenty','thirty','forty','fifty'
i,R=input(),range
l+=sum([[m[x]]+[m[x]+l[y]for y in R(9)]for x in R(3)],[])
for x in R(1,l.index(i)+2)if i in l else l[:int(i)]:print(x)

Try it online 3!

Python 2, 327 320 313 308 bytes

l='one two three four five six seven eight nine ten eleven twelve thir#four#fif#six#seven#eigh#nine#'.replace('#','teen ').split()
m='twenty','thirty','forty'
i,R=raw_input(),range
l+=sum([[m[x]]+[m[x]+l[y]for y in R(9)]for x in R(3)],[])+['fifty']
for x in R(1,l.index(i)+2)if i in l else l[:int(i)]:print x

Try it online 2!

163 170 177 bytes shorter than my original answer so I am posting it as an alternative. This uses for on the two lists to build up a complete list of all of the string representations of the numbers then identifies the right one in the list and prints everything up to it either by value or by index. Outputs a new line for each value.

JavaScript (ES6), 261 bytes

Note: the string assigned to z is encoded with atob. In the encoded string there are 11 bytes that I can not post to this site, even if they are valid characters in a javascript string. So I used an hex escape in the form \xHH. Each one of these escapes is counted as 1 byte.
The original uncompressed string is the less golfed version.

x=>(z=btoa('ö\x89ÞöÜ(öØkyï_¢êý~+Þöȱöǯz\x7f^\x8a\x08möx§{Û^\x9f×¥z÷§öÜ\x1e\x96÷½¶\x18«÷×â\x7fß}z(!÷Ûpz\x7f}~\x8aý').split(9),o=(0+z.map((v,i)=>i<20?i<13?v:(v||z[i-10])+'teen':z.slice(0,10).map(d=>(z[i]||z[i-8]||z[i-18])+'ty'+d))).split`,`,p=o.indexOf(x),o.slice(1,-~x+p+!~p).map((x,i)=>~p?i+1:x))

Less golfed

x => (
  z = '9one9two9three9four9five9six9seven9eight9nine9ten9eleven9twelve9thir99fif999eigh99twen99for9'
      .split(9),
  o = (0 + // 0 + array to build a comma separated string
       z.map( (v, i) => 
         i < 20 
         ? i < 13 
           ? v // 1 to 13 are 'as is'
           : (v||z[i-10])+'teen' // compose for 14 to 19
         : z.slice(0,10).map(d=>(v||z[i-8]||z[i-18])+'ty'+d)) // 20s, 30s, 40s, 50s
      ).split`,`, // from comma separated to array again
  // o contains strings from one to fiftynine
  p = o.indexOf(x), // look for input
  o.slice(1, -~x+p+!~p).map((x,i) => ~p?i+1:x)
)

Test

F=
x=>(z=btoa('ö\x89ÞöÜ(öØkyï_¢êý~+Þöȱöǯz\x7f^\x8a\x08möx§{Û^\x9f×¥z÷§öÜ\x1e\x96÷½¶\x18«÷×â\x7fß}z(!÷Ûpz\x7f}~\x8aý').split(9),o=(0+z.map((v,i)=>i<20?i<13?v:(v||z[i-10])+'teen':z.slice(0,10).map(d=>(v||z[i-8]||z[i-18])+'ty'+d))).split`,`,p=o.indexOf(x),o.slice(1,-~x+p+!~p).map((x,i)=>~p?i+1:x))

function update() {
  var i=I.value
  O.textContent = F(i)
}  

update()

<input id=I value=25 oninput='update()'><pre id=O></pre>

Python 2, 262 bytes

x="one two three four five six seven eight nine ten eleven twelve thir#four#fif#six#seven#eigh#nine#".replace("#","teen ").split()
x+=[a+"ty"+b for a in"twen","thir","for","fif"for b in['']+x[:9]]
v=input()
for s in range(1,x.index(v)+2)if v>50else x[:v]:print s

repl.it

Input and output strings are lower case and concatenated*, so to test a string input enter, for example, "thirtyfive" at the prompt.

Builds the list of all the words (plus "fiftyone" to "fiftynine"), x, then tests if input is a word with the proxy v>50 (strings are greater than numbers in Python 2, and all the numbers in the valid input range from the specification are <=50) and prints the appropriate values by either slicing the list, x[:v], or building a range of integers, range(1,x.index(v)+2).

* Adding hyphenation on both costs 11 bytes, by replacing a+"ty"b with a+"ty"+'-'*(b>'')+b.

Common Lisp, 297 253 243 242 144 128

(lambda(s)(#1=dotimes(u(or(#1#(i 51)(if(equal(#2=format()"~R"i)s)(return i)))s))(#2#t"~[~:;, ~]~:[~R~;~D~]"u(stringp s)(1+ u))))

Details

(lambda (s) 
  (dotimes                         ; iterate...                                                                          
      (u                           ; for u from zero below ...                
       (or                         ; if s is a string, then                   
        (dotimes (i 51)            ;   parse s by iterating from 0 to 50      
          (if (equal               ;   until we find a match between          
               (format nil "~R" i) ;   the English word(s) for i              
               s)                  ;   and the given s                        
              (return i)))         ;   (exit loop)                            
        s))                        ; otherwise, use s, which is a number      
    (format t                      ; for each U, print to standard output     
            "~[~:;, ~]~:[~R~;~D~]" ; (see below for details)                  
            u                      ; ...                                      
            (stringp s)            ; ... arguments to format                  
            (1+ u))))              ; ...                                      

• ~[ 0 ~; 1 ~; ... ~:; else ~] is a switch, based on the next available argument's value, which jumps to the appropriate sub-control format. Here, I only have a case of "0" and for "else". This is used to insert a separator before each number except the first one, thanks to U starting from zero.

• ~:[ FALSE ~; TRUE ~] is a conditional format; here we output things differently whether the input s is a string or not.

• ~R write a number as a cardinal English number, whereas ~D simply prints the number.

Examples

CL-USER> (test "five")
1, 2, 3, 4, 5

CL-USER> (test 32)
one, two, three, four, five, six, seven, eight, nine, ten, eleven, twelve, thirteen, fourteen, fifteen, sixteen, seventeen, eighteen, nineteen, twenty, twenty-one, twenty-two, twenty-three, twenty-four, twenty-five, twenty-six, twenty-seven, twenty-eight, twenty-nine, thirty, thirty-one, thirty-two

Swift3, 402 bytes

let f=["one","two","three","four","five","six","seven","eight","nine"]
let g=["twenty","thirty","forty","fifty"]
let v=[f,["ten","eleven","twelve"],["thir","four","fif","six","seven","eigh","nine"].map{$0+"teen"},[g[0]],f.map{g[0]+$0},[g[1]],f.map{g[1]+$0},[g[2]],f.map{g[2]+$0},[g[3]]].flatMap{$0}
func c(s:String){if let i=Int(s){print(v.prefix(upTo:i))}else{for j in 1...v.index(of:s)!+1{print(j)}}}

Ungolfed:

let f = ["one","two","three","four","five","six","seven","eight","nine"]
let g = ["twenty","thirty","forty","fifty"]

let values = [f,["ten","eleven","twelve"],["thir","four","fif","six","seven","eigh","nine"].map{$0+"teen"},
              [g[0]], f.map{g[0]+$0},
              [g[1]], f.map{g[1]+$0},
              [g[2]], f.map{g[2]+$0},
              [g[3]]].flatMap{$0}

func count(s:String){
    if let i = Int(s) {
        print(values.prefix(upTo: i))
    } else {
        for j in 1...values.index(of: s)!+1{
            print(j)
        }
    }
}

count(s:"29")
count(s:"twentyeight")

Nothing special here, just using an array to back up the written-out numbers.

I originally thought this solution using this other way to calculate the values array:

let values = f + ["eleven","twelve"]
    + ["thir","four","fif","six","seven","eigh","nine"].map{$0+"teen"}
    + [g[0]] + f.map{g[0]+$0}
    + [g[1]] + f.map{g[1]+$0}
    + [g[2]] + f.map{g[2]+$0}
    + [g[3]]

Which could be golfed to:

let v=f+["eleven","twelve"]+["thir","four","fif","six","seven","eigh","nine"].map{$0+"teen"}+[g[0]]+f.map{g[0]+$0}+[g[1]]+f.map{g[1]+$0}+[g[2]]+.map{g[2]+$0}+[g[3]]

replacing the 3rd line in the golfed code

I could have scored 381 bytes, but, there is a compiler error that says: "expression was too complex to be solved in reasonable time", more info on the error can be found here

PHP - 397 372 349 344 329 bytes

Inspired by TomDevs's JS solution

Saved 25 bytes by replacing $a=[...] by $a=explode(...)

Saved another 23 bytes by switching back to an array without string delimiters and by storing teen in a variable, thanks to @user59178

Saved another 5 bytes by removing the (int) typecasting

Saved another 15 bytes by dropping $b, the $i in the for declarations, and the curly braces, thanks to @user59178 again

$a=[one,two,three,four,five,six,seven,eight,nine,ten,eleven,twelve,thir.$t=teen,four.$t,fif.$t,six.$t,seven.$t,eigh.$t,nine.$t];foreach([twenty,thirty,forty,fifty] as$c){$a[]=$c;for($i=0;$i<9;)$a[]=$c.'-'.$a[$i++];}if($argv[1]!=0)for($i=0;$i<$argv[1];)echo$a[$i++].' ';else for($i=1;$i<=array_search($argv[1],$a)+1;)echo$i++.' ';

Ungolfed :

$a =[one,two,three,four,five,six,seven,eight,nine,ten,eleven,‌​twelve,thir.$t=teen,‌​four.$t,fif.$t,six.$‌​t,seven.$t,eigh.$t,n‌​ine.$t];
foreach ([twenty,thirty,forty,fifty] as $c){
    $a[] = $c;
    for ($i=0;$i<9;)
        $a[] = $c . '-' . $a[$i++];
}
if( $argv[1] !=0 )
    for ($i=0;$i<$argv[1];)
        echo $a[$i++] . ' ';
else
    for ($i=1;$i<=array_search($argv[1], $a)+1;)
        echo $i++ . ' ';

Try it for an input string or for an input number

Scheme, 161, 152, 149

(define (c x)(let((r(string->number x)))(let l((i 1))(let((n (format #f "~r" i)))(display(if r n i))(newline)(or(eq? r i)(equal? x n)(l (+ i 1)))))))

Uncompressed:

(define (count limit)
  (let ((numerical-limit (string->number limit)))
    (let l ((i 1))
      (let ((current-number (format #f "~r" i)))
        (display (if numerical-limit current-number i))
        (newline)
        (or (eq? numerical-limit i)
            (equal? limit current-number)
            (l (+ i 1)))))))

Wolfram Language, 92 bytes

If[NumberQ@#, Do[Print@IntegerName@i, {i, #}], 
  Do[Print@i, {i, Interpreter["SemanticNumber"]@#}]] &

(I'm new to this, let me know if I did something wrong)

Python 2, 503 499 494 490 479 bytes

-5 with thanks to @JonathanAllan

l='one two three four five six seven eight nine ten eleven twelve thir#four#fif#six#seven#eigh#nine#'.replace('#','teen ').split()
m='twenty','thirty','forty','fifty'
i,z,R=raw_input(),' ',range
try:n=int(i);p=(n/10)-2;o=(l+sum([[m[x]]+[m[x]+z+l[y]for y in R(9)]for x in R(p)],[])+[m[p]]+[m[p]+z+l[y]for y in R(n%10)],l[:n])[n<20]
except:j=i.split();o=map(str,R(1,(m.index(j[0])+2)*10+l.index(j[1])+2if z in i else l.index(i)+2if i in l else(m.index(i)+2)*10+1))
print','.join(o)

Try it online!

Input either a number or a space seperated spelling of a number.

Slightly less golfed and more readable version:

l='one two three four five six seven eight nine ten eleven twelve thirteen fourteen fifteen sixteen seventeen eighteen nineteen'.split()
m='twenty','thirty','forty','fifty'
i=raw_input()
try:
 n=int(i)
 if n<20:
  o=l[0:n]
 else:
  o=l
  for x in range((n/10)-2):
   o+=[m[x]]+[m[x]+' '+l[y]for y in' '*9]
  p=m[(n/10)-2]
  o+=[p]+[p+' '+l[y]for y in' '*n%10]
except:
 if' 'in i:
  t=i.split()
  s=((m.index(t[0])+2)*10)+l.index(t[1])+2
 else:
  s=l.index(i)+2 if i in l else((m.index(i)+2)*10)+1
 r=range(1,s)
 o=map(str,r)
print','.join(o)

PowerShell, 362 bytes

$z=0..50|%{("0twenty0thirty0forty0fifty"-split0)[+(($b="$_"[0])-gt49)*($_-gt19)*(+"$b"-1)]+($x=(("0one0two0three0four0five0six0seven0eight0nine0ten0eleven0twelve"-split0)+(-split'thir four fif six seven eigh nine'|%{$_+'teen'})))[($_%10)*($_-gt19)]+$x[$_*($_-le19)]}
if(($n=-split$args)[0][0]-in48..57){$z[$n[0]..$n[2]]}else{$z.IndexOf($n[0])..$z.IndexOf($n[2])}

Try it online! words input or numbers input

This is a right mess, and I'm not terribly happy with it, but here it is. Golfing suggestions welcome.

The first line sets $z to be an array of the full English words. You can see the -split0 for numbers 1 to 12, and the loop to construct all the teens, and then there's a bunch of logic to put everything together right. Try it online!

The second line starts with some logic. We take the input $args (as a string), -split it on whitespace, store it into $n for use later, take the first [0] word, and the first [0] character of that, and check if it is -in a range 48..57 (i.e., ASCII 0 to 9). So, we're checking if we have decimal input or English input. Try it online!

In the first case, we build a range based on the decimal inputs $n[0]..$n[2] and use that to index into $z[...]. In the other case, we find the .indexOf() the first word and the last word, and build just a numerical range from that. In either situation, we now have an array of objects on the pipeline (either strings or integers), and an implicit Write-Output at program completion gives us a newline between elements.

JavaScript ES6, 559 526 381 368 364 358 332 327 315 bytes

a="one0two0three0four0five0six0seven0eight0nine0ten0eleven0twelve0thir10four10fif10six10seven10eigh10nine1".replace(/1/g,'teen').split(0),b="twenty0thirty0forty0fifty".split(0),c=(n,d=Array,e=b.forEach(i=>a=a.concat(i,a.slice(0,9).map(x=>i+x))))=>1/n?a.slice(0,n).join():d.from(d(a.indexOf(n)+1),(x,i)=>i+1).join();

Thanks to Kritixi Lithos for the idea of splitting the array and Arnauld for the 1/n trick.

a="one0two0three0four0five0six0seven0eight0nine0ten0eleven0twelve0thir10four10fif10six10seven10eigh10nine1".replace(/1/g,'teen').split(0),b="twenty0thirty0forty0fifty".split(0),c=(n,d=Array,e=b.forEach(i=>a=a.concat(i,a.slice(0,9).map(x=>i+x))))=>1/n?a.slice(0,n).join():d.from(d(a.indexOf(n)+1),(x,i)=>i+1).join();

console.log(c("twentyfive"));
console.log(c("fifty"));
console.log(c(50));

Perl 6, 119 113 bytes

{my \n=<① ⑳ ㉑ ㉟ ㊱ ㊿>.map:{|map *.uniname.words[2..*].join,$^a..$^b}
/\d/??n[^$_]!!1..1+first $_,n,:k}

Unicode database FTW!

Uses upper-case written-out numbers without hypen, e.g. TWENTYTWO.
Returns a list of strings, or a range of numbers. (Both use space as separator when printed with put.)