Some general points which do not address the OPed question (BTW: I think hippo is on the mark with the observation that  $i3 is never used), but may be of interest:

• $textarray[$iwrdnum][1] = $tagfields[0]; #the ... 
  $textarray[$iwrdnum][2] = $tagfields[1];
  $textarray[$iwrdnum][3] = $tagfields[2];
  $textarray[$iwrdnum][4] = $tagfields[3];
  $textarray[$iwrdnum][5] = $tagfields[4];
  [download]

  This can, IMHO, be expressed with more clarity/concision/maintainability with array slices (see Slices in perldata) (untested):
      @{ $textarray[$iwrdnum] }[ 1 .. 5 ] = @tagfields[ 0 .. 4 ];
  
  
• Also in the interest of concision, the statement
      print (OUTFILE "$textarray[$i][0]$textarray[$i+1][0]$textarray[$i+2][0]$textarray[$i+3][0]$textarray[$i+4][0]$textarray[$i+5]\n ");
  could be re-written as (also untested)
      print OUTPUT join '', map $textarray[ $i+$_ ][0], 0 .. 5;
      print OUTPUT "\n ";
  (I assume the final  $textarray[$i+5] in the OPed code should really have been  $textarray[$i+5][0]). See map.
  
  
• The capturing groups in a couple of regexes that look like  /\b(nn|nns|nvbg|np|nps|npl)\b/ can be replaced by non-capturing  (?:pattern) groupings. The captured substrings never seem to be used, and if this is true, the conventional wisdom is that non-capturing grouping is a bit faster. This could make a significant difference if you're crunching lotsa data. See  "(?:pattern)" in Extended Patterns in perlre; also Non-capturing groupings in perlretut.